CN111241237B - Intelligent question-answer data processing method and device based on operation and maintenance service - Google Patents
Intelligent question-answer data processing method and device based on operation and maintenance service Download PDFInfo
- Publication number
- CN111241237B CN111241237B CN201911403393.XA CN201911403393A CN111241237B CN 111241237 B CN111241237 B CN 111241237B CN 201911403393 A CN201911403393 A CN 201911403393A CN 111241237 B CN111241237 B CN 111241237B
- Authority
- CN
- China
- Prior art keywords
- word segmentation
- model
- scene
- matching
- library
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/3332—Query translation
- G06F16/3335—Syntactic pre-processing, e.g. stopword elimination, stemming
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Economics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Tourism & Hospitality (AREA)
- Molecular Biology (AREA)
- Human Computer Interaction (AREA)
- Development Economics (AREA)
- Evolutionary Computation (AREA)
- Technology Law (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Resources & Organizations (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent question-answer data processing method and device based on operation and maintenance service. The method comprises the following steps: obtaining corpus data to be processed of a target object; word segmentation is carried out on the corpus data to be processed to obtain word segmentation results, wherein the word segmentation results comprise at least two word segmentation data and corpus analysis results; inputting the word segmentation result into a classification model for service classification, wherein the classification model is obtained by performing machine learning training on a plurality of service sample data, and the service sample data carries corresponding service classification labeling information; when the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model; and acquiring an answer associated with the target question, and taking the associated answer as a target answer to be returned.
Description
Technical Field
The invention relates to the technical field of internet communication, in particular to an intelligent question-answer data processing method and device based on operation and maintenance service.
Background
With the rapid development of internet communication technology, intelligent question-answering systems have been developed. The intelligent question-answering system can bring convenience to life and work of people. Based on the intelligent question-answering system, communication between the user and the machine can be established, and accordingly, the machine is required to understand and use natural language of human society such as Chinese, english and the like, and then the machine responds according to the natural language input by the user.
In the related art, a Long Short Term Memory (LSTM) network is trained based on a question sample and an answer sample to obtain a target model. When a question is input to the target model, the target model performs answer output. However, the answers thus obtained often lack linguistic logic and require further processing. Thus, a more accurate and efficient answer return scheme is needed.
Disclosure of Invention
In order to solve the problems of low accuracy rate of returned answers and the like in an intelligent question-answering system in the prior art, the invention provides an intelligent question-answering data processing method and device based on operation and maintenance service, which comprises the following steps:
in one aspect, the invention provides an intelligent question-answer data processing method based on operation and maintenance service, which comprises the following steps:
Obtaining corpus data to be processed of a target object;
word segmentation is carried out on the corpus data to be processed to obtain word segmentation results, the word segmentation results comprise at least two word segmentation data and corpus analysis results, and the corpus analysis results comprise at least one of the following: part-of-speech analysis results, grammar analysis results, entity analysis results and emotion analysis results;
inputting the word segmentation result into a classification model for service classification, wherein the classification model is obtained by performing machine learning training on a plurality of service sample data, and the service sample data carries corresponding service classification labeling information;
when the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model;
acquiring an answer associated with the target question, and taking the associated answer as a target answer to be returned;
the problem matching models comprise scene problem matching models with at least two scene dimensions, the scene problem matching models with at least two scene dimensions are used for performing scene problem matching on the word segmentation result based on corresponding priority sequences, each scene problem matching model is configured with a corresponding candidate problem library, and each scene problem matching model is obtained through machine learning training of the corresponding candidate problem library.
In another aspect, an intelligent question-answering data processing device based on operation and maintenance service is provided, where the device includes:
corpus data acquisition module: the method comprises the steps of obtaining corpus data to be processed of a target object;
the word segmentation processing module: the method comprises the steps of performing word segmentation on corpus data to be processed to obtain word segmentation results, wherein the word segmentation results comprise at least two word segmentation data and corpus analysis results, and the corpus analysis results comprise at least one of the following: part-of-speech analysis results, grammar analysis results, entity analysis results and emotion analysis results;
and a service classification module: the classification model is obtained through machine learning training of a plurality of service sample data, and the service sample data carries corresponding service classification labeling information;
and a problem matching module: when the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model;
answer acquisition module: the method comprises the steps of obtaining an answer associated with the target question, and taking the associated answer as a target answer to be returned;
The problem matching models comprise scene problem matching models with at least two scene dimensions, the scene problem matching models with at least two scene dimensions are used for performing scene problem matching on the word segmentation result based on corresponding priority sequences, each scene problem matching model is configured with a corresponding candidate problem library, and each scene problem matching model is obtained through machine learning training of the corresponding candidate problem library.
In another aspect, an electronic device is provided, where the electronic device includes a processor and a memory, where at least one instruction or at least one section of program is stored in the memory, where the at least one instruction or the at least one section of program is loaded and executed by the processor to implement an intelligent question-answer data processing method based on an operation and maintenance service as described above.
In another aspect, a computer readable storage medium is provided, where at least one instruction or at least one program is stored, where the at least one instruction or the at least one program is loaded and executed by a processor to implement an intelligent question-answer data processing method based on an operation and maintenance service as described above.
The intelligent question-answering data processing method and device based on the operation and maintenance service provided by the invention have the following technical effects:
According to the invention, a machine learning model for service classification and problem matching is introduced into the intelligent question-answering system, so that the accuracy and adaptability of selecting matched problems for corpus data to be processed from a question library can be improved. Returning the answer associated with the question as the target answer can improve the accuracy and efficiency of determining the target answer. Based on the difference of the service and the dimension of the (service) scene, the corresponding problem library is configured for the related model, so that the problem matching efficiency is improved, and meanwhile, the problem library is convenient to manage and maintain.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of an application environment provided by an embodiment of the present invention;
fig. 2 is a schematic flow chart of an intelligent question-answer data processing method based on operation and maintenance service according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart of a method for inputting the word segmentation result into a problem matching model matched with the operation and maintenance service and determining a target problem matched with the word segmentation result from a candidate problem library by using the problem matching model according to the embodiment of the invention;
FIG. 4 is a schematic flow chart of a method for inputting the word segmentation result into a problem matching model matched with the operation and maintenance service and determining a target problem matched with the word segmentation result from a candidate problem library by using the problem matching model according to the embodiment of the invention;
FIG. 5 is a diagram of a model architecture for scene problem matching using a transformer-based bi-directional coded representation model (BERT model) provided by an embodiment of the present invention;
FIG. 6 is a schematic flow chart of the scene problem matching by inputting the word segmentation result into the second-level submodel according to the embodiment of the present invention;
FIG. 7 is a flowchart of determining a target answer when the result obtained by classification indicates a non-operation service according to an embodiment of the present invention;
FIG. 8 is a block diagram of an intelligent question-answer data processing device based on operation and maintenance services according to an embodiment of the present invention;
fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It is noted that the terms "comprises" and "comprising," and any variations thereof, in the description and claims of the present invention and in the foregoing figures, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server comprising a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.
Referring to fig. 1, fig. 1 is a schematic diagram of an application environment provided in an embodiment of the present invention, which may include a client 01 and a server 02, where the client and the server are connected through a network. The client sends the corpus data to be processed of the target object to the server, and the server returns corresponding answers based on the corpus data to be processed. It should be noted that fig. 1 is only an example.
Specifically, the client 01 may include a smart phone, a desktop computer, a tablet computer, a notebook computer, an augmented reality (augmented reality, AR)/Virtual Reality (VR) device, a digital assistant, a smart wearable device, or other types of physical devices, and may also include software running in the physical devices, such as a computer program. The operating system running on the client 01 may include, but is not limited to, android system (Android system), IOS system (mobile operating system developed by apple corporation), linux (an operating system), microsoft Windows (microsoft windows operating system), and the like.
In particular, the server 02 may include a server that operates independently, or a distributed server, or a server cluster that is composed of a plurality of servers. The server 02 may include a network communication unit, a processor, a memory, and the like. The server 02 may provide background services for the clients described above.
In practical application, a user can perform man-machine conversation through an application program with an intelligent question-answering function installed on a terminal device, and the user inputs a problem on a user interaction interface of the application program.
Fig. 2 is a schematic flow chart of an intelligent question-answer data processing method based on an operation and maintenance service according to an embodiment of the present invention, where the description provides the steps of the method according to the embodiment or the flowchart, but may include more or fewer steps based on conventional or non-creative labor. The order of steps recited in the embodiments is merely one way of performing the order of steps and does not represent a unique order of execution. When implemented in a real system or server product, the methods illustrated in the embodiments or figures may be performed sequentially or in parallel (e.g., in a parallel processor or multithreaded environment). As shown in fig. 2, the method may include:
S201: obtaining corpus data to be processed of a target object;
in the embodiment of the present invention, the target object may indicate the current login account id of the client (here and below refer to software running in the terminal setting, as opposed to the server) or the identity of the client. Accordingly, the corpus data to be processed of the target object may represent a query question entered by a registered user using the current login account or a query question entered by a guest performing a service experience based on the identity of the client.
The corpus data to be processed can be text data, such as "university A is located at B place", "Hello", and the like. The corpus to be processed can also be non-text data of data types such as images, audios or videos, and the non-text data can be converted into text data through a voice recognition technology or an image recognition technology.
S202: word segmentation is carried out on the corpus data to be processed to obtain word segmentation results, the word segmentation results comprise at least two word segmentation data and corpus analysis results, and the corpus analysis results comprise at least one of the following: part-of-speech analysis results, grammar analysis results, entity analysis results and emotion analysis results;
In the embodiment of the invention, the vocabulary and the syntax corresponding to the corpus data to be processed can be focused in the word segmentation processing of the corpus data to be processed, and the corresponding vocabulary and syntax are analyzed. Correspondingly, the word segmentation result comprises at least two word segmentation data and a corpus analysis result, and the corpus analysis result comprises at least one of the following: part of speech analysis results, grammar analysis results, entity analysis results and emotion analysis results. By adopting a morphology and syntax combined analysis mode, the word segmentation result can be expressed with the meaning of the whole sentence of the query question concerned. As the word segmentation result of the input of the subsequent classification model, the method can help classification to more accurately classify the business.
For example, the corpus data to be processed corresponds to "university a is located at B"; the at least two word segmentation data correspond to university a, located at ground B; the part of speech analysis results indicate that: "university A" and "B" are nouns, "located" is a preposition; the parsing result indicates: "university A" is the subject, "located" is the predicate and "B ground" is the object; the entity analysis result indicates that: "university A" is the organization name and "B land" is the place name; the emotion analysis result indicates that: neutral emotion (versus positive emotion, negative emotion).
In practical application, stanford CoreNLP (NLP: natural Language Processing, natural language processing) technology (a natural language processing technology) can be adopted when the corpus data to be processed is subjected to word segmentation, part-of-speech analysis, grammar analysis and entity recognition analysis, so that the query problem is further subjected to word segmentation. Based on Stanford CoreNLP technology, basic properties (such as company name, person name, normalized date, etc.) of the word segmentation can be given, sentence structure of the query question is marked according to grammar, and relationships among entities in the sentence are found, etc. The Stanford CoreNLP technique provides a set of tools for processing natural language that point to numerous grammars that can improve the robustness and efficiency of word segmentation processing based on the Stanford CoreNLP technique.
S203: inputting the word segmentation result into a classification model for service classification, wherein the classification model is obtained by performing machine learning training on a plurality of service sample data, and the service sample data carries corresponding service classification labeling information;
in the embodiment of the invention, the classification model obtained by training has high generalization capability. The plurality of traffic sample data may include positive sample data (corresponding to an operation and maintenance traffic) and negative sample data (corresponding to a non-operation and maintenance traffic, such as boring). For example, "how a mailbox files" corresponding positive sample data, "what is you eating in the noon? "corresponds to negative sample data".
In a specific embodiment, before the step of inputting the word segmentation result into the classification model for service classification, mapping processing may be performed on the word segmentation result to obtain a word segmentation result vector. Specifically, the word2vec model (a group of related models used to generate word vectors) obtained by training may be used to map the word segmentation result into a vector matrix, that is, the word segmentation result vector. The word2vec model is obtained through machine learning training through a large amount of text data.
Correspondingly, inputting the word segmentation result vector into the classification model; extracting business features of the word segmentation result vector by using a convolution layer (convolution layer) of the classification model; outputting, by a fully connected layer (fully connected layer) of the classification model, the traffic classification result based on the traffic characteristics. Specifically, the classification model can be obtained by training a TextCNN network (the network is a convolutional neural network proposed in 2014 for text classification) based on massive operation and maintenance problem samples and boring problem samples, and the TextCNN network has the characteristics of simple structure and good classification effect, and is widely applied to the NLP fields of text classification, recommendation and the like.
In practical applications, the normal operation of financial institutions, such as banks, securities companies, insurance companies, trust investments companies, and fund management companies, is critical to the normal performance of various transactions. The operation and maintenance business related to the financial institution can help to maintain market stability and the like, and correspondingly, the accurate classification of the query questions related to the operation and maintenance business can help to timely and effectively return target answers later.
S204: when the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model;
in the embodiment of the invention, the problem matching model comprises at least two scene problem matching models with scene dimensions, the scene problem matching models with at least two scene dimensions are used for performing scene problem matching on the word segmentation result based on the corresponding priority sequence, each scene problem matching model is configured with a corresponding candidate problem library, and each scene problem matching model is obtained by performing machine learning training through the corresponding candidate problem library. In particular, the candidate problem entered into the initial model, which outputs the predictive label for that candidate problem, may carry the actual label. And adjusting model parameters of the initial model according to the difference between the prediction label and the actual label, and taking the initial model corresponding to the adjusted model parameters as a classification model.
The problem matching model is set in a grading manner with finer granularity based on (service) scenes, and can match different scene dimensions in different problem libraries for query problems related to operation and maintenance services. The priorities of the different scenario dimensions may be set in connection with the actual experience (e.g., historical accuracy of returned answers, historical efficiency). Each scene problem matching model is configured with a corresponding candidate problem library, so that management and maintenance work can be performed on the corresponding candidate problem library based on scene dimensions for massive candidate problems.
In a specific embodiment, the at least two scene-dimensional scene question matching models include a multi-round dialog scene question matching model and a single-round dialog scene question matching model. After determining the target problem, the multi-round dialogue scene problem matching model not only returns an answer associated with the target problem, but also determines adjacent intention by combining context information, historical experience and the like, and further returns the recommended problem together with the associated problem. The recommendation questions may guide the user through the next dialog in a more natural and user-friendly manner. The data mining of the multi-round session can improve the efficiency and quality of operation and maintenance and save the labor cost.
As shown in fig. 3, the inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by using the problem matching model includes:
s301: inputting the word segmentation result into the multi-round dialogue scene problem matching model to perform scene problem matching, wherein the multi-round dialogue scene problem matching model corresponds to a first candidate problem library;
s302: when the multi-round dialogue scene problem matching model determines a matched first candidate problem from the first candidate problem library, the first candidate problem is used as the target problem;
s303: when the multi-round dialogue scene problem matching model does not determine the matched problem from the first candidate problem library, inputting the word segmentation result into the single-round dialogue scene problem matching model to perform scene problem matching, wherein the single-round dialogue scene problem matching model corresponds to a second candidate problem library;
s304: and when the single-round dialogue scene problem matching model determines a matched second candidate problem from the second candidate problem library, the second candidate problem is used as the target problem.
The multi-round session engine (corresponding to the multi-round dialogue scene problem matching model) and the single-round session engine (corresponding to the multi-round dialogue scene problem matching model) are adopted, and in the specific embodiment, the opposite-round session engine is utilized first and then the single-round session engine is utilized, so that the accuracy of determining the intention of the user can be improved, and the efficiency of determining the target problem and then returning the target answer to the user can be improved.
In addition, the recommendation problem can be carried in the return of the multi-round session engine, and compared with the return of the single-round session engine only to the target answer, the data size is larger. The method can be combined with the current communication network state between the clients, and when the communication network state is poor, a multi-round session engine is skipped and the single-round session engine is directly used for problem matching.
Further, as shown in fig. 4, when the single-turn dialogue scene question matching model does not determine the matched question from the second candidate question library, the method further includes:
s305: constructing a problem template based on the word segmentation result;
the problem template is constructed by performing the steps of: firstly, determining a problem keyword based on the word segmentation result and configuring a corresponding second weight for the problem keyword; then, acquiring a weight threshold value, and taking the problem key word with the corresponding second weight larger than the weight threshold value as a core key word; then, a question template is obtained based on the word segmentation result and the core keyword.
The question keywords may be obtained by performing word segmentation processing on each candidate question in the second candidate question library by using a Stanford CoreNLP technology. The second weight corresponding to the question key may be configured using TF-IDF (Term Frequency) -Inverse Document Frequency (inverse text Frequency index) technique, a common weighting technique for information retrieval and data mining. In determining the core keywords, on one hand, the problem keywords with the corresponding second weight greater than the first weight threshold (such as 0.6) can be directly used as the core keywords; on the one hand, when the query problem is relatively long, the first weight threshold and the target word number (for example, the number of times corresponding to the at least two word segmentation data is greater than 1/2) can be combined to determine the core keyword, so that the matching accuracy is ensured. After the core keywords are determined, the core keywords can be arranged according to the syntactic relation of the core keywords in the query questions to generate question templates, and the question templates are generally in the format of: a B C D.
S306: and matching the problem template based on the second candidate problem library and the synonym library.
A group of words with the same meaning are classified into a class in the synonym library, and are shown by [ $X ]. For example, [ $fail ], which represents a group of words that have the same meaning as 'fail'.
And the problem templates are generated, so that the problem library is more convenient to maintain. When matching the problem templates, the problem templates may be parsed first; then, the synonym corresponding to each core keyword is taken out from the synonym library; then, converting the problem template into a complete template by using the corresponding synonyms; furthermore, matching the complete template based on the second candidate problem library, and returning a matching failure if no candidate problem with the matching degree exceeding a matching degree threshold value exists in the second candidate problem library; if so, the candidate problem is determined to be the target problem.
In another specific embodiment, each scene problem matching model includes a first-level sub-model and a second-level sub-model, and the process of inputting the word segmentation result into the scene problem matching model to perform scene problem matching is implemented by executing the following steps: 1) Inputting the word segmentation result into the first-level submodel to perform scene problem matching; 2) When the first-level submodel determines a matched third candidate problem from the corresponding candidate problem library, the third candidate problem is used as the target problem; 3) When the first-stage submodel does not determine the matched problem from the corresponding candidate problem library, inputting the word segmentation result into the second-stage submodel for scene problem matching; 4) When the second-level sub-model determines a fourth candidate problem which is matched with the second-level sub-model from the corresponding candidate problem library, the fourth candidate problem is used as the target problem; 5) And when the second-level sub-model does not determine the matched problem from the corresponding candidate problem library, inputting the word segmentation result into a scene problem matching model with a priority lower than that of the current scene problem matching model to perform scene problem matching.
In particular, the first level sub-model employs a transformer-based bi-directional coded representation model (BERT model). Short text similarity analysis tasks performed by the BERT model may be categorized on Natural Language Processing (NLP) as classification tasks, with classes mainly similar (1) and dissimilar (0). Scene problem matching is carried out based on the BERT model, so that accuracy of determining a target answer can be ensured.
Before the step of inputting the word segmentation result into the first-level submodel for scene problem matching, a word vector corresponding to each word segmentation data can be obtained according to the word segmentation result; performing sentence meaning segmentation on the corpus data to be processed according to the word segmentation result to obtain at least one piece of segmentation data and a segment vector corresponding to the at least one piece of segmentation data; and obtaining a position vector corresponding to each word segmentation data according to the word segmentation result.
The input representation of the first level sub-model may represent a single text sentence or a pair of texts (e.g., two text sentences) in a word sequence. For a given word, its input representation may be composed by a three-part summation: 1) Token symbols: word vectors, the first word being a CLS flag (which may be used for later classification tasks; word vectors may be ignored for non-classification tasks); 2) Segment Embeddings: segment vectors for distinguishing two sentences; 3) Position Embeddings: a position vector.
As shown in fig. 5, the word segmentation result may be preprocessed by using a pre-training model, where fine-tuning (fine-tuning) is implemented based on a pyrerch framework (a neural network framework). For example, the corpus data to be processed corresponds to: "how to transact construction credit card" and "how to transact construction bank credit card", the at least two word segmentation data correspond to: how, transact, build, credit card, and how, transact, build, bank, credit card. Accordingly, the result of preprocessing the word segmentation result by using the pre-training model can be seen in the following table 1:
TABLE 1
Where [ CLS ] is the starting symbol for the BERT model, [ SEP ] is used to segment two sentences. Of course, the corpus data to be processed may also be input to the pre-training model. The pre-training model also performs classification tasks on the input in two text sentences.
Correspondingly, the corresponding word vector, the corresponding segment vector and the corresponding position vector are input into the first-level submodel to perform scene problem matching. And taking the candidate problem with the largest similarity with the query problem in the corresponding candidate problem library as a target problem in the first-stage submodel.
When training the first-level submodel, segments_ids may be used to label whether the problem of the current input model belongs to the corresponding candidate problem library, and the segment of the problem of the current input model is segmented by a token (refer to the above-mentioned processing procedure of obtaining the word vector, the segment vector and the position vector) to obtain the label send_token, and then send_ token, segments _ids and the problem of the current input model together into the initial network for training.
Specifically, the second-level sub-model adopts a probability retrieval model, and the probability retrieval model can be trained based on a BM25 algorithm (used for scoring retrieval relevance). The probability search model is utilized to match scene problems, so that the actual search results in the corresponding candidate problem library can be focused more, efficiency and accuracy can be better considered when the target problems are determined, and the hit rate ignored by the BERT model can be made up to a certain extent. As shown in fig. 6, the inputting the word segmentation result into the second-stage submodel for scene problem matching includes:
s601: calculating a first relevance of each word segmentation data and each candidate problem in the corresponding candidate problem library;
For example, the corpus data to be processed corresponds to Query question Query "how to archive mailbox", and the at least two word segmentation numbersAccording to the corresponding mailbox, how and archive, each word is divided by q i And (3) representing. D for each candidate question in the corresponding candidate question library D j And (3) representing. Correspondingly, calculate q i And d j Is the first correlation of (1), i.e., q 1 (corresponding to "mailbox") and d j Is the first correlation, q 2 (corresponding to how) and d j Is the first correlation, q 3 (corresponding "archiving") and d j Is a first correlation of (a).
The following is q i And d j Correlation degree R (q) i ，d j ) Is calculated according to the formula:
wherein,,
k 1 、k 2 b is an adjustment factor, and the parameter b is used to control the influence degree of the document length on the relevance value, and is generally set to k according to experience 1 ＝1、k 2 ＝2、b＝0.75；f i Is q i At d j Frequency (times), qf of occurrence i Is q i Frequency (number of times) of occurrence in Query. dl is d j Avgdl is all D in D j Is used for the average document length of (a). Q in most cases i Only once in Query, qf i =1, so the formula can be reduced to:
from the formula of K, we find: the larger b is, the larger the influence of the document length on the relevance value is; and the longer the document length is, the larger the K is, and the smaller the relevance value is. When the document length is longer, the word q is included i The greater the probability of (1), and therefore the longer the document and q at the same frequency i Is not as relevant as a short document to q i Is related to the degree of correlation of (2)High.
S602: configuring a corresponding first weight for each word segmentation data based on the corresponding candidate question library;
a corresponding first weight may be configured for each of the word segmentation data based on an IDF (inverse text frequency index) technique:
where N represents the total number of documents (corresponding to all candidate questions) of the corresponding candidate question library D, N (q i ) Representing that the corresponding candidate question library D contains the segmentation word q i Is a number of documents of candidate questions. From the formula, it can be found that when the word q is included i Q is the more the number of documents is i The lower the weight of (i) when many documents have a word q i At the time, the word q is described i The word q is a word, which is commonly used and has no special meaning i The importance is lower when the correlation judgment is made.
S603: obtaining a second correlation degree between the corpus data to be processed and each candidate problem in the corresponding candidate problem library based on the corresponding first weight and the first correlation degree between each word segmentation data and each candidate problem in the corresponding candidate problem library;
the following are Q and d j Correlation Score (Q, d) j ) Is calculated according to the formula:
obtaining a Query question Query and each candidate question D in the corresponding candidate question library D according to the formula j And a second degree of correlation. In combination with the above W i And K is calculated according to the formula:
s604: and determining target relevance meeting the relevance threshold requirement from the second relevance corresponding to the corpus data to be processed, and taking the candidate problem corresponding to the target relevance as the target problem.
And taking the candidate problem with the largest similarity with the query problem in the corresponding candidate problem library as a target problem in the first-stage submodel.
In practical applications, the problem matching model may include a multi-round session engine (corresponding to a multi-round dialog scene problem matching model) and a single-round session engine (corresponding to a multi-round dialog scene problem matching model). The multi-round session engine comprises a BERT model and a probability retrieval model: 1) firstly matching in a multi-round conversation problem library by using a BERT model, 2) returning corresponding answers to users based on target questions if the matching is successful, 3) matching in the multi-round conversation problem library by using a probability retrieval model if the matching is unsuccessful, 4) returning corresponding answers to users based on the target questions if the matching is successful, 5) triggering a step of matching scene questions by using a single-round conversation engine if the matching is unsuccessful. The single-round session engine comprises a BERT model and a probability retrieval model: 1) firstly matching in a single-round conversation problem library by using a BERT model, 2) returning corresponding answers to users based on target questions if matching is successful, 3) matching in the single-round conversation problem library by using a probability retrieval model if matching is unsuccessful, 4) returning corresponding answers to users based on the target questions if matching is successful, 5) triggering the steps of constructing a problem template and matching the problem template based on the single-round conversation problem library and a synonym library if matching is unsuccessful.
S205: acquiring an answer associated with the target question, and taking the associated answer as a target answer to be returned;
in the embodiment of the invention, the candidate questions and the associated answers thereof in the candidate question library are stored in a database, and the candidate questions and the associated answers thereof are stored in the database based on the association relation. And after determining the target questions matched with the word segmentation results from the candidate question library, determining associated answers based on the association relation, and returning the associated answers to the user as target answers.
As shown in fig. 7, when the result obtained by classification indicates a non-operation service, the method further includes:
s701: obtaining a characteristic word set based on the word segmentation result;
s702: determining word frequency and inverse text frequency index corresponding to each feature word in the feature word set based on the reference question library;
s703: obtaining a weight value corresponding to each feature word based on the corresponding word frequency and the inverse text frequency index;
s704: constructing a first vector based on the corresponding weight value and a reference word set of the reference question library;
s705: respectively calculating the similarity of the first vector and a reference vector corresponding to each reference problem in the reference problem library;
S706: and determining a reference question corresponding to the maximum similarity, and taking an answer associated with the reference question corresponding to the maximum similarity as a target answer to be returned.
When the classification result indicates a non-operation service (such as boring), a feature word set can be obtained based on the word segmentation result, and then a corresponding weight value is configured for each feature word in the feature word set by using a TF-IDF technology:
1) Based on the reference question library D (e.g., the boring question library), each feature word T in the feature word set T is determined as follows i The corresponding word frequency:
wherein,,
is the characteristic word t i Number of occurrences, W, in reference problem library D d Is the total word number in the reference question library D;
2) Based on a reference question library D (e.g., a boring question library), each of the feature word sets T is determined as followsFeature word t i The corresponding inverse text frequency index:
where N is the total number of documents (corresponding to all reference questions) of the reference question library D,
for inclusion of feature words t in the reference question library D i Reference to the number of documents of the problem.
3) The term frequency may characterize the ability of the feature word to describe the content of the document, and the inverse text frequency index may characterize the ability of the feature word to distinguish between documents. Feature word t i Weight value=tf (t i ，D)*IDF(t i D) is described. When a feature word appears more frequently in a particular reference question, and also appears less frequently in the reference question, its weight value is also greater.
Then, constructing a first vector based on the corresponding weight value and the reference word set of the reference question library, and the first vector can be a one-dimensional vector; and then respectively calculating the similarity of the first vector and the reference vector corresponding to each reference problem in the reference problem library, wherein the similarity calculation can be performed by adopting one of the following steps: euclidean distance, cosine similarity, relative entropy; and finally, determining a reference question corresponding to the maximum similarity, and taking an answer associated with the reference question corresponding to the maximum similarity as a target answer to be returned.
The technical scheme provided by the embodiment of the specification can be seen that the machine learning model for service classification and problem matching is introduced into the intelligent question-answering system in the embodiment of the specification, the learning ability of the machine learning model for data representation is stronger, and the accuracy and adaptability of selecting matched problems from the problem library for the corpus data to be processed can be improved. Returning the answer associated with the question as the target answer can improve the accuracy and efficiency of determining the target answer. Based on the difference of the service and the dimension of the (service) scene, the corresponding problem library is configured for the related model, so that the problem matching efficiency is improved, and meanwhile, the problem library is convenient to manage and maintain.
The embodiment of the invention also provides an intelligent question-answer data processing device based on the operation and maintenance service, as shown in fig. 8, the device comprises:
corpus data acquisition module 810: the method comprises the steps of obtaining corpus data to be processed of a target object;
word segmentation processing module 820: the method comprises the steps of performing word segmentation on corpus data to be processed to obtain word segmentation results, wherein the word segmentation results comprise at least two word segmentation data and corpus analysis results, and the corpus analysis results comprise at least one of the following: part-of-speech analysis results, grammar analysis results, entity analysis results and emotion analysis results;
the traffic classification module 830: the classification model is obtained through machine learning training of a plurality of service sample data, and the service sample data carries corresponding service classification labeling information;
problem-matching module 840: when the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model;
Answer acquisition module 850: the method comprises the steps of obtaining an answer associated with the target question, and taking the associated answer as a target answer to be returned;
the problem matching models comprise scene problem matching models with at least two scene dimensions, the scene problem matching models with at least two scene dimensions are used for performing scene problem matching on the word segmentation result based on corresponding priority sequences, each scene problem matching model is configured with a corresponding candidate problem library, and each scene problem matching model is obtained through machine learning training of the corresponding candidate problem library.
It should be noted that the apparatus and method embodiments in the apparatus embodiments are based on the same inventive concept.
The embodiment of the invention provides electronic equipment, which comprises a processor and a memory, wherein at least one instruction or at least one section of program is stored in the memory, and the at least one instruction or the at least one section of program is loaded and executed by the processor to realize the intelligent question-answer data processing method based on operation and maintenance services, which is provided by the embodiment of the method.
Further, fig. 9 shows a schematic hardware structure of an electronic device for implementing the intelligent question-answer data processing method based on the operation and maintenance service provided by the embodiment of the present invention, where the electronic device may participate in forming or including the intelligent question-answer data processing device based on the operation and maintenance service provided by the embodiment of the present invention. As shown in fig. 9, the electronic device 90 may include one or more processors 902 (shown in the figures as 902a, 902b, … …,902 n) (the processor 902 may include, but is not limited to, a microprocessor MCU or a processing device such as a programmable logic device FPGA), a memory 904 for storing data, and a transmission device 906 for communication functions. In addition, the method may further include: a display, an input/output interface (I/O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the I/O interface), a network interface, a power supply, and/or a camera. It will be appreciated by those skilled in the art that the configuration shown in fig. 9 is merely illustrative and is not intended to limit the configuration of the electronic device. For example, the electronic device 90 may also include more or fewer components than shown in FIG. 9, or have a different configuration than shown in FIG. 9.
It should be noted that the one or more processors 902 and/or other data processing circuitry described above may be referred to herein generally as "data processing circuitry. The data processing circuit may be embodied in whole or in part in software, hardware, firmware, or any other combination. Further, the data processing circuitry may be a single stand-alone processing module, or incorporated, in whole or in part, into any of the other elements in the electronic device 90 (or mobile device). As referred to in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of the path of the variable resistor termination to interface).
The memory 904 may be used to store software programs and modules of application software, and the processor 902 executes the software programs and modules stored in the memory 94 by executing program instructions/data storage devices corresponding to the methods according to the embodiments of the present invention, so as to perform various functional applications and data processing, that is, implement an intelligent question-answer data processing method based on operation and maintenance services. The memory 904 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 904 may further include memory remotely located relative to the processor 902, which may be connected to the electronic device 90 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The transmission means 906 is used for receiving or transmitting data via a network. Specific examples of the network described above may include a wireless network provided by a communications provider of the electronic device 90. In one example, the transmission means 906 includes a network adapter (NetworkInterfaceController, NIC) that can be connected to other network devices through a base station to communicate with the internet. In one embodiment, the transmission device 906 may be a radio frequency (RadioFrequency, RF) module for communicating wirelessly with the internet.
The display may be, for example, a touch screen type Liquid Crystal Display (LCD) that may enable a user to interact with a user interface of the electronic device 90 (or mobile device).
The embodiment of the invention also provides a storage medium which can be arranged in the electronic equipment to store at least one instruction or at least one section of program related to the intelligent question-answer data processing method based on the operation and maintenance service in the method embodiment, and the at least one instruction or the at least one section of program is loaded and executed by the processor to realize the intelligent question-answer data processing method based on the operation and maintenance service provided by the method embodiment.
Alternatively, in this embodiment, the storage medium may be located in at least one network server among a plurality of network servers of the computer network. Alternatively, in the present embodiment, the storage medium may include, but is not limited to: a U-disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing program codes.
It should be noted that: the sequence of the embodiments of the present invention is only for description, and does not represent the advantages and disadvantages of the embodiments. And the foregoing description has been directed to specific embodiments of this specification. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for the apparatus and electronic device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and references to the parts of the description of the method embodiments are only required.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (11)
1. An intelligent question-answer data processing method based on operation and maintenance service is characterized by comprising the following steps:
obtaining corpus data to be processed of a target object;
word segmentation is carried out on the corpus data to be processed to obtain word segmentation results, the word segmentation results comprise at least two word segmentation data and corpus analysis results, and the corpus analysis results comprise at least one of the following: part-of-speech analysis results, grammar analysis results, entity analysis results and emotion analysis results;
inputting the word segmentation result into a classification model for service classification, wherein the classification model is obtained by performing machine learning training on a plurality of service sample data, and the service sample data carries corresponding service classification labeling information;
When the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model;
acquiring an answer associated with the target question, and taking the associated answer as a target answer to be returned;
the problem matching models comprise at least two scene-dimensional scene problem matching models, wherein the at least two scene-dimensional scene problem matching models comprise a multi-round dialogue scene problem matching model and a single-round dialogue scene problem matching model, the at least two scene-dimensional scene problem matching models match the word segmentation result based on a corresponding priority sequence, each scene problem matching model is configured with a corresponding candidate problem library, and each scene problem matching model is obtained through machine learning training of the corresponding candidate problem library; each scene problem matching model comprises a first-level sub-model and a second-level sub-model, the first-level sub-model adopts a bidirectional coding representation model based on a transformer, the second-level sub-model adopts a probability retrieval model, and the process of inputting the word segmentation result into the scene problem matching model for scene problem matching is realized by executing the following steps: inputting the word segmentation result into the first-level submodel to perform scene problem matching; when the first-level submodel determines a matched third candidate problem from the corresponding candidate problem library, the third candidate problem is used as the target problem; when the first-stage submodel does not determine the matched problem from the corresponding candidate problem library, inputting the word segmentation result into the second-stage submodel for scene problem matching; when the second-level sub-model determines a fourth candidate problem which is matched with the second-level sub-model from the corresponding candidate problem library, the fourth candidate problem is used as the target problem; and when the second-level sub-model does not determine the matched problem from the corresponding candidate problem library, inputting the word segmentation result into a scene problem matching model with a priority lower than that of the current scene problem matching model to perform scene problem matching.
2. The method of claim 1, wherein before the inputting the word segmentation result into the classification model for traffic classification, the method further comprises:
mapping the word segmentation result to obtain a word segmentation result vector;
correspondingly, the step of inputting the word segmentation result into a classification model to classify the business comprises the following steps:
inputting the word segmentation result vector into the classification model;
extracting service features of the word segmentation result vector by utilizing a convolution layer of the classification model;
outputting the service classification result based on the service characteristics by the full connection layer of the classification model.
3. The method of claim 1, wherein inputting the word segmentation result into a question matching model that matches the operation and maintenance business, and determining a target question that matches the word segmentation result from a candidate question library using the question matching model, comprises:
inputting the word segmentation result into the multi-round dialogue scene problem matching model to perform scene problem matching, wherein the multi-round dialogue scene problem matching model corresponds to a first candidate problem library;
when the multi-round dialogue scene problem matching model determines a matched first candidate problem from the first candidate problem library, the first candidate problem is used as the target problem;
When the multi-round dialogue scene problem matching model does not determine the matched problem from the first candidate problem library, inputting the word segmentation result into the single-round dialogue scene problem matching model to perform scene problem matching, wherein the single-round dialogue scene problem matching model corresponds to a second candidate problem library;
and when the single-round dialogue scene problem matching model determines a matched second candidate problem from the second candidate problem library, the second candidate problem is used as the target problem.
4. The method of claim 1, wherein the first level sub-model employs a transformer-based bi-directional coded representation model, and wherein prior to inputting the word segmentation result into the first level sub-model for scene problem matching, the method further comprises:
obtaining word vectors corresponding to each word segmentation data according to the word segmentation result;
performing sentence meaning segmentation on the corpus data to be processed according to the word segmentation result to obtain at least one piece of segmentation data and a segment vector corresponding to the at least one piece of segmentation data;
obtaining a position vector corresponding to each word segmentation data according to the word segmentation result;
correspondingly, the step of inputting the word segmentation result into the first-stage submodel to perform scene problem matching includes:
And inputting the corresponding word vector, the corresponding segment vector and the corresponding position vector into the first-level submodel to perform scene problem matching.
5. The method according to claim 1, wherein the second level sub-model adopts a probability retrieval model, and the inputting the word segmentation result into the second level sub-model for scene problem matching comprises:
calculating a first relevance of each word segmentation data and each candidate problem in the corresponding candidate problem library;
configuring a corresponding first weight for each word segmentation data based on the corresponding candidate question library;
obtaining a second correlation degree between the corpus data to be processed and each candidate problem in the corresponding candidate problem library based on the corresponding first weight and the first correlation degree between each word segmentation data and each candidate problem in the corresponding candidate problem library;
and determining target relevance meeting the relevance threshold requirement from the second relevance corresponding to the corpus data to be processed, and taking the candidate problem corresponding to the target relevance as the target problem.
6. The method of claim 3, wherein when the single round dialog scenario problem matching model does not determine a matching problem from the second candidate problem library, the method further comprises:
Constructing a problem template based on the word segmentation result;
and matching the problem template based on the second candidate problem library and the synonym library.
7. The method of claim 6, wherein the problem template is constructed by performing the steps of:
determining a problem keyword based on the word segmentation result, and configuring a corresponding second weight for the problem keyword;
acquiring a weight threshold value, and taking the problem key word with the corresponding second weight larger than the weight threshold value as a core key word;
and obtaining a problem template based on the word segmentation result and the core keyword.
8. The method of claim 1, wherein when the result of the classification indicates non-operational services, the method further comprises:
obtaining a characteristic word set based on the word segmentation result;
determining word frequency and inverse text frequency index corresponding to each feature word in the feature word set based on the reference question library;
obtaining a weight value corresponding to each feature word based on the corresponding word frequency and the inverse text frequency index;
constructing a first vector based on the corresponding weight value and a reference word set of the reference question library;
Respectively calculating the similarity of the first vector and a reference vector corresponding to each reference problem in the reference problem library;
and determining a reference question corresponding to the maximum similarity, and taking an answer associated with the reference question corresponding to the maximum similarity as a target answer to be returned.
9. An intelligent question-answering data processing device based on operation and maintenance service, which is characterized by comprising:
corpus data acquisition module: the method comprises the steps of obtaining corpus data to be processed of a target object;
the word segmentation processing module: the method comprises the steps of performing word segmentation on corpus data to be processed to obtain word segmentation results, wherein the word segmentation results comprise at least two word segmentation data and corpus analysis results, and the corpus analysis results comprise at least one of the following: part-of-speech analysis results, grammar analysis results, entity analysis results and emotion analysis results;
and a service classification module: the classification model is obtained through machine learning training of a plurality of service sample data, and the service sample data carries corresponding service classification labeling information;
and a problem matching module: when the classified result indicates operation and maintenance service, inputting the word segmentation result into a problem matching model matched with the operation and maintenance service, and determining a target problem matched with the word segmentation result from a candidate problem library by utilizing the problem matching model;
Answer acquisition module: the method comprises the steps of obtaining an answer associated with the target question, and taking the associated answer as a target answer to be returned;
the problem matching models comprise at least two scene-dimensional scene problem matching models, wherein the at least two scene-dimensional scene problem matching models comprise a multi-round dialogue scene problem matching model and a single-round dialogue scene problem matching model, the at least two scene-dimensional scene problem matching models match the word segmentation result based on a corresponding priority sequence, each scene problem matching model is configured with a corresponding candidate problem library, and each scene problem matching model is obtained through machine learning training of the corresponding candidate problem library; each scene problem matching model comprises a first-level sub-model and a second-level sub-model, the first-level sub-model adopts a bidirectional coding representation model based on a transformer, the second-level sub-model adopts a probability retrieval model, and the process of inputting the word segmentation result into the scene problem matching model for scene problem matching is realized by executing the following steps: inputting the word segmentation result into the first-level submodel to perform scene problem matching; when the first-level submodel determines a matched third candidate problem from the corresponding candidate problem library, the third candidate problem is used as the target problem; when the first-stage submodel does not determine the matched problem from the corresponding candidate problem library, inputting the word segmentation result into the second-stage submodel for scene problem matching; when the second-level sub-model determines a fourth candidate problem which is matched with the second-level sub-model from the corresponding candidate problem library, the fourth candidate problem is used as the target problem; and when the second-level sub-model does not determine the matched problem from the corresponding candidate problem library, inputting the word segmentation result into a scene problem matching model with a priority lower than that of the current scene problem matching model to perform scene problem matching.
10. An electronic device comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the intelligent question-answering data processing method based on operation and maintenance services according to any one of claims 1-8.
11. A computer readable storage medium having stored therein at least one instruction or at least one program loaded and executed by a processor to implement the intelligent query and answer data processing method based on an operation and maintenance service according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911403393.XA CN111241237B (en) | 2019-12-31 | 2019-12-31 | Intelligent question-answer data processing method and device based on operation and maintenance service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911403393.XA CN111241237B (en) | 2019-12-31 | 2019-12-31 | Intelligent question-answer data processing method and device based on operation and maintenance service |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111241237A CN111241237A (en) | 2020-06-05 |
| CN111241237B true CN111241237B (en) | 2023-05-23 |
Family
ID=70871865
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911403393.XA Active CN111241237B (en) | 2019-12-31 | 2019-12-31 | Intelligent question-answer data processing method and device based on operation and maintenance service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111241237B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111858887B (en) * | 2020-07-13 | 2022-09-20 | 北京航空航天大学 | A Community Question Answering System for Airport Service |
| CN111782792A (en) * | 2020-08-05 | 2020-10-16 | 支付宝(杭州)信息技术有限公司 | Method and apparatus for information processing |
| CN112100355A (en) * | 2020-09-17 | 2020-12-18 | 中国建设银行股份有限公司 | Intelligent interaction method, device and equipment |
| CN112214654A (en) * | 2020-10-19 | 2021-01-12 | 厦门渊亭信息科技有限公司 | Universal intelligent question-answering automatic operation and maintenance system and method |
| CN112380853B (en) * | 2020-11-16 | 2024-04-26 | 中国平安人寿保险股份有限公司 | Service scene interaction method and device, terminal equipment and storage medium |
| CN112632242A (en) * | 2020-12-17 | 2021-04-09 | 竹间智能科技（上海）有限公司 | Intelligent conversation method and device and electronic equipment |
| CN115081441B (en) * | 2021-03-11 | 2025-01-07 | 中国移动通信集团江苏有限公司 | Method, device, electronic device and storage medium for identifying operation and maintenance instruction intention |
| CN112905752A (en) * | 2021-03-30 | 2021-06-04 | 中国建设银行股份有限公司 | Intelligent interaction method, device, equipment and storage medium |
| CN113377904B (en) * | 2021-06-04 | 2024-05-10 | 百度在线网络技术（北京）有限公司 | Industry action recognition method and device, electronic equipment and storage medium |
| CN113094474B (en) * | 2021-06-08 | 2022-05-10 | 深圳追一科技有限公司 | Intelligent question answering method and device, server and computer readable storage medium |
| CN113553405A (en) * | 2021-06-11 | 2021-10-26 | 中国农业银行股份有限公司浙江省分行 | Implementation method and system of intelligent robot based on Chinese bert model |
| CN113553412B (en) * | 2021-06-30 | 2023-07-25 | 北京百度网讯科技有限公司 | Question answering method, device, electronic device and storage medium |
| CN113590778A (en) * | 2021-06-30 | 2021-11-02 | 上海一谈网络科技有限公司 | Intelligent customer service intention understanding method, device, equipment and storage medium |
| CN114064876A (en) * | 2021-11-26 | 2022-02-18 | 弘成科技发展有限公司 | A method of realizing intelligent student service based on artificial intelligence BERT model |
| CN115344690A (en) * | 2021-11-30 | 2022-11-15 | 中国银联股份有限公司 | Data processing method and device for business problems |
| CN115689477A (en) * | 2022-10-12 | 2023-02-03 | 中国工商银行股份有限公司 | Service scheme generation method and device, computer equipment and storage medium |
| CN116521869A (en) * | 2023-04-26 | 2023-08-01 | 北京华玉通软科技有限公司 | A Problem Management Method Based on Functional Safety Standards and Basic Quality Standards |
| CN116955568A (en) * | 2023-08-02 | 2023-10-27 | 浙江吉利控股集团有限公司 | Question answering method and device based on instruction manual, electronic equipment and storage medium |
| CN119007723B (en) * | 2024-08-13 | 2025-04-01 | 广州视声智能科技有限公司 | Response method and system based on artificial intelligence |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102637192A (en) * | 2012-02-17 | 2012-08-15 | 清华大学 | Method for answering with natural language |
| CN108153780A (en) * | 2016-12-05 | 2018-06-12 | 阿里巴巴集团控股有限公司 | A man-machine dialogue device and method for realizing man-machine dialogue |
| CN109522393A (en) * | 2018-10-11 | 2019-03-26 | 平安科技（深圳）有限公司 | Intelligent answer method, apparatus, computer equipment and storage medium |
| CN110309283A (en) * | 2019-06-28 | 2019-10-08 | 阿里巴巴集团控股有限公司 | A kind of answer of intelligent answer determines method and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9690776B2 (en) * | 2014-12-01 | 2017-06-27 | Microsoft Technology Licensing, Llc | Contextual language understanding for multi-turn language tasks |
-
2019
- 2019-12-31 CN CN201911403393.XA patent/CN111241237B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102637192A (en) * | 2012-02-17 | 2012-08-15 | 清华大学 | Method for answering with natural language |
| CN108153780A (en) * | 2016-12-05 | 2018-06-12 | 阿里巴巴集团控股有限公司 | A man-machine dialogue device and method for realizing man-machine dialogue |
| CN109522393A (en) * | 2018-10-11 | 2019-03-26 | 平安科技（深圳）有限公司 | Intelligent answer method, apparatus, computer equipment and storage medium |
| CN110309283A (en) * | 2019-06-28 | 2019-10-08 | 阿里巴巴集团控股有限公司 | A kind of answer of intelligent answer determines method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111241237A (en) | 2020-06-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111241237B (en) | Intelligent question-answer data processing method and device based on operation and maintenance service | |
| CN113569011B (en) | Training method, device and equipment of text matching model and storage medium | |
| US11334635B2 (en) | Domain specific natural language understanding of customer intent in self-help | |
| CN112131350B (en) | Text label determining method, device, terminal and readable storage medium | |
| CN111695338B (en) | Methods, apparatus, equipment, and media for refining interview content based on artificial intelligence. | |
| US20200372025A1 (en) | Answer selection using a compare-aggregate model with language model and condensed similarity information from latent clustering | |
| WO2021121198A1 (en) | Semantic similarity-based entity relation extraction method and apparatus, device and medium | |
| CN114580382A (en) | Text error correction method and device | |
| CN112287069B (en) | Information retrieval method and device based on voice semantics and computer equipment | |
| CN111931517A (en) | Text translation method and device, electronic equipment and storage medium | |
| CN110347790B (en) | Text duplicate checking method, device and equipment based on attention mechanism and storage medium | |
| CN116913278B (en) | Voice processing method, device, equipment and storage medium | |
| CN113342944B (en) | Corpus generalization method, apparatus, device and storage medium | |
| CN113505293B (en) | Information pushing method and device, electronic equipment and storage medium | |
| CN110825867A (en) | Similar text recommendation method, device, electronic device and storage medium | |
| CN115438149A (en) | End-to-end model training method and device, computer equipment and storage medium | |
| CN117933231B (en) | Text error correction and error correction model training method and device, electronic equipment and medium | |
| CN119783672B (en) | Corpus expansion method, corpus expansion equipment and storage medium | |
| CN116361638A (en) | Question answer search method, device and storage medium | |
| CN115879458A (en) | Corpus expansion method, apparatus and storage medium | |
| CN114328820A (en) | Information searching method and related equipment | |
| CN114707489A (en) | Method and device for acquiring marked data set, electronic equipment and storage medium | |
| CN119088963A (en) | Text analysis method and device | |
| CN117216203B (en) | Interactive question-answering method, device, computer equipment and medium | |
| CN119903830A (en) | Method, device, storage medium, electronic device and program product for determining tag information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |