CN117788118A

CN117788118A - Product recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN117788118A
Application number: CN202311857642.9A
Authority: CN
Inventors: 王佳璐
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-12-29
Filing date: 2023-12-29
Publication date: 2024-03-29

Abstract

The application provides a product recommendation method, a product recommendation device, electronic equipment and a storage medium, and relates to the technical field of financial science and technology. The method comprises the following steps: according to a received recommendation request sent by a user, a user portrait set and a product portrait set are obtained, a first recommendation list is obtained according to identity information of a target user, the user portrait set and the product portrait set through a collaborative filtering recommendation algorithm based on a time attenuation function, a second recommendation list is obtained through a random walk pattern recommendation algorithm based on improvement, the identity information of the target user, the first recommendation list and the second recommendation list are input into a preset reordering recommendation model, a target score of each candidate product is generated, and at least one target product is pushed to the target user according to the target score of each candidate product. By the method, the accuracy and individuation degree of product recommendation are improved, and timeliness, flexibility and stability are also improved.

Description

Product recommendation method and device, electronic equipment and storage medium

Technical Field

The application relates to the technical field of financial science and technology, in particular to a product recommendation method, a device, electronic equipment and a storage medium.

Background

With the development of the financial market, banks are faced with a strong competition and diversified customer demands, and how to provide personalized recommendations for customers from massive financial products becomes an important problem. The recommendation system is used as an intelligent information filtering and matching technology, and aims to help customers find valuable financial products, improve customer experience and satisfaction, and create value and benefits for banks.

In the prior art, the recommendation of the bank financial products is usually realized through a mode of carrying out product linkage on line and off line together, or a new method of the old, however, the method is relatively passive, and can not accurately recommend proper products for target users, and meanwhile, timeliness and flexibility are lacking.

In summary, how to implement high-precision product recommendation for users is a problem to be solved in the art.

Disclosure of Invention

The application provides a product recommendation method, a product recommendation device, electronic equipment and a storage medium, which are used for solving the problem of how to realize high-precision product recommendation for a user.

In a first aspect, the present application provides a product recommendation method, applied to a server of a banking system, including:

Receiving a recommendation request sent by a user, wherein the recommendation request comprises identity information of a target user;

acquiring a user portrait set and a product portrait set according to the recommendation request, wherein the user portrait set comprises user portraits of all users in the banking system, and the product portrait set comprises product portraits of all products in the banking system;

according to the identity information of the target user, the user portrait set and the product portrait set, a first recommendation list is obtained through calculation through a collaborative filtering recommendation algorithm based on a time attenuation function, and the first recommendation list comprises at least one alternative product;

according to the identity information of the target user, the user portrait set and the product portrait set, calculating to obtain a second recommendation list through a recommendation algorithm based on an improved random walk pattern, wherein the second recommendation list comprises at least one alternative product;

inputting the identity information of the target user, the first recommendation list and the second recommendation list into a preset reordering recommendation model, and generating a target score of each alternative product, wherein the reordering recommendation model is a model for calculating product scores obtained by training an XGBoost model according to the user portrait set and the product portrait set, and the alternative product is any one of the first recommendation list and the second recommendation list;

And pushing at least one target product to the user according to the target score of each candidate product.

With reference to the first aspect, in some embodiments, the calculating, according to the identity information of the target user, the user portrait set, and the product portrait set, the first recommendation list through a collaborative filtering recommendation algorithm based on a time decay function includes:

acquiring product evaluation data of the banking system;

determining a target product set according to the identity information of the target user, the user portrait set, the product evaluation data and the product portrait set, wherein the target product set comprises: the target user and at least one other user jointly evaluate the product image of the first product, the score of each user on each first product and the scoring time stamp of each user on each first product;

according to the identity information of the target user and the target product set, calculating at least one user similarity through a Pearson correlation coefficient based on time attenuation, wherein the at least one user similarity comprises the similarity between the target user and the at least one other user;

Determining at least one neighbor user from the at least one other user according to a preset similarity threshold and the at least one user similarity, wherein the at least one neighbor user comprises a user with the user similarity greater than the similarity threshold with the target user;

according to the identity information of the target user, the target product set, the at least one user similarity and the at least one neighbor user, calculating and obtaining a prediction score of the target user on each first product through a weighted average algorithm based on time attenuation;

and sequencing each first product from large to small according to the predictive score, and determining a first recommendation list according to a preset scoring threshold.

With reference to the first aspect, in some embodiments, the calculating, according to the identity information of the target user, the user portrait set, and the product portrait set, the second recommendation list based on the improved random walk pattern recommendation algorithm includes:

carrying out graph construction processing according to the identity information of the target user, the user portrait set and the product portrait set to obtain an heterogram, wherein the heterogram comprises user nodes, product nodes and attribute nodes;

Starting from a user node of the target user, performing wandering on the heterogeneous graph according to a preset random walk rule to obtain a wandering result, wherein the wandering result comprises the access times of all nodes;

and determining the second recommendation list according to the identity information of the target user, the user portrait set, the product portrait set and the wandering result.

With reference to the first aspect, in some embodiments, the random walk rule includes:

when the current node is a user node, determining a next node needing to walk according to a preset first probability, wherein the first probability comprises a user node probability, a product node probability and an attribute node probability which are matched with the current node, and the sum of the user node probability, the product node probability and the attribute node probability is 1;

when the current node is a product node, determining a next node needing to walk according to a preset second probability, wherein the second probability comprises a user node probability, a product node probability and an attribute node probability which are matched with the current node, and the sum of the user node probability, the product node probability and the attribute node probability is 1;

When the current node is an attribute node, determining a next node needing to walk according to a preset third probability, wherein the third probability comprises a user node probability and a product node probability which are matched with the current node, and the sum of the user node probability and the product node probability is 1.

With reference to the first aspect, in some embodiments, the determining the second recommendation list according to the identity information of the target user, the user portrait set, the product portrait set, and the walk result includes:

according to the identity information of the target user, the user portrait set and the product portrait set, carrying out matching degree calculation on the target user and each product to obtain the matching degree of each product;

screening all products according to a preset matching degree threshold value of each product to obtain at least one screened second product, wherein the matching degree of each second product is larger than the matching degree threshold value;

and sequencing the at least one second product according to the access times from large to small, and determining the second recommendation list according to a preset access time threshold.

With reference to the first aspect, in some embodiments, the acquiring a user portrait set and a product portrait set according to the recommendation request includes:

reading a user data set and a product data set from a database, wherein the user data set comprises data of all users in the banking system, and the product data set comprises data of all products in the banking system;

respectively carrying out data cleaning on the user data set and the product data set to obtain a cleaned user data set and a cleaned product data set;

performing feature extraction on the cleaned user data set to obtain a user feature set, and performing feature extraction on the cleaned product data set to obtain a product feature set;

and labeling the user characteristics to obtain the user portrait set, and labeling the product characteristics to obtain the product portrait set.

With reference to the first aspect, in some embodiments, the pushing at least one target product to the user according to the target score of each candidate product includes:

determining at least one target product from a plurality of candidate products according to a preset scoring threshold, wherein the at least one target product is a candidate product with a target score greater than the scoring threshold;

And pushing a target product list to the user according to the at least one target product, wherein the target product list comprises the at least one target product which is ranked from large to small according to a target score.

In a second aspect, the present application provides a product recommendation device, comprising:

the receiving module is used for receiving a recommendation request sent by a user, wherein the recommendation request comprises the identity information of a target user;

the acquisition module is used for acquiring a user portrait set and a product portrait set according to the recommendation request, wherein the user portrait set comprises user portraits of all users in the banking system, and the product portrait set comprises product portraits of all products in the banking system;

the first calculation module is used for calculating a first recommendation list according to the identity information of the target user, the user portrait set and the product portrait set through a collaborative filtering recommendation algorithm based on a time attenuation function, wherein the first recommendation list comprises at least one alternative product;

the second calculation module is used for calculating a second recommendation list according to the identity information of the target user, the user portrait set and the product portrait set through a recommendation algorithm based on an improved random walk pattern, and the second recommendation list comprises at least one alternative product;

The third calculation module is used for inputting the identity information of the target user, the first recommendation list and the second recommendation list into a preset reordering recommendation model, and generating a target score of each alternative product, wherein the reordering recommendation model is a model for calculating product scores obtained by training an XGBoost model according to the user portrait set and the product portrait set, and the alternative product is any one of the first recommendation list and the second recommendation list;

and the pushing module is used for pushing at least one target product to the user according to the target score of each candidate product.

In a third aspect, the present application provides an electronic device, comprising: the device comprises a memory, a processor, a communication interface and a display;

the memory stores computer-executable instructions;

the processor executes computer-executable instructions stored in the memory to implement the method of any one of the first aspects.

In a fourth aspect, the present application provides a computer-readable storage medium having stored therein computer-executable instructions for implementing the method of any one of the first aspects when executed by a processor.

In a fifth aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the product recommendation method of any one of the first aspects.

According to the product recommending method, device, electronic equipment and storage medium, a user portrait set and a product portrait set are obtained according to a received recommending request sent by a user, a first recommending list is obtained according to identity information of a target user, the user portrait set and the product portrait set through a collaborative filtering recommending algorithm based on a time attenuation function, a second recommending list is obtained through a recommending algorithm based on an improved random walk pattern, identity information of the target user, the first recommending list and the second recommending list are input into a preset reordering recommending model, target scores of each candidate product are generated, and at least one target product is pushed to the target user according to the target scores of each candidate product. By the method, the accuracy and individuation degree of product recommendation are improved, and timeliness, flexibility and stability are also improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

Fig. 1 is an application scenario diagram of a product recommendation method provided in an embodiment of the present application;

fig. 2 is a schematic flow chart of a first embodiment of a product recommendation method provided in the embodiment of the present application;

fig. 3 is a schematic flow chart of a second embodiment of a product recommendation method provided in the embodiment of the present application;

fig. 4 is a schematic flow chart of a third embodiment of a product recommendation method provided in the embodiment of the present application;

fig. 5 is a schematic flow chart of a fourth embodiment of a product recommendation method provided in the embodiment of the present application;

fig. 6 is a schematic flow chart of a fifth embodiment of a product recommendation method provided in the embodiment of the present application;

fig. 7 is a schematic flow chart of a sixth embodiment of a product recommendation method provided in the embodiment of the present application;

fig. 8 is a schematic structural diagram of a first embodiment of a product recommendation device provided in an embodiment of the present application;

fig. 9 is a schematic structural diagram of a second embodiment of a product recommendation device provided in the embodiment of the present application;

fig. 10 is a schematic structural diagram of a third embodiment of a product recommendation device provided in the embodiment of the present application;

fig. 11 is a schematic structural diagram of a fourth embodiment of a product recommendation device provided in the embodiment of the present application;

Fig. 12 is a schematic structural diagram of a fifth embodiment of a product recommendation device provided in the embodiment of the present application;

fig. 13 is a schematic structural diagram of an electronic device provided in the present application.

Specific embodiments thereof have been shown by way of example in the drawings and will herein be described in more detail. These drawings and the written description are not intended to limit the scope of the inventive concepts in any way, but to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present application as detailed in the accompanying claims.

With the development of the financial market, banks are faced with a strong competition and diversified customer demands, and how to provide personalized recommendations for customers from massive financial products becomes an important problem. The recommendation system is used as an intelligent information filtering and matching technology, and aims to help customers find valuable financial products, improve customer experience and satisfaction, and create value and benefits for banks. For the recommendation of bank financial products, the product linkage is usually realized through a mode of carrying out product linkage on line and off line together, or a new method of the old, however, the method is relatively passive, and can not accurately recommend proper products for target users, and meanwhile, timeliness and flexibility are lacking.

Aiming at the problems, the application provides a product recommending method, a device, electronic equipment and a storage medium, which realize the accuracy, flexibility and timeliness of recommending products for users. Specifically, for bank products, the product linkage mode is usually realized through online and offline, or the recommendation is performed by a new method in the old zone, however, the method has low accuracy and lack of intellectualization, and cannot accurately recommend a proper product for a target user, and meanwhile, the inventor researches whether a collaborative filtering algorithm, a random walk pattern recommendation algorithm and a XGBoost model-based reordering recommendation algorithm can be integrated on the basis of user images and product images as data in consideration of the problems, so that the product recommendation for the target user is realized.

It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) referred to in the present application are information and data authorized by the user or fully authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards, and provide corresponding operation entries for the user to select authorization or rejection.

It should be noted that the product recommendation method, device, electronic equipment and storage medium of the present application may be used in the technical field of financial technology, and may also be used in any field other than the technical field of financial technology, and the application fields of the product recommendation method, device, electronic equipment and storage medium of the present application are not limited.

In the application scenario diagram of the product recommendation method provided in the embodiment of the application, as shown in fig. 1, the scenario at least includes a server, a financial platform and a user, where the server may be electronic devices such as a notebook computer and a desktop computer, the financial platform may be electronic devices such as a notebook computer and a smart phone with a data storage function, or may be a functional module with a storage function, the server may be connected with a front-end device, the front-end device may be electronic devices such as a notebook computer, a mobile phone and a smart watch, the user may perform data communication with the server through control operation of the front-end device, the scenario may further include a terminal, the terminal is configured with an application program, the application program may implement data transmission with the server through a communication link, and then the user may implement data communication with the server through operation of the terminal. Based on the requirements of users, the server can process data stored by the financial platform, and further, products are recommended to the users through front-end equipment or terminals.

The following describes the technical solutions of the present application and how the technical solutions of the present application solve the above technical problems in detail with specific embodiments. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.

Fig. 2 is a schematic flow chart of a first embodiment of a product recommendation method provided in the embodiment of the present application, as shown in fig. 2, where the product recommendation method provided in the embodiment is applied to a server of a banking system, and specifically includes:

s201: and receiving a recommendation request sent by the user.

In this step, in order to bring better consumption experience for the user and further realize more accurate product recommendation for the user, the server of the banking system responds to the user operation and receives a recommendation request sent by the user, wherein the recommendation request comprises the identity information of the target user.

Specifically, the server of the banking system responds to the user operation, which may be that the target user operates an application program configured by the terminal device, and then the application program generates a recommendation request according to the user operation, and then sends the recommendation request to the server of the banking system through a communication link, or that the target user goes to a banking hall, and then sends the recommendation request to the server of the banking system through control operation performed by a banking staff at a front-end display device.

Optionally, the application program operation configured by the user through the terminal device may be that the user opens the application program and logs in personal information, or that the user clicks the product recommendation control or interface through the application program. The operation of the bank staff can be that the staff inputs the identity information of the target user through the front-end equipment, or can be that the face recognition is carried out on the target user through a camera configured by the front-end equipment under the condition that the target user is allowed to be authorized.

S202: and acquiring a user portrait set and a product portrait set according to the recommendation request.

In the step, after a server of a banking system receives a recommendation request sent by a user, the server acquires a user portrait set and a product portrait set through a financial platform according to the recommendation request in order to improve the accuracy of recommending a target user product. The user image set comprises user images of all users in the bank system, and the product image set comprises product images of all products in the bank system.

Specifically, a user data set and a product data set are read from a database, data cleaning is respectively carried out on the user data set and the product data set to obtain a cleaned user data set and a cleaned product data set, then feature extraction is carried out on the cleaned user data set to obtain a user feature set, feature extraction is carried out on the cleaned product data set to obtain a product feature set, finally, labeling treatment is carried out on the user feature to obtain a user image set, and labeling treatment is carried out on the product feature to obtain a product image set.

It should be noted that, the financial platform is a data management platform, the historical data of the banking system are all stored in the financial platform, and the historical data of other devices are also stored in the financial platform, and under the authorization of other devices, the server of the banking system can also obtain the user portrait and the product portrait of other devices.

S203: and calculating to obtain a first recommendation list through a collaborative filtering recommendation algorithm based on a time decay function according to the identity information of the target user, the user image set and the product image set.

In this step, in order to recommend a product that is more matched with the target user to the target user, filtering is performed on the target user, the user portraits set and the product portraits set according to the identity information of the target user through a collaborative filtering recommendation algorithm based on a time decay function, so as to obtain a first recommendation list, wherein the first recommendation list comprises at least one candidate product.

Specifically, product evaluation data of a banking system are obtained from a database of the banking system, then a target product set is determined according to identity information of a target user, a user portrait set, product evaluation data and a product portrait set, at least one user similarity is obtained through calculation based on a pearson correlation coefficient of time attenuation according to the identity information of the target user and the target product set, at least one neighbor user is determined from at least one other user according to a preset similarity threshold and the at least one user similarity, a prediction score of the target user for each first product is obtained through calculation according to the identity information of the target user, the target product set, the at least one user similarity and the at least one neighbor user through a weighted average algorithm based on time attenuation, and finally each first product is ranked from large to small according to the prediction score, and a first recommendation list is determined according to the preset score threshold.

S204: and calculating a second recommendation list according to the identity information of the target user, the user image set and the product image set through a recommendation algorithm based on an improved random walk pattern.

In the step, after the first recommendation list is obtained, in order to improve the recommendation accuracy, identity information of a target user, a user image set and a product image set are walked according to a random walk rule through a recommendation algorithm based on an improved random walk pattern, so that a second recommendation list is obtained. Wherein the second recommendation list includes at least one alternative product.

Specifically, according to the identity information of the target user, the user image set and the product image set, performing image construction processing to obtain an abnormal image, then starting from a user node of the target user, performing wandering on the heterogeneous image according to a preset random wandering rule to obtain a wandering result, and finally determining a second recommendation list according to the identity information of the target user, the user image set, the product image set and the wandering result.

It should be noted that, the candidate product included in the first recommendation list and the candidate product included in the second recommendation list may have duplicate products or may be completely different products.

S205: and inputting the identity information of the target user, the first recommendation list and the second recommendation list into a preset reordered recommendation model, and generating a target score of each candidate product.

In this step, after the first recommendation list and the second recommendation list are obtained, a product more suitable for the target user can be selected from the candidate products in the two lists, and then the identity information of the target user, the first recommendation list and the second recommendation list are input into a preset reorder recommendation model, and the candidate products in the two recommendation lists are reordered through the reorder recommendation model, so that the target score of each candidate product is generated. The reorder recommendation model is a model for calculating product scores, which is obtained by training the XGBoost model according to a user image set and a product image set, and the candidate products are all candidate products in a first recommendation list and a second recommendation list.

S206: and pushing at least one target product to the target user according to the target score of each candidate product.

In this step, after obtaining the target score of each candidate product, in order to recommend a product more suitable for the target user to the target user, the target scores are sorted from large to small, so as to obtain at least one target product, and the at least one target product is pushed to the target user.

Specifically, according to a preset scoring threshold, at least one target product is determined from a plurality of candidate products, and then a target product list is pushed to a target user according to the at least one target product. The target product list comprises at least one target product which is ranked according to the target score from large to small.

Optionally, a number threshold may be preset, the target products are sorted according to the target scores from large to small, and a target product list is generated by selecting a preset number threshold.

According to the product recommendation method provided by the embodiment, a user portrait set and a product portrait set are obtained according to a received recommendation request sent by a user, a first recommendation list is obtained according to identity information of a target user, the user portrait set and the product portrait set through a collaborative filtering recommendation algorithm based on a time attenuation function, a second recommendation list is obtained through a recommendation algorithm based on an improved random walk pattern, identity information of the target user, the first recommendation list and the second recommendation list are input into a preset reordering recommendation model, a target score of each candidate product is generated, and at least one target product is pushed to the target user according to the target score of each candidate product. By the method, the accuracy and individuation degree of product recommendation are improved, and timeliness, flexibility and stability are also improved.

Fig. 3 is a schematic flow chart of a second embodiment of a product recommendation method provided in the embodiment of the present application, as shown in fig. 3, based on the foregoing embodiment, step S203 specifically includes:

s301: product evaluation data of the banking system is obtained.

In this step, in order to obtain the first recommendation list by calculation through the collaborative filtering algorithm, calculation is required based on the real evaluation data of the bank product by the user, and further the evaluation data of the product is obtained through the historical database of the bank system.

S302: and determining a target product set according to the identity information of the target user, the user portrait set, the product evaluation data and the product portrait set.

In this step, after obtaining the evaluation data of the products, in order to screen out the products suitable for the target users, the products jointly evaluated by other users and the target users need to be summarized to obtain a target product set, where the target product set includes: the product image of the first product that the target user has rated in common with at least one other user, the score of each user for each first product, and the time stamp of the score of each user for each first product.

Specifically, user figures of other users are screened according to the identity information of the target user and the user figures of the target user, further a first product is determined, product figures of the first product are screened out in a product figure set according to the determined first product, further product evaluation data of the first product are screened out in product evaluation data, and the product figures of the first product and the product evaluation data are summarized to obtain a target product set.

S303: and calculating at least one user similarity through the Person correlation coefficient based on time attenuation according to the identity information of the target user and the target product set.

In this step, after the target product set is determined, the similarity between other users and the target user is calculated by pearson correlation coefficient based on time attenuation.

Specifically, the user similarity is calculated by the following formula:

wherein I is _uv Is a target product set, u represents a target user, v represents other users, r _ui Representing the target user's score for product i,representing the average score, w, of the target user for the first product _ui A time decay weight representing the target user's score for product i.

The time decay weight of a user's score for a product can be calculated using the following formula:

wherein T is _max For the largest timestamp in the target product set, T _ui And a time stamp representing the product scoring of the target user, wherein lambda is a time attenuation factor, and the time attenuation factor can be adjusted according to the target product set.

S304: and determining at least one neighbor user from at least one other user according to the preset similarity threshold and at least one user similarity.

In this step, the similarity of the user is calculated according to the calculation formula in the previous step, and after the similarity between other users and the target user is obtained, in order to further determine the user more similar to the target user, further, the other users are screened according to a preset similarity threshold, and at least one other user with the user similarity greater than the similarity threshold is determined as at least one neighbor user.

Optionally, a number threshold may be preset to control the number of neighbor users, so that the data is more accurate.

By way of example, setting the similarity threshold to 0.5 and the number threshold to 10, the first 10 other users with similarity greater than 0.5 are determined to be neighbor users.

S305: according to the identity information of the target user, the target product set, at least one user similarity and at least one neighbor user, calculating to obtain a prediction score of the target user on each first product through a weighted average algorithm based on time attenuation.

In this step, after at least one neighbor user is determined, in order to determine the matching degree between the product and the target user, the prediction score of the target user for each first product is calculated by a weighted average algorithm based on time attenuation.

Specifically, the predictive score is calculated by the following formula:

wherein,representing the predictive score of the target user for the product, +.>Representing the average score of the target user on the product, N _u Representing a set of neighbor users, r _vi Representing the score of other users on the product, +.>Representing the average score, w, of other users on the product _ui The time decay weight representing the target user's score for product i, sim (u, v) represents the similarity of the target user to other users.

S306: and sequencing each first product from large to small according to the predictive score, and determining a first recommendation list according to a preset scoring threshold.

In the step, after the predictive score of the target user on the first products is calculated, in order to determine the products recommended to the target user, the first recommendation list is determined according to the predictive score of each first product from large to small and a preset scoring threshold.

Specifically, a first product with a predictive score greater than a scoring threshold value of the first product is determined to be an alternative product, and the alternative products are summarized to obtain a first recommendation list.

Optionally, the number threshold may be preset, so as to control the number of the candidate products, and further make the data more accurate.

For example, if the score threshold is set to 70 and the quantity threshold is set to 10, the first 10 first products with predictive scores greater than 70 are determined as candidate products.

According to the product recommendation method provided by the embodiment, product evaluation data of a banking system are obtained from a database of the banking system, then a target product set is determined according to identity information of a target user, a user portrait set, the product evaluation data and the product portrait set, further at least one user similarity is calculated according to the identity information of the target user and the target product set through a pearson correlation coefficient based on time attenuation, at least one neighbor user is determined from at least one other user according to a preset similarity threshold and the at least one user similarity, then a prediction score of the target user for each first product is calculated according to the identity information of the target user, the target product set, the at least one user similarity and the at least one neighbor user through a weighted average algorithm based on time attenuation, and finally each first product is ranked from large to small according to the prediction score, and a first recommendation list is determined according to the preset score threshold. By the method, stability, flexibility and accuracy of product recommendation are improved.

Fig. 4 is a schematic flow chart of a third embodiment of a product recommendation method provided in the embodiment of the present application, as shown in fig. 4, on the basis of the foregoing embodiment, step S204 specifically includes:

s401: and carrying out graph construction processing according to the identity information of the target user, the user image set and the product image set to obtain the heterogram.

In this step, in order to make the product recommended to the target user more suitable for the target user, after the first recommendation list is obtained, the product is continuously screened according to the user image set and the product image set, so that the different composition is obtained by performing the graph construction processing according to the identity information of the target user, the user image set and the product image set. The heterogeneous graph comprises user nodes, product nodes and attribute nodes.

Specifically, a user-user similarity graph is drawn through the similarity between the target user and other users obtained through calculation in the above embodiment, and similarly, the product similarity is calculated according to the formula for calculating the similarity in the above embodiment to obtain a product-product similarity graph, and according to the product evaluation data, a user-product score graph is obtained, a user-attribute graph is obtained according to a user image set, and according to a product image set, a product-attribute graph is obtained. And carrying out fusion construction on the 5 graphs to obtain the iso-graph.

S402: and starting from a user node of the target user, performing wandering on the heterogeneous graph according to a preset random wandering rule to obtain a wandering result.

In the step, after the heterogram is obtained, random walk rules preset by taking a target user node as a starting point are used for carrying out follow-up walk, and the visit times of each node in the walk process are recorded to obtain a walk result.

Optionally, the random walk rule specifically includes:

(1) When the current node is a user node, determining the next node needing to walk according to a preset first probability.

Illustratively, with probability p1, selecting the user node most similar to the current user, and continuing to walk;

selecting the product node scored by the current user according to the probability p2, and continuing to walk;

with probability p3, the attribute node that best matches the representation of the current user is selected and the walk is continued.

Wherein p1+p2+p3=1, and p1, p2, p3 can be freely set according to actual conditions.

(2) When the current node is a product node, determining the next node needing to walk according to a preset second probability.

Illustratively, with probability q1, selecting user nodes scored for the current product, and continuing to walk;

Selecting a product node most similar to the current product according to the probability q2, and continuing to walk;

and selecting an attribute node which is most matched with the portrait of the current product with probability q3, and continuing to walk.

Wherein q1+q2+q3=1, and q1, q2, q3 can be freely set according to actual conditions.

(3) When the current node is an attribute node, determining the next node needing to walk according to a preset third probability.

Illustratively, with probability r1, selecting the user node that best matches the current attribute, and continuing to walk;

and selecting a product node which is most matched with the current attribute by using the probability r2, and continuing to walk.

Wherein r1+r2=1, and r1, r2 can be freely set according to actual conditions.

S403: and determining a second recommendation list according to the identity information of the target user, the user portrait set, the product portrait set and the wandering result.

In this step, after the walk result is obtained, the number of accesses of the product node in the walk result represents the correlation with the target user, and then the second recommendation list is determined according to the identity information of the target user, the number of accesses of the product node in the walk result, the user image set and the product image set.

Specifically, according to the identity information of the target user, the user image set and the product image set, matching degree calculation is carried out on the target user and each product to obtain the matching degree of each product, all products are screened according to the matching degree of each product and a preset matching degree threshold value, at least one screened second product is obtained, the at least one second product is ranked from large to small according to the access times, and a second recommendation list is determined according to the preset access times threshold value.

According to the product recommendation method provided by the embodiment, the different composition is obtained by conducting image construction processing according to the identity information of the target user, the user image set and the product image set, then the heterogeneous image is walked according to the preset random walk rule from the user node of the target user to obtain a walk result, and finally the second recommendation list is determined according to the identity information of the target user, the user image set, the product image set and the walk result. By the method, product recommendation intellectualization is realized, and timeliness and flexibility of product recommendation are improved.

Fig. 5 is a schematic flow chart of a fourth embodiment of a product recommendation method provided in the embodiment of the present application, as shown in fig. 5, based on the foregoing embodiments, step S403 specifically includes:

S501: and calculating the matching degree of the target user and each product according to the identity information of the target user, the user image set and the product image set to obtain the matching degree of each product.

In this step, after the walk result is obtained, in order to improve the accuracy of product recommendation, before determining the second recommendation list according to the number of product node accesses, the matching degree between the user and the product needs to be screened, and then the matching degree between the target user and the product is calculated to obtain the matching degree of each product.

Specifically, the matching degree between the target user and the product can be obtained by the following formula:

wherein L is _u Representing a user representation, L _i Representing a product representation. match (u, i) represents the degree of match between the user and the product, and the value range is [0,1 ]]The larger the more matched.

S502: and screening all the products according to the matching degree of each product and a preset matching degree threshold value to obtain at least one screened second product.

In the step, the matching degree of the target user and the product is calculated through the steps, products are screened according to a preset matching degree threshold, and the product with the matching degree larger than the matching degree threshold is determined to be at least one second product.

S503: and sequencing at least one second product according to the access times from large to small, and determining a second recommendation list according to a preset access times threshold.

In the step, after at least one second product is determined, at least one second product is ordered according to the number of times of access from large to small, a threshold value of the number of times of access is preset, the second product with the number of times of access being greater than the threshold value of the number of times of access is determined as an alternative product, and a second recommendation list is obtained through summarizing the alternative products.

According to the product recommendation method provided by the embodiment, the matching degree calculation is carried out on the target user and each product according to the identity information of the target user, the user image set and the product image set to obtain the matching degree of each product, all products are screened according to the matching degree of each product and a preset matching degree threshold value to obtain at least one screened second product, the at least one second product is ranked according to the access times from large to small, and a second recommendation list is determined according to the preset access times threshold value. By the method, accuracy and flexibility of product recommendation are improved.

Fig. 6 is a schematic flow chart of a fifth embodiment of a product recommendation method provided in the embodiment of the present application, as shown in fig. 6, on the basis of the foregoing embodiments, step S202 specifically includes:

S601: the user data set and the product data set are read from the database.

In this step, in order to obtain the user set and the product portrait set, all user data of the banking system are collected in the database to be the user data set, and all product data are collected to be the product data set.

Optionally, the user data set includes basic information, behavior data, preference data, and the like of at least one user, where the basic information of the user includes identity information, registration information, login information, browsing information, purchase information, and rating information of the user. The product data set comprises basic information, functional data, risk data, income data and the like of at least one product, and the basic information of the product comprises identification of the product.

S602: and respectively cleaning the data of the user data set and the product data set to obtain a cleaned user data set and a cleaned product data set.

In this step, after the user data set and the product data set are obtained, in order to enable the subsequent calculation to be more accurate, data cleaning is performed on the user data set and the product data set, respectively, to obtain a cleaned user data set and a cleaned product data set.

Specifically, the data cleaning includes duplicate data removal, missing value supplementation, outlier screening, and data normalization.

S603: and carrying out feature extraction on the cleaned user data set to obtain a user feature set, and carrying out feature extraction on the cleaned product data set to obtain a product feature set.

In this step, after the user data set and the product data set are cleaned, in order to obtain features related to product purchase, feature extraction is performed on the cleaned user data set, and feature extraction is performed on the cleaned product data set.

Specifically, the user data set is extracted to extract the characteristics of the user such as age, gender, income, occupation, consumption capability, risk preference, investment target and the like, the product data set is extracted to extract the characteristics of the product type, deadline, yield, risk grade, applicable crowd and the like, and after the characteristic extraction is completed, the characteristics are selected to obtain the characteristics capable of reflecting the relevance of the user and the product, so that the user characteristic set and the product characteristic set are obtained.

Alternatively, feature selection may be achieved by correlation analysis, information gain, chi-square test, and the like.

S604: and labeling the user characteristics to obtain a user image set, and labeling the product characteristics to obtain a product image set.

In this step, after the user feature set and the product feature set are obtained, in order to obtain the user portrait and the product portrait, the user feature set and the product feature set are respectively subjected to labeling processing, so as to obtain the user portrait set and the product portrait set.

Illustratively, the ages of the users are divided into six sections of [18,25], [26,35], [36,45], [46,55], [56,65], [66,75], the types of the products are divided into six categories of periodic financial management, active financial management, funds, stocks, bonds and insurance, and the characteristics are converted into discrete labels, for example, the age labels of the users are [18,25], and the type labels of the products are periodic financial management.

According to the product recommendation method provided by the embodiment, a user data set and a product data set are read from a database, data cleaning is respectively carried out on the user data set and the product data set to obtain a cleaned user data set and a cleaned product data set, feature extraction is carried out on the cleaned user data set to obtain a user feature set, feature extraction is carried out on the cleaned product data set to obtain a product feature set, finally, labeling treatment is carried out on the user feature to obtain a user image set, and labeling treatment is carried out on the product feature to obtain a product image set. By the method, the coverage rate and diversity of product recommendation are improved.

Fig. 7 is a schematic flow chart of a sixth embodiment of a product recommendation method provided in the embodiment of the present application, as shown in fig. 7, based on the foregoing embodiments, step S206 specifically includes:

s701: and determining at least one target product from the plurality of candidate products according to a preset scoring threshold value.

S702: and pushing the target product list to the user according to at least one target product.

After the target score of each candidate product is obtained, in order to improve the accuracy and stability of recommending the products for the user, a scoring threshold value is preset, products with target scores larger than the scoring threshold value in the candidate products are used as target products, the determined target products are summarized to obtain a target product list, and then the target product list is recommended to the target user.

Optionally, in order to facilitate the user to select a product suitable for the user, a quantity threshold may be set, and a preset quantity of target products are selected from the determined at least one target product and summarized into a target product list for recommendation.

In one possible design, after the target product list is obtained, the list may be displayed to the target user through a display screen of the front-end equipment of the bank hall, or the target product list may be sent to a terminal equipment of the target user and displayed through a configured application program interface.

According to the product recommendation method provided by the embodiment, at least one target product is determined from a plurality of candidate products according to the preset scoring threshold, and a target product list is pushed to a user according to the at least one target product. By the method, accuracy and timeliness of product recommendation are improved.

Fig. 8 is a schematic structural diagram of a first embodiment of a product recommendation device provided in an embodiment of the present application, and as shown in fig. 8, a product recommendation device 800 includes:

the receiving module 801 is configured to receive a recommendation request sent by a user, where the recommendation request includes identity information of a target user.

An obtaining module 802, configured to obtain, according to the recommendation request, a user portrait set and a product portrait set, where the user portrait set includes user portraits of all users in the banking system, and the product portrait set includes product portraits of all products in the banking system.

The first calculating module 803 is configured to calculate, according to the identity information of the target user, the user image set, and the product image set, a first recommendation list according to a collaborative filtering recommendation algorithm based on a time decay function, where the first recommendation list includes at least one candidate product.

The second calculating module 804 is configured to calculate, according to the identity information of the target user, the user image set, and the product image set, a second recommendation list according to a recommendation algorithm based on the improved random walk pattern, where the second recommendation list includes at least one candidate product.

The third calculation module 805 is configured to input the identity information of the target user, the first recommendation list, and the second recommendation list into a preset reorder recommendation model, and generate a target score of each candidate product, where the reorder recommendation model is a model for calculating a product score obtained by training the XGBoost model according to the user image set and the product image set, and the candidate product is any one of the first recommendation list and the second recommendation list.

A pushing module 806, configured to push at least one target product to the target user according to the target score of each candidate product.

Fig. 9 is a schematic structural diagram of a second embodiment of a product recommendation device provided in the embodiment of the present application, as shown in fig. 9, a first computing module 803 includes:

an acquisition unit 901 for acquiring product evaluation data of a banking system.

The first determining unit 902 is configured to determine a target product set according to identity information of a target user, a user portrait set, product evaluation data, and a product portrait set, where the target product set includes: the product image of the first product that the target user has rated in common with at least one other user, the score of each user for each first product, and the time stamp of the score of each user for each first product.

The first calculating unit 903 is configured to calculate, according to the identity information of the target user and the set of target products, at least one user similarity through pearson correlation coefficients based on time attenuation, where the at least one user similarity includes a similarity between the target user and at least one other user.

A second determining unit 904, configured to determine at least one neighbor user from at least one other user according to a preset similarity threshold and at least one user similarity, where the at least one neighbor user includes a user whose user similarity with the target user is greater than the similarity threshold.

The second calculating unit 905 is configured to calculate, according to the identity information of the target user, the set of target products, at least one user similarity, and at least one neighboring user, a prediction score of the target user for each first product by using a weighted average algorithm based on time attenuation.

And a third determining unit 906, configured to rank each first product according to the prediction scores from large to small, and determine the first recommendation list according to a preset score threshold.

Fig. 10 is a schematic structural diagram of a third embodiment of a product recommendation device provided in the embodiment of the present application, as shown in fig. 10, a second computing module 804 includes:

The composition unit 1001 is configured to perform a graph construction process according to identity information of a target user, a user image set, and a product image set to obtain an iso-graph, where the iso-graph includes a user node, a product node, and an attribute node.

The wander unit 1002 is configured to walk the heterogeneous graph according to a preset random walk rule from a user node of the target user, to obtain a walk result, where the walk result includes the access times of all nodes.

The determining unit 1003 is configured to determine the second recommendation list according to the identity information of the target user, the user portrait set, the product portrait set and the walk result.

In one possible design, on the basis of the foregoing embodiments, in the product recommendation device provided in this embodiment, the random walk rule includes: when the current node is a user node, determining a next node needing to walk according to a preset first probability, wherein the first probability comprises user node probability, product node probability and attribute node probability matched with the current node, and the sum of the user node probability, the product node probability and the attribute node probability is 1;

When the current node is an attribute node, determining the next node needing to walk according to a preset third probability, wherein the third probability comprises a user node probability and a product node probability which are matched with the current node, and the sum of the user node probability and the product node probability is 1.

Optionally, the determining unit 1003 is specifically configured to:

according to the identity information of the target user, the user image set and the product image set, carrying out matching degree calculation on the target user and each product to obtain the matching degree of each product;

screening all products according to a preset matching degree threshold value according to the matching degree of each product to obtain at least one screened second product, wherein the matching degree of each second product is larger than the matching degree threshold value;

and sequencing at least one second product according to the access times from large to small, and determining a second recommendation list according to a preset access times threshold.

Fig. 11 is a schematic structural diagram of a fourth embodiment of a product recommendation device provided in the embodiment of the present application, as shown in fig. 11, on the basis of the foregoing embodiments, an obtaining module 802 specifically includes:

the reading unit 1101 is configured to read, from the database, a user data set including data of all users in the banking system and a product data set including data of all products in the banking system.

And the cleaning unit 1102 is configured to perform data cleaning on the user data set and the product data set respectively, so as to obtain a cleaned user data set and a cleaned product data set.

The feature extraction unit 1103 is configured to perform feature extraction on the cleaned user data set to obtain a user feature set, and perform feature extraction on the cleaned product data set to obtain a product feature set.

The labeling processing unit 1104 is configured to perform labeling processing on the user feature to obtain a user image set, and perform labeling processing on the product feature to obtain a product image set.

Fig. 12 is a schematic structural diagram of a fifth embodiment of a product recommendation device provided in the embodiment of the present application, as shown in fig. 12, on the basis of the foregoing embodiments, a pushing module 806 specifically includes:

and the determining unit 1201 is configured to determine at least one target product from the multiple candidate products according to a preset scoring threshold, where the at least one target product is a candidate product whose target score is greater than the scoring threshold.

And a pushing unit 1202, configured to push, to a target user, a target product list according to at least one target product, where the target product list includes at least one target product ordered according to the target scores from large to small.

The product recommendation device provided by the application can execute the product recommendation method in the embodiment of the method, and the implementation principle and the technical effect are similar and are not repeated here.

Fig. 13 is a schematic structural diagram of an electronic device provided in the present application. As shown in fig. 13, the electronic device 1300 may include at least one memory 1301, a processor 1302, which may be, for example, a computer, a server, or the like, having processing capabilities.

A memory 1301 for storing a program. In particular, the program may include program code including computer-operating instructions. The memory 502 may comprise high-speed RAM memory or may further comprise non-volatile memory (non-volatile memory), such as at least one disk memory.

The processor 1302 is configured to execute computer-executable instructions stored in the memory 1301 to implement the backup redundancy processing method described in the foregoing method embodiment. The processor 1302 may be a central processing unit (Central Processing Unit, abbreviated as CPU), or an application specific integrated circuit (Application Specific Integrated Circuit, abbreviated as ASIC), or one or more integrated circuits configured to implement embodiments of the present application.

The electronic device 1300 may also include a communication interface 1303 to enable communication interactions with external devices through the communication interface 1303. The external device may be, for example, an electronic device such as a computer.

In a specific implementation, if the communication interface 1303, the memory 1301, and the processor 1302 are implemented independently, the communication interface 1303, the memory 1301, and the processor 1302 may be connected to each other through a bus and perform communication with each other. The bus may be an industry standard architecture (Industry Standard Architecture, abbreviated ISA) bus, an external device interconnect (Peripheral Component, abbreviated PCI) bus, or an extended industry standard architecture (Extended Industry Standard Architecture, abbreviated EISA) bus, among others. Buses may be divided into address buses, data buses, control buses, etc., but do not represent only one bus or one type of bus.

Alternatively, in a specific implementation, if the communication interface 1303, the memory 1301, and the processor 1302 are integrated on a single chip, the communication interface 1303, the memory 1301, and the processor 1302 may complete communication through internal interfaces.

The present application also provides a computer-readable storage medium, which may include: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk or an optical disk, etc., in which program codes can be stored, and specifically, the computer-readable storage medium stores therein program instructions for the product recommendation method in the above-described embodiment.

The present application also provides a computer program product comprising a computer program stored in a readable storage medium. The at least one processor of the electronic device 1300 may read the computer program from the readable storage medium, and execution of the computer program by the at least one processor causes the electronic device 1300 to implement the product recommendation method provided by the various embodiments described above.

The application also provides a chip, and a computer program is stored on the chip, and when the computer program is executed by the chip, the product recommendation method provided by various embodiments is realized.

Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

It is to be understood that the present application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims

1. A product recommendation method, applied to a server of a banking system, comprising:

And pushing at least one target product to the target user according to the target score of each candidate product.

2. The method according to claim 1, wherein the calculating the first recommendation list according to the identity information of the target user, the user portrayal set, and the product portrayal set by a collaborative filtering recommendation algorithm based on a time decay function includes:

acquiring product evaluation data of the banking system;

3. The method of claim 1, wherein the calculating a second recommendation list based on the identity information of the target user, the user portrayal set, the product portrayal set by a modified random walk pattern recommendation algorithm comprises:

4. A method according to claim 3, wherein the random walk rule comprises:

5. The method of claim 3, wherein the determining the second recommendation list based on the identity information of the target user, the set of user portraits, the set of product portraits, and the walk-through result comprises:

6. The method according to any one of claims 1 to 5, wherein the obtaining a user portrait collection and a product portrait collection according to the recommendation request includes:

7. The method of claim 1, wherein pushing at least one target product to the target user based on the target score for each of the candidate products comprises:

and pushing a target product list to the target user according to the at least one target product, wherein the target product list comprises the at least one target product which is ranked from large to small according to a target score.

8. A product recommendation device, comprising:

the acquisition module is used for acquiring a user portrait set and a product portrait set according to the recommendation request, wherein the user portrait set comprises user portraits of all users in a banking system, and the product portrait set comprises product portraits of all products in the banking system;

and the pushing module is used for pushing at least one target product to the target user according to the target score of each candidate product.

9. An electronic device, comprising: the device comprises a memory, a processor, a communication interface and a display;

the memory stores computer-executable instructions;

the processor executes computer-executable instructions stored in the memory to implement the method of any one of claims 1 to 7.

10. A computer readable storage medium having stored therein computer executable instructions which when executed by a processor are adapted to carry out the method of any one of claims 1 to 7.

11. A computer program product, characterized in that the computer program product comprises a computer program which, when executed by a processor, implements the method according to any of claims 1 to 7.