CN110472995A - To shop prediction technique, device, readable storage medium storing program for executing and electronic equipment - Google Patents

Abstract

An embodiment of the present application provides a store arrival prediction method, device, readable storage medium, and electronic equipment, the method including: obtaining a search request initiated by a user terminal; performing feature extraction on the search request, and determining the user terminal's The characteristics of the user, the characteristics of each store in the store list corresponding to the search request, and the user-store intersection feature; A store probability prediction model, determining the probability that the user arrives at each store; and determining whether the user arrives at a target store according to the probability that the user arrives at each store. The present application determines the target store that the user arrives through the probability of each store output by the store arrival probability prediction model, realizes the accurate positioning of the user's current location, and improves the reliability of the stores near the current location recommended by the user terminal to the user.

Description

Arrival prediction method, device, readable storage medium and electronic equipment

technical field

The present application relates to the technical field of information processing, and in particular to a store arrival prediction method, device, readable storage medium and electronic equipment.

Background technique

With the development of the mobile Internet, people can easily access the Internet through mobile devices to obtain services, and a number of O2O (Online-to-Offline) local life-oriented services (such as: O2O nearby search) have emerged, which have a great impact It facilitates people's life. Taking O2O nearby search as an example, users can view food, entertainment and other life information near the previous location through this search function. According to the distance between the location and the merchants marked with latitude and longitude, the merchants whose distance meets the user's screening range are sorted according to a certain algorithm to obtain the screening results, and the screening results are returned to the mobile device used by the user for display.

However, there is a problem in the specific implementation process of the above-mentioned search function: the screening results are overly dependent on the user's location. Merchants cannot be used as screening results and returned to users, which seriously affects users' decision-making and experience.

Therefore, in order to make the calculated distance between the user's location and each merchant marked with latitude and longitude satisfy a certain degree of accuracy, it is necessary to ensure that the user's location meets a relatively high accuracy requirement. In actual situations, the user's real-time positioning is greatly affected by the network conditions. When the user initiates an O2O nearby search in some large buildings (for example: in a shopping mall), the floor information of the large building cannot be located due to ordinary positioning technology. , and the mobile device cannot receive GPS signals indoors, and can only rely on the base station of the network operator to realize the real-time positioning of the user, but the real-time positioning of the user obtained by relying on the base station of the network operator has drift phenomenon, and the real-time positioning of the user indoors cannot be guaranteed. Positioning meets high precision requirements.

Therefore, how to more accurately locate the current location of the user is an urgent problem in this field.

Contents of the invention

Embodiments of the present application provide a store arrival prediction method, device, readable storage medium, and electronic device, which can predict whether a user arrives at a store in real time, and realize accurate positioning of the user's current location.

The first aspect of the embodiment of the present application provides a store arrival prediction method, the method comprising:

obtaining a search request initiated by a user terminal;

Perform feature extraction on the search request, determine the user characteristics of the user terminal, the characteristics of each store in the store list corresponding to the search request, and the user-store intersection feature, the user-store intersection feature is the Obtained by feature crossing between the user's features and the features of each store;

Inputting the characteristics of the user, the characteristics of each store and the user-store cross-feature into a pre-trained store arrival probability prediction model, and determining the probability that the user arrives at each store;

According to the probability that the user arrives at each store, it is determined whether the user arrives at a target store, and the target store is one of the various stores.

Optionally, the step of determining whether the user arrives at the target store according to the probability of the user arriving at each store includes:

If the probability that the user arrives at the target store is greater than a preset probability threshold, determining that the user arrives at the target store; or

When the probability that the user arrives at the target store is greater than a preset probability threshold and the parameter value corresponding to the search request is within a preset effective parameter value range, it is determined that the user arrives at the target store.

Optionally, the method also includes:

Determine the type of user behavior associated with a user visit to the store;

Extract a first type of search record and a second type of search record from the search log of the user terminal, the first type of search record is a search record conforming to the user behavior type, and the second type of search record is a corresponding Search records whose time difference between the search time and the search time of the first type of search records is within a preset duration;

Mark the search records of the first type of search record and the second type of search records that conform to the user behavior type as positive samples, and mark the search records of the second type of search records that do not conform to the user behavior type Search records are marked as negative samples;

According to the positive sample and the negative sample, the preset model is trained to obtain the store arrival probability prediction model.

Optionally, the step of training a preset model according to the positive samples and the negative samples to obtain the prediction model of the probability of arrival at the store includes:

Perform feature extraction on the positive sample and the negative sample respectively, determine the characteristics of the sample users corresponding to the positive sample and the negative sample, and each sample in the store list corresponding to the positive sample and the negative sample The characteristics of the store and the sample user-sample store intersection feature, the sample user-sample store intersection feature is obtained by performing feature intersection on the characteristics of the sample user and the characteristics of each sample store;

Using the features of the sample user, the features of each sample store, and the cross feature of the sample user-sample store as training samples, the preset model is trained to obtain the store arrival probability prediction model.

Optionally, after the step of determining that the user has arrived at the target store, the method further includes:

Extracting from the search log of the user terminal the corresponding search record after the user arrives at the target store at the corresponding search time;

In the case that the extracted search record is a search record for the target store, the extracted search record is marked as a positive sample, and the weight of the positive sample is increased;

If the extracted search record is not a search request for the target store, mark the extracted search record as a negative sample, and reduce the weight of the negative sample; and

The store arrival probability prediction model is updated according to the positive samples after the weight increase and the negative samples after the weight reduction.

Optionally, the characteristics of the user include at least one of the following: the name of the WIFI scanned or connected to by the user terminal and the corresponding signal strength, the device type of the user terminal, the latitude and longitude of the user terminal, the user terminal The IP address of the terminal, the user portrait of the user, and the consumption preference of the user.

Optionally, the characteristics of each store include at least one of the following: the logo of each store, the WIFI name of each store, the average connection or scanning intensity of WIFI of each store, the latitude and longitude of each store, The category to which each store belongs, the price range of commodities sold by each store, the click rate of each store, and the visit rate of each store.

Optionally, the user-store intersection feature is obtained by at least one of the following methods:

determining the linear distance between the user terminal and each store according to the latitude and longitude of the user terminal and the latitude and longitude of each store;

The signal strength of the WIFI of the shop that the user terminal scans or connects to, performs feature crossover with the WIFI average connection or scan strength of the shop that the user terminal scans or connects to;

performing feature intersection on the WIFI signal strength of the store scanned or connected to by the user terminal, and the linear distance between the user terminal and the store scanned or connected to by the user terminal; and/or

The feature intersection is performed on the user click or consumption price of the user terminal and the per capita price of each store.

The second aspect of the embodiment of the present application provides a store arrival prediction device, the device comprising:

An obtaining module, configured to obtain a search request initiated by a user terminal;

The feature extraction module is used to perform feature extraction on the search request, determine the features of the user of the user terminal, the features of each store in the store list corresponding to the search request, and the user-store intersection feature, the user-store The intersection feature is obtained by performing feature intersection on the features of the user and the features of each store;

The probability prediction module is used to input the characteristics of the user, the characteristics of the various stores and the cross-features of the user-store into the pre-trained store arrival probability prediction model to determine the probability that the user arrives at the various stores; and

A determining module, configured to determine whether the user arrives at a target store according to the probability that the user arrives at each store, and the target store is one of the various stores.

Optionally, the determination module includes:

A first determining module, configured to determine that the user arrives at the target store when the probability of the user arriving at the target store is greater than a preset probability threshold; or

The second determination module is configured to determine the target store when the probability of the user arriving at the target store is greater than a preset probability threshold and the parameter value corresponding to the search request is within the preset effective parameter value range. The user arrives at the target store.

Optionally, the device also includes:

The third determination module is used to determine the type of user behavior associated with the user's visit to the store;

A first extracting module, configured to extract a first type of search record and a second type of search record from the search log of the user terminal, the first type of search record is a search record conforming to the user behavior type, the second type of search record The second type of search record is a search record in which the time difference between the corresponding search time and the search time of the first type of search record is within a preset duration;

A marking module, configured to mark the search records of the first type of search record and the second type of search records that match the user behavior type as positive samples, and mark the search records of the second type of search records that do not meet the user behavior type. The search records of the above user behavior types are marked as negative samples;

The training module is used to train a preset model according to the positive samples and the negative samples to obtain the prediction model of the arrival probability.

Optionally, the training module includes:

The feature extraction sub-module is used to perform feature extraction on the positive samples and the negative samples respectively, determine the characteristics of the sample users corresponding to the positive samples and the negative samples, and determine the characteristics of the sample users corresponding to the positive samples and the negative samples respectively. The features of each sample store in the corresponding store list and the sample user-sample store cross feature, the sample user-sample store cross feature is obtained by feature crossing the features of the sample user and the features of each sample store;

The training sub-module is used to use the characteristics of the sample user, the characteristics of each sample store, and the cross feature of the sample user-sample store as training samples to train the preset model to obtain the store arrival probability predictive model.

Optionally, the device also includes:

The second extraction module is used to extract from the search log of the user terminal the corresponding search record after the user arrives at the target store at the corresponding search time;

The first weight adjustment module is used to mark the extracted search record as a positive sample and increase the weight of the positive sample when the extracted search record is a search record for the target store;

A second weight adjustment module, configured to mark the extracted search record as a negative sample and reduce the weight of the negative sample when the extracted search record is not a search record for the target store; and

An update module, configured to update the store arrival probability prediction model according to the weighted positive samples and the weighted negative samples.

Optionally, the characteristics of the user include at least one of the following: the name of the WIFI scanned or connected to by the user terminal and the corresponding signal strength, the device type of the user terminal, the latitude and longitude of the user terminal, the user terminal The IP address of the terminal, the user portrait of the user, and the consumption preference of the user.

Optionally, the characteristics of each store include at least one of the following: the logo of each store, the WIFI name of each store, the average connection or scanning intensity of WIFI of each store, the latitude and longitude of each store, The category to which each store belongs, the price range of commodities sold by each store, the click rate of each store, and the visit rate of each store.

Optionally, the user-store intersection feature is obtained by at least one of the following methods:

determining the linear distance between the user terminal and each store according to the latitude and longitude of the user terminal and the latitude and longitude of each store;

The signal strength of the WIFI of the shop that the user terminal scans or connects to, performs feature crossover with the WIFI average connection or scan strength of the shop that the user terminal scans or connects to;

performing feature intersection on the WIFI signal strength of the store scanned or connected to by the user terminal, and the linear distance between the user terminal and the store scanned or connected to by the user terminal; and/or

The feature intersection is performed on the user click or consumption price of the user terminal and the per capita price of each store.

The third aspect of the embodiments of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in the method described in the first aspect of the present application are implemented.

The fourth aspect of the embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and operable on the processor, and the processor implements the method described in the first aspect of the present application when executed A step of.

Using a store arrival prediction method provided by the embodiment of the present application, firstly obtain the search request initiated by the user, then perform feature extraction on the search request, and extract the features (including: user terminal user features, search request corresponding The characteristics of each store in the store list and the user-store intersection feature) are input into the store probability prediction model to obtain the probability of the user arriving at each store in the store list, and finally the target store that the user arrives is predicted according to these probabilities. This application extracts the features of the search request initiated by the user terminal from the three dimensions of user dimension, store dimension and user-store intersection dimension, which improves the accuracy of the results output by the store arrival probability prediction model and the accuracy of the final predicted target store. In addition, the target store that the user arrives at is determined through the probability of each store output by the store arrival probability prediction model, which realizes accurate positioning of the user's current location and improves the reliability of the stores near the current location recommended by the user terminal to the user. performance, enhancing the user experience.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application , for those skilled in the art, other drawings can also be obtained according to these drawings without paying creative labor.

FIG. 1 is a schematic diagram of interaction between a user terminal and a background server provided by various embodiments of the present application;

Fig. 2 is a flowchart of a store arrival prediction method shown in an embodiment of the present application;

Fig. 3 is a flow chart of obtaining user-store intersection features shown in an embodiment of the present application;

Fig. 4 is a flowchart of a method for training a store arrival probability prediction model shown in an embodiment of the present application;

Fig. 5 is a flowchart of another method for training a store arrival probability prediction model shown in an embodiment of the present application;

FIG. 6 is a flow chart of a method for updating a store arrival probability prediction model shown in an embodiment of the present application;

Fig. 7 is a schematic diagram of a store arrival prediction device according to an embodiment of the present application.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of this application.

Before describing the various embodiments of the present application, the related technologies are described first. In order to obtain a more accurate positioning of the user, in related technologies, a method of correcting the obtained user's positioning is adopted, for example, one of the correction methods is: searching for a reference source within a certain range of the user's current location, for example: If the user's current network status is connected to WIFI, the coordinates of the WIFI can be used as a reference source, and then the user's positioning can be corrected; another example: when the user is inside a large building (for example: inside a large shopping mall), Since the GPS signal is blocked, the user's location can be corrected according to the surrounding geomagnetic information of the user's historical location.

However, the above two ways of correcting the obtained user's location need to obtain the corresponding data information, which is expensive. For example, when using the coordinates of WIFI as the reference source, it is necessary to additionally obtain the coordinates of each WIFI in the area where the user is located. Coordinate information. When using geomagnetic information to correct the user's positioning, it is necessary to additionally obtain geomagnetic information in the area where the user is located.

In related technologies, another way to obtain the location of the user is to collect the WIFI of the store the user is connected to, the time the user stays in the store, and the purchase information of the user in the store, and then formulate corresponding rules based on the business. Then determine whether the user is in the store according to the rules. However, this method only considers the WIFI of the store connected by the user, the user's stay time in the store, and the user's purchase information in the store, that is, it only judges whether the user is in the store from the user's side. In addition, the solution relies too much on Whether there is WIFI in the store or whether the user has a purchase behavior does not have strong universality. Therefore, the accuracy of the positioning result obtained by this positioning method for user positioning is still not high.

In order to improve the accuracy of the user's positioning, the embodiment of the present application utilizes the characteristic that many businesses in the O2O (Online-to-Offline) local life service can only be completed in the store, obtains the user's search request in real time and marks the user's target store. Consumption data of typical behaviors (for example: participating in store group buying coupon verification activities, purchasing flash discounts, signing in, uploading UGC, etc.), and then analyzing these consumption data from multiple dimensions such as users, stores, and the intersection of users and stores Feature extraction, and then input the extracted features into the pre-trained store arrival probability prediction model to obtain the probability of the user arriving at each store, and finally obtain the target store that the user actually arrives based on these probabilities.

A method for predicting a store arrival provided by an embodiment of the present application will be described in detail below.

FIG. 1 is a schematic diagram of interaction between a user terminal and a background server provided by various embodiments of the present application. Referring to FIG. 1 , the background server communicates with one or more user terminals (eg, user terminal 1 to user terminal n in FIG. 1 ) through a network to realize communication interaction. The background server may be a web server, a database server, or the like. The user terminal may be a personal computer (personal computer, PC), a tablet computer, a smart phone, and the like.

Fig. 2 is a flow chart of a store arrival prediction method according to an embodiment of the present application, and the method is applied to the background server in Fig. 1 . Referring to Fig. 2, the arrival prediction method provided by an embodiment of the present application includes the following steps:

Step S11: Obtain the search request initiated by the user terminal.

In this embodiment, the user may input a search request on the user terminal, and the background server is configured to receive the search request sent by the user terminal. Wherein, the user terminal is installed with terminal application software that can support a nearby search function, such as: search application software, shopping application software, or application software that can provide users with other services. Through this type of application software, users can search for any target within the surrounding range of the current location. After receiving the search request input by the user, the user terminal sends the search request to the background server, and then receives the search result returned by the background server for the search request and displays the search result to the user in the form of a store list on the page.

After the user terminal receives the user's search request, it actively sends the user data for this search request to the background server, or, after the user terminal receives the user's search request, the next time the background server requests the user data from the user terminal, then The user data for this search request is sent to the background server. Of course, other methods may also be used to send the user data for the search request to the background server, which is not specifically limited by various embodiments of the present application.

Wherein, the user data includes user behavior data, user information, and device state information of the user terminal. User behavior data refers to: data dynamically generated when a user performs user behavior on any store in the store list. User behavior can be general user behavior, such as clicking, collecting, sharing, etc.; it can also be behavior related to visiting a store (i.e. behaviors that can only be completed after arriving at the store), such as: paying for flash benefits, checking coupons for group purchases, self-service ordering, queuing for number pickup, uploading UGC, user sign-in, etc. Of course, other types of user behaviors can also be used. Applications include but are not limited to various user behaviors listed above. User information refers to: user profile (for example: user's age, occupation, gender, etc.), user's consumption preferences (for example: frequently purchased commodity categories, price consumption range), and other information that characterizes the user's personal characteristics. The device status information of the user terminal refers to: the name of the WIFI scanned or connected to by the user terminal and the corresponding signal strength, the device type of the user terminal, the latitude and longitude of the user terminal, the IP address of the user terminal, etc.

Exemplarily, the background server is a background server of a consumer APP. Correspondingly, the terminal application software installed on the user terminal that supports the nearby search function is a consumer APP that communicates and interacts with the background server. The user enters "hot pot" in the search bar of the consumer APP, and the consumer APP will pop up a page showing multiple stores related to "hot pot". The user can click on a store at will to query the relevant information of the store , Participate in activities such as flash discount purchase, group purchase coupon inspection, user sign-in and other activities in the store. The life consumption APP records the user behavior data generated by the user in response to the search request in real time, and at the same time obtains user information through the user's personal account, and obtains the device status information of the user terminal through the application program in the user terminal. information and the device status information of the user terminal to generate user data for this search request and send it to the background server.

Step S12: Perform feature extraction on the search request, determine the characteristics of the user of the user terminal, the characteristics of each store in the store list corresponding to the search request, and the user-store intersection feature, the user-store intersection feature is It is obtained by performing feature intersection on the features of the user and the features of each store.

In each embodiment of the present application, feature extraction is performed on the search request from three dimensions, namely: the user dimension, the store dimension, and the intersection dimension of the user and the store. Performing feature extraction on the search request is specifically: performing feature extraction on user data corresponding to the search request.

Specifically, determining the characteristics of the user of the user terminal refers to: extracting data that conforms to the user dimension from the user data, as the characteristics of the user of the user terminal, for example, the characteristics of the user of the user terminal may be: the user terminal scanned or connected to WIFI name and corresponding signal strength, device type of the user terminal, latitude and longitude of the user terminal, IP address of the user terminal, user portrait of the user, and consumption preference of the user. Of course, the user's features may also include other features that can characterize the user's personality, and this application includes but not limited to the user's features listed above.

The background server pre-stores information of all stores registered on the terminal application software. For example, when the terminal application software is a consumer APP, the background server pre-stores information of all stores registered on the consumer APP. After the user terminal initiates a search request, the background server will return a list of stores that meet the search criteria to the user terminal. Therefore, determining the characteristics of each store in the store list corresponding to the search request refers to: from the pre-stored information of all registered stores Extract the characteristics of each store in the store list, wherein, the characteristics of each store in the store list can be: the logo of each store, the WIFI name of each store, the average connection or scanning intensity of WIFI of each store, the latitude and longitude of each store, the The category it belongs to, the price range of the products sold in each store, the click-through rate of each store, and the visit rate of each store. Specifically, the category to which each store belongs represents the type of product sold in each store. For example, if the product sold in a certain store is clothing, then the category to which the store belongs is clothing, and for example: the product sold in a certain store If it is snacks or beverages, then the category of the store is catering. The visit-to-purchase rate of each store refers to the percentage of customers who made purchases among all customers who visited the store within a certain period of time.

The user-store intersection feature is the association feature between the user and each store. Fig. 3 is a flow chart of obtaining user-store intersection features according to an embodiment of the present application. Referring to Figure 3, the user-store cross feature can be obtained through the following steps:

Step S121: According to the latitude and longitude of the user terminal and the latitude and longitude of each store, determine the linear distance between the user terminal and each store.

Step S122: performing feature intersection on the WIFI signal strength of the store scanned or connected to by the user terminal and the average WIFI connection or scan strength of the store scanned or connected to by the user terminal.

The feature crossing of feature A and feature B refers to: using a preset calculation method to calculate feature A and feature B to obtain the associated feature C of feature A and feature B, for example: feature A is scanned by the user terminal or connected to The signal strength of the store's WIFI, feature B is the average connection or scan strength of the store's WIFI scanned or connected to the user terminal, the preset calculation method is the ratio of feature A to feature B, and the associated feature C is feature A divided by The resulting value for feature B.

Step S123: performing feature intersection on the WIFI signal strength of the store scanned or connected to by the user terminal, and the linear distance between the user terminal and the store scanned or connected to by the user terminal; and/or the user terminal The user's click or consumption price of the terminal and the per capita price of each store are subjected to feature intersection.

The per capita price of each store is the per capita consumption price of each store, the user click price is the price of the product clicked by the user, and the user consumption price is the price of the product purchased by the user. In this embodiment, in order to ensure the accuracy of the final predicted target store, the signal strength of the WIFI of the store that the user terminal scans or connects to, performs feature crossover with the WIFI average connection or scan strength of the store that the user terminal scans or connects to. Afterwards, feature intersection can also be performed in any of the following three ways:

1) The signal strength of the WIFI of the store scanned or connected to by the user terminal is characteristically intersected with the straight-line distance between the user terminal and the store scanned or connected to by the user terminal.

2) The user click price (or user consumption price) of the user terminal and the per capita price of each store are subjected to feature intersection.

3) The signal strength of the WIFI of the store scanned or connected to the user terminal is intersected with the linear distance between the user terminal and the store scanned or connected to the user terminal, and the user click price (or user consumption price) of the user terminal is calculated Perform feature intersection with the per capita price of each store.

Step S13: Input the user's features, the features of each store and the user-store intersection feature into a pre-trained store arrival probability prediction model to determine the probability of the user arriving at each store.

Wherein, the store arrival probability prediction model is obtained by using the search log of the user terminal to train the preset model. The specific training process will be described in detail below.

After the user's characteristics, the characteristics of each store, and the user-store cross-feature are input into the store probability prediction model as input values, the store arrival probability prediction model will output the probability value of the user's arrival at each store in the store list. The probability value of the user arriving at each store is proportional to the possibility of the user arriving at the store. The higher the probability value, the greater the possibility of the user arriving at the store, and the lower the probability value, the less likely the user is arriving at the store.

Step S14: According to the probability that the user arrives at each store, determine whether the user has arrived at a target store, and the target store is one of the various stores.

The target store is the store where the user is currently located, and determining the target store means determining the current precise location of the user. The specific steps for determining whether the user has reached the target store will be described below.

In one embodiment, the background server is the background server of a consumer APP, and the software installed on the user terminal that supports the nearby search function is a consumer APP that communicates and interacts with the background server. When the user is in a large shopping mall And when searching for nearby stores through the consumer APP, the background server will return a number of items that meet the search conditions according to the user's search request (the search conditions can be set from multiple aspects, such as: the distance from the user's current location, the price range of the store, User's favorable rating ranking) stores for users to view. For example, a user can enter "hotpot" in the search bar and set the search condition to be within a kilometer from the current location. The consumer APP will display all shops related to "hotpot" that meet the search condition on the page. Users can View the stores you are interested in or make user behaviors related to a certain store (as mentioned above, user behaviors can include: general user behaviors and behaviors related to store arrivals), and consumer apps record the user-generated information in real time. User behavior data, and user behavior data, user information, and device status information of the user terminal are sent to the background server as user data for the user's current search request (search request with the search name "hot pot"). The background server extracts features from the user data, and inputs the extracted features into the store arrival probability prediction model to obtain the probability of the user arriving at each store related to "hotpot" in the search results, and then predicts the user's current status based on each probability. store at all times. In each embodiment of the present application, the current time is a short duration including the time when the user initiates the search request.

In the embodiment of the present application, the search request initiated by the user is firstly obtained, and then the search request is subjected to feature extraction, and the extracted features (including: the user's feature of the user terminal, the feature of each store in the store list corresponding to the search request) and user-store intersection features) into the store probability prediction model to obtain the probability of the user arriving at each store in the store list, and finally predict the target store that the user arrives based on these probabilities. This application extracts the features of the search request initiated by the user terminal from the three dimensions of user dimension, store dimension and user-store intersection dimension, which improves the accuracy of the results output by the store arrival probability prediction model and the accuracy of the final predicted target store. In addition, the target store that the user arrives at is determined through the probability of each store output by the store arrival probability prediction model, which realizes the positioning of the user's current location and improves the reliability of the stores near the current location recommended by the user terminal to the user. , enhancing the user experience.

Specifically, step S14 may include:

Step S141: If the probability of the user arriving at the target store is greater than a preset probability threshold, determine that the user has arrived at the target store; or

Step S142: When the probability of the user arriving at the target store is greater than the preset probability threshold and the parameter value corresponding to the search request is within the preset effective parameter value range, determine that the user has reached the target store. target store.

In this embodiment, there are two ways to determine whether the user has arrived at the target store. The first way to determine is: compare the probability output by the arrival probability prediction model with the preset probability threshold, and determine whether the probability is greater than the preset probability threshold The store is the target store that the user arrives at; the second judgment method is: among the stores whose probability is greater than the preset probability threshold, the store whose parameter value corresponding to the search request is within the preset effective parameter value range is the target of the user's arrival Store, the relationship between the parameter value corresponding to the search request and the preset effective parameter value range is used to assist in determining whether the user has reached the target store.

For example, if the preset probability threshold is 0.8, and the probability of only one store M among all the probabilities output by the store arrival probability prediction model is 0.9, then according to the first judgment method, the store M is the target store that the user arrives at; There are two judgment methods. If the store M is a catering store, the parameter value corresponding to the search request can be set to the current time of the user terminal, and the preset effective parameter value range can be set to lunch time (for example: 11:00- 13:00), if the current time of the user terminal just falls within the preset effective parameter value range, it can be determined that the user arrives at the store M. Certainly, the parameter value corresponding to the search request and the preset effective parameter value range may be determined according to the requirements in the actual application process of the present application.

In actual situations, if the period for the background server to obtain the search request initiated by the user terminal is long, there may be a phenomenon that the probability values of multiple stores output by the store arrival probability prediction model are greater than the preset threshold. If the period for the background server to obtain the search request initiated by the user terminal is relatively short, the number of shops that satisfy the probability value greater than the preset threshold is usually small, or even zero. In this case, if there is only one probability value greater than the preset threshold If there is a store with a threshold value, the store is taken as the target store where the user is currently located. If there are multiple stores with a probability value greater than the preset threshold value, the store with the highest probability value among the multiple stores is taken as the store where the user is currently located. The preset threshold is an empirical threshold obtained by calculating the accuracy rate and recall rate of the store arrival probability prediction model to determine whether the user's arrival at the store is effective, that is, the preset threshold is the target store predicted by the store arrival probability prediction model, After comparative analysis with the store actually arrived by the user, combined with the optimal probability value obtained by the specific business, only under the optimal probability value, the accuracy of the target store predicted by the user to arrive is the highest.

In the embodiment of the present application, two methods are set up to determine whether the user has arrived at the target store, which enhances the flexibility of the store arrival prediction method of the present application in the actual application process. In addition, the second determination method adds a search The auxiliary determination condition of whether the parameter value corresponding to the request is within the preset effective parameter value range improves the accuracy of the determination result.

The training process of the store arrival probability prediction model will be described below.

Fig. 4 is a flow chart of a method for training a store arrival probability prediction model according to an embodiment of the present application. With reference to Fig. 4, this training method comprises:

Step S21: Determine the user behavior type associated with the user's visit to the store.

The user behavior associated with the user's visit to the store refers to the user's behavior that can only be completed after the user arrives at the store, such as: paying for flash benefits, checking coupons for group buying, self-service ordering, queuing for number pickup, uploading UGC, user sign-in, connecting to WIFI, etc. . For a store, if the user generates a user behavior related to the store, it is more likely that the user actually arrives at the store.

Step S22: Extracting a first type of search record and a second type of search record from the search log of the user terminal, the first type of search record is a search record conforming to the user behavior type, and the second type of search record It is a search record whose time difference between the corresponding search time and the search time of the first type of search record is within a preset duration.

The search log of the user terminal contains all search records, and a search request initiated by the user may correspond to multiple search records, and the multiple search records generated each time a search request is initiated constitute the user behavior data of the search request. For example, if the user enters "clothing" in the search bar, then the search request named "clothing" is the search request X. After clicking Search, the user terminal will display multiple search names that match "Chinese food". If user 1 views store A and completes check-in in store A, then user 1-search request X-store A-check-in is a search record, if user 1 views store A and participates in group buying in store A Coupon verification activity, then user 1-search request X-store A-group purchase coupon verification is another search record, similarly, user 1 can also generate multiple search records based on other stores.

Among them, the first type of search record is a search record that conforms to the type of user behavior. As long as the user behavior carried in the search record is a user behavior that can only be generated after the user arrives at the corresponding store, this search record can be used as the first type of search record. Record. For example: a search record is user 1-search request X-store A-group purchase coupon verification, since the group purchase coupon verification can only be completed after the user arrives at the corresponding store, so this search record can be used as a first-type search record, Another example: a search record is user 1-search request X-store A-share, since sharing does not require the user to reach the corresponding store, this search record cannot be regarded as a first-type search record.

Optionally, after multiple first-type search records are selected, filter conditions can be further set, for example, the filter condition can be a time range condition, and the corresponding search time in all search records is in the first category within a certain period of time Search records are filtered out as new first-class search records.

In this embodiment, the first type of search records is only a part of the search records in the search log that conform to the user behavior type. In order to make the collected search records reasonably distributed, the second type of search records need to be collected. The second type of search record is a search record in which the time difference between the corresponding search time and the search time of the first type of search record is within a preset duration. For example, the search time of a first-type search record is 10:00, if the preset duration is 1 minute, then the second-type search record obtained from the first-type search record can be for the user at 9:59- Search records generated between 10:01.

The method of extracting the first type of search records and the second type of search records adopted in this embodiment ensures that the extracted search records have a better distribution (for example, the first type of search records with a reasonable time distribution are extracted in advance, Then extract the second type of search records based on these first type of search records), avoiding the phenomenon that a certain number of search records that meet the user behavior type cannot be extracted, or the time distribution of the extracted search records is unreasonable (for example: search The phenomenon that records are excessively concentrated in a certain period of time).

Step S23: mark the search records matching the user behavior type among the first type search records and the second type search records as positive samples, and mark the second type search records not matching the user behavior type as positive samples. Behavior-type search records are marked as negative samples.

A positive sample indicates that the user has theoretically arrived at the store, and a negative sample indicates that the user has not theoretically arrived at the store. Since the user behaviors in all the first-type search records are user behaviors associated with the user's visit to the store, all the first-type search records are positive samples. Since the first type of search record is only a part of the search records in the search log that match the user behavior type, there may also be multiple search records that fit the user behavior type in the second type of search record, so when dividing For negative samples, it is also necessary to mark the search records matching the user behavior type in the second type of search records as positive samples, and mark the search records not matching the user behavior type as negative samples.

Step S24: According to the positive samples and the negative samples, train the preset model to obtain the prediction model of the arrival probability.

In this embodiment, the divided positive samples and negative samples are used as input values, and a machine learning algorithm (for example: logistic regression algorithm) is used to train the preset model to obtain a two-category prediction model (ie: arrival probability prediction model), Its function is: when the features extracted from the search request are input, the probability of the user arriving at each store in the store list can be output, and the store list is the search result corresponding to the search request.

Fig. 5 is a flow chart of another method for training a store arrival probability prediction model according to an embodiment of the present application. With reference to Fig. 5, step S24 comprises:

Step S241: Perform feature extraction on the positive sample and the negative sample respectively, determine the characteristics of the sample users corresponding to the positive sample and the negative sample, and the store list corresponding to the positive sample and the negative sample The features of each sample store and the sample user-sample store cross feature, the sample user-sample store cross feature is obtained by performing feature crossing on the sample user features and the features of each sample store.

Step S242: Using the characteristics of the sample user, the characteristics of each sample store, and the cross feature of the sample user-sample store as training samples, train the preset model to obtain the store arrival probability prediction model.

In this embodiment, each search record also carries user information and device status information of the user terminal. After obtaining positive samples and negative samples, the characteristics of the corresponding sample user and the corresponding sample user can be extracted from each sample. Store features and sample user-sample store cross features, and then input these features into the preset model and train the preset model to obtain a store arrival probability prediction model.

Fig. 6 is a flowchart of a method for updating a store arrival probability prediction model according to an embodiment of the present application. Referring to 6, the method includes:

Step S31: extracting from the search log of the user terminal the corresponding search records after the user arrives at the target store at a corresponding search time.

After the target store that the user arrives is predicted according to the store arrival probability prediction model, the corresponding search record is extracted from the search log of the user terminal after the user arrives at the target store, and this part of the search record is used as a feedback record. The record can realize the adjustment of the sample weight, which is convenient for updating the store arrival probability prediction model. For example: the user actually only arrived at store A, but may have made user behaviors related to the user's visit to other stores. At this time, the samples related to store A can be marked as positive samples and the weight value can be adjusted. The samples of other stores are marked as positive and negative and the weight value is adjusted to ensure the reliability of each sample when used.

Step S32: If the extracted search record is a search record for the target store, mark the extracted search record as a positive sample, and increase the weight of the positive sample.

For example, if it is predicted that the target store that the user arrives at is store A, then in the feedback record, the search record related to store A can be used as a positive sample with a larger weight value, for example: when the search record related to store A When the user behavior in the record is group buying and coupon checking behavior (or other user behavior, for example: uploading UGC, connecting to WIFI), the search record can be marked as a positive sample with a larger weight value.

Optionally, according to the different user behaviors in the feedback records, this embodiment sets different weight values for different positive samples (search records related to the target store in the feedback records), for example: in the positive samples, the user behavior is The weight value corresponding to the sample of the behavior of group buying coupon checking is greater than the weight value corresponding to the sample of the behavior of the user clicking.

Step S33: if the extracted search record is not a search record for the target store, mark the extracted search record as a negative sample, and reduce the weight of the negative sample.

In this embodiment, the weights of search records irrelevant to store A can also be adjusted, specifically, the search records unrelated to store A are marked as negative samples, and the weight of the negative samples is reduced. For example: when the user behavior in the extracted search record is not group checking coupon behavior (or other user behavior, for example: uploading UGC, connecting to WIFI), the search record can be marked as a negative sample with a smaller weight value.

Optionally, the weight values of different negative samples can also be adjusted. According to the different user behaviors in the feedback records, this embodiment sets the size of different negative samples (search records that have nothing to do with the target store in the feedback records) different weight values. For example, in the negative samples, the weight value corresponding to the sample whose user behavior is group buying coupon inspection is greater than the weight value corresponding to the sample whose user behavior is click behavior.

Step S34: update the store arrival probability prediction model according to the positive samples after the weight increase and the negative samples after the weight reduction.

In this embodiment, all the feedback records are used as samples when the store arrival probability prediction model is updated later, and the feedback records are mined from the search logs and reacted to the prediction process of the model. According to the determined user's arrival goal The store and the feedback behavior generated by the user after the target store is determined, the weight value of the positive sample and the negative sample in the feedback record is adjusted to ensure the reliability of the sample used when updating the store probability prediction model. In the case of a large number, the continuous iterative optimization of the arrival probability prediction model has been realized, which has effectively improved the accuracy of the prediction results of the arrival probability prediction model, and the target predicted based on the prediction results of the arrival probability prediction model store accuracy.

The store arrival prediction method in the embodiment of the present application can locate users more accurately, improve the accuracy of distance calculation in nearby O2O searches, and is applicable to various scenarios that need to calculate the distance from a store to a user, such as various services The peripheral search service of the software (including: nearby food, fresh fruits and vegetables, flower delivery, etc.), can provide users with a better search experience. In addition, the store arrival prediction method proposed in this application is not only simple and easy to implement and low in cost, but also can use feedback records to continuously iteratively update the store arrival prediction model, so that the accuracy of the output results of the store arrival prediction model can increase with the search business With the development of scale and the accumulation of search data (for example: search records), it is getting higher and higher, so as to achieve more precise positioning of users.

Based on the same inventive concept, an embodiment of the present application provides a store arrival prediction device. Fig. 7 is a schematic diagram of a store arrival prediction device according to an embodiment of the present application. Referring to Figure 7, the device 700 includes:

Obtaining module 701, configured to obtain a search request initiated by a user terminal;

The feature extraction module 702 is configured to perform feature extraction on the search request, determine the features of the user of the user terminal, the features of each store in the store list corresponding to the search request, and the user-store intersection feature, the user- The store intersection feature is obtained by performing feature intersection on the features of the user and the features of each store;

The probability prediction module 703 is used to input the characteristics of the user, the characteristics of each store, and the user-store cross-feature into a pre-trained store arrival probability prediction model, and determine the probability that the user arrives at each store;

The determination module 704 is configured to determine whether the user arrives at a target store according to the probability that the user arrives at each store, and the target store is one of the various stores.

Optionally, the determination module includes:

A first determining module, configured to determine that the user arrives at the target store when the probability of the user arriving at the target store is greater than a preset probability threshold; or

The second determination module is configured to determine the target store when the probability of the user arriving at the target store is greater than a preset probability threshold and the parameter value corresponding to the search request is within the preset effective parameter value range. The user arrives at the target store.

Optionally, the device also includes:

The third determination module is used to determine the type of user behavior associated with the user's visit to the store;

A first extracting module, configured to extract a first type of search record and a second type of search record from the search log of the user terminal, the first type of search record is a search record conforming to the user behavior type, the second type of search record The second type of search record is a search record in which the time difference between the corresponding search time and the search time of the first type of search record is within a preset duration;

A marking module, configured to mark the search records of the first type of search record and the second type of search records that match the user behavior type as positive samples, and mark the search records of the second type of search records that do not meet the user behavior type. The search records of the above user behavior types are marked as negative samples;

The training module is used to train a preset model according to the positive samples and the negative samples to obtain the prediction model of the arrival probability.

Optionally, the training module includes:

The feature extraction sub-module is used to perform feature extraction on the positive samples and the negative samples respectively, determine the characteristics of the sample users corresponding to the positive samples and the negative samples, and determine the characteristics of the sample users corresponding to the positive samples and the negative samples respectively. The features of each sample store in the corresponding store list and the sample user-sample store cross feature, the sample user-sample store cross feature is obtained by feature crossing the features of the sample user and the features of each sample store;

The training sub-module is used to use the characteristics of the sample user, the characteristics of each sample store, and the cross feature of the sample user-sample store as training samples to train the preset model to obtain the store arrival probability predictive model.

Optionally, the device also includes:

The second extraction module is used to extract from the search log of the user terminal the corresponding search record after the user arrives at the target store at the corresponding search time;

The first weight adjustment module is used to mark the extracted search record as a positive sample and increase the weight of the positive sample when the extracted search record is a search record for the target store;

A second weight adjustment module, configured to mark the extracted search record as a negative sample and reduce the weight of the negative sample when the extracted search record is not a search record for the target store; and

An update module, configured to update the store arrival probability prediction model according to the weighted positive samples and the weighted negative samples.

Optionally, the characteristics of the user include at least one of the following: the name of the WIFI scanned or connected to by the user terminal and the corresponding signal strength, the device type of the user terminal, the latitude and longitude of the user terminal, the user terminal The IP address of the terminal, the user portrait of the user, and the consumption preference of the user.

Optionally, the characteristics of each store include at least one of the following: the logo of each store, the WIFI name of each store, the average connection or scanning intensity of WIFI of each store, the latitude and longitude of each store, The category to which each store belongs, the price range of commodities sold by each store, the click rate of each store, and the visit rate of each store.

Optionally, the user-store intersection feature is obtained by at least one of the following methods:

determining the linear distance between the user terminal and each store according to the latitude and longitude of the user terminal and the latitude and longitude of each store;

The signal strength of the WIFI of the shop that the user terminal scans or connects to, performs feature crossover with the WIFI average connection or scan strength of the shop that the user terminal scans or connects to;

performing feature intersection on the WIFI signal strength of the store scanned or connected to by the user terminal, and the linear distance between the user terminal and the store scanned or connected to by the user terminal; and/or

The feature intersection is performed on the user click or consumption price of the user terminal and the per capita price of each store.

Based on the same inventive concept, another embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in the method described in any of the above-mentioned embodiments of the present application are implemented. .

Based on the same inventive concept, another embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and operable on the processor. Steps in the methods described in the examples.

As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

Those skilled in the art should understand that the embodiments of the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the embodiment of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present application are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor or processor of other programmable data processing terminal equipment to produce a machine such that instructions executed by the computer or processor of other programmable data processing terminal equipment Produce means for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the The instruction means implements the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded into a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce computer-implemented processing, thereby The instructions executed above provide steps for implementing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

While the preferred embodiments of the embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once the basic inventive concept is understood. Therefore, the appended claims are intended to be interpreted to cover the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the application.

Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or terminal equipment comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements identified, or also include elements inherent in such a process, method, article, or end-equipment. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

The above is a detailed introduction to the store arrival prediction method, device, readable storage medium and electronic equipment provided by the application. In this paper, specific examples are used to illustrate the principle and implementation of the application. The above embodiments The description is only used to help understand the method of the present application and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation and application scope, in summary , the contents of this specification should not be construed as limiting the application.

Claims (11)

1. A store arrival prediction method, characterized in that the method comprises: obtaining a search request initiated by a user terminal; Perform feature extraction on the search request, determine the user characteristics of the user terminal, the characteristics of each store in the store list corresponding to the search request, and the user-store intersection feature, the user-store intersection feature is the Obtained by feature crossing between the user's features and the features of each store; Inputting the characteristics of the user, the characteristics of each store and the user-store cross-feature into a pre-trained store arrival probability prediction model, and determining the probability that the user arrives at each store; According to the probability that the user arrives at each store, it is determined whether the user arrives at a target store, and the target store is one of the various stores. 2. The method according to claim 1, wherein the step of determining whether the user arrives at the target store according to the probability of the user arriving at each store comprises: If the probability that the user arrives at the target store is greater than a preset probability threshold, determining that the user arrives at the target store; or When the probability that the user arrives at the target store is greater than a preset probability threshold and the parameter value corresponding to the search request is within a preset effective parameter value range, it is determined that the user arrives at the target store. 3. The method according to claim 1, characterized in that the method further comprises: Determine the type of user behavior associated with a user visit to the store; Extract a first type of search record and a second type of search record from the search log of the user terminal, the first type of search record is a search record conforming to the user behavior type, and the second type of search record is a corresponding Search records whose time difference between the search time and the search time of the first type of search records is within a preset duration; Mark the search records of the first type of search record and the second type of search records that conform to the user behavior type as positive samples, and mark the search records of the second type of search records that do not conform to the user behavior type Search records are marked as negative samples; According to the positive sample and the negative sample, the preset model is trained to obtain the store arrival probability prediction model. 4. The method according to claim 3, wherein the step of training a preset model according to the positive samples and the negative samples to obtain the predictive model of the arrival probability comprises: Perform feature extraction on the positive sample and the negative sample respectively, determine the characteristics of the sample users corresponding to the positive sample and the negative sample, and each sample in the store list corresponding to the positive sample and the negative sample The characteristics of the store and the sample user-sample store intersection feature, the sample user-sample store intersection feature is obtained by performing feature intersection on the characteristics of the sample user and the characteristics of each sample store; Using the features of the sample user, the features of each sample store, and the cross feature of the sample user-sample store as training samples, the preset model is trained to obtain the store arrival probability prediction model. 5. The method according to claim 2, characterized in that, after the step of determining that the user arrives at the target store, the method further comprises: Extracting from the search log of the user terminal the corresponding search record after the user arrives at the target store at the corresponding search time; In the case that the extracted search record is a search record for the target store, the extracted search record is marked as a positive sample, and the weight of the positive sample is increased; In the case that the extracted search record is not a search record for the target store, marking the extracted search record as a negative sample and reducing the weight of the negative sample; and The store arrival probability prediction model is updated according to the positive samples after the weight increase and the negative samples after the weight reduction. 6. The method according to claim 1, wherein the characteristics of the user include at least one of the following: the name of the WIFI scanned or connected to by the user terminal and the corresponding signal strength, the device type of the user terminal , the latitude and longitude of the user terminal, the IP address of the user terminal, the user portrait of the user, and the consumption preference of the user. 7. The method according to claim 1, wherein the characteristics of each store include at least one of the following: the logo of each store, the WIFI name of each store, the average WIFI connection of each store Or scan intensity, the latitude and longitude of each store, the category to which each store belongs, the price range of the commodities sold by each store, the click rate of each store, and the visit rate of each store. 8. The method according to claim 7, wherein the user-store intersection feature is obtained in at least one of the following ways: determining the linear distance between the user terminal and each store according to the latitude and longitude of the user terminal and the latitude and longitude of each store; The signal strength of the WIFI of the shop that the user terminal scans or connects to, performs feature crossover with the WIFI average connection or scan strength of the shop that the user terminal scans or connects to; performing feature intersection on the WIFI signal strength of the store scanned or connected to by the user terminal, and the linear distance between the user terminal and the store scanned or connected to by the user terminal; and/or The feature intersection is performed on the user click or consumption price of the user terminal and the per capita price of each store. 9. A store arrival prediction device, characterized in that the device comprises: An obtaining module, configured to obtain a search request initiated by a user terminal; A feature extraction module, configured to perform feature extraction on the search request, determine the features of the user of the user terminal, the features of each store in the store list corresponding to the search request, and the user-store intersection feature, the user-store The intersection feature is obtained by performing feature intersection on the features of the user and the features of each store; The probability prediction module is used to input the characteristics of the user, the characteristics of each store and the user-store cross-feature into a pre-trained store arrival probability prediction model to determine the probability that the user arrives at each store; and A determining module, configured to determine whether the user arrives at a target store according to the probability that the user arrives at each store, and the target store is one of the various stores. 10. A computer-readable storage medium, on which a computer program is stored, wherein, when the program is executed by a processor, the steps in the method according to any one of claims 1-8 are implemented. 11. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, characterized in that, when the processor executes, it realizes any one of claims 1-8 method steps.