CN115774813A

CN115774813A - Product recommendation method and device, computer equipment and storage medium

Info

Publication number: CN115774813A
Application number: CN202211579632.9A
Authority: CN
Inventors: 钟亚洲
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2022-12-09
Filing date: 2022-12-09
Publication date: 2023-03-10

Abstract

The application relates to a product recommendation method, a product recommendation device, computer equipment and a storage medium, which are applied to the field of financial technology or other related fields, wherein the method comprises the following steps: acquiring object characteristic information of multiple dimensions of a target object; determining a predicted portrait label for the target object based on the object feature information for the plurality of dimensions; and determining a target product from a plurality of candidate products according to the predicted portrait label, and pushing the target product to a client associated with the target object. By the method, the adaptation degree between the recommended product and the target object can be improved, accurate pushing of the product is realized, and the utilization rate of the pushing information is improved.

Description

Product recommendation method and device, computer equipment and storage medium

Technical Field

The present application relates to the field of financial technology or other related fields, and in particular, to a method, an apparatus, a computer device, a storage medium, and a computer program product for recommending a product.

Background

With the rapid development of the financial industry, at present, financial institutions recommend some financial products to be recommended to each user, specifically, there are various recommendation modes, for example, a short message group sending mode is used to send links and brief contents of a financial product to all users of the financial institutions, or popularization information of the financial product is set in a preset position of a financial APP home page.

However, the current product recommendation method cannot accurately push the financial products to the users interested in the financial products, which leads to resource waste when pushing information.

Disclosure of Invention

Accordingly, it is desirable to provide a product recommendation method, an apparatus, a computer device, a computer-readable storage medium, and a computer program product for solving the technical problems of single recommendation mode and poor recommendation result of the product recommendation method.

In a first aspect, the present application provides a product recommendation method. The method comprises the following steps:

acquiring object characteristic information of multiple dimensions of a target object;

determining a predicted portrait label for the target object based on the object feature information for the plurality of dimensions;

and determining a target product from a plurality of candidate products according to the predicted portrait label, and pushing the target product to a client associated with the target object.

In one embodiment, the determining a target product from a plurality of candidate products according to the predicted portrait label includes:

acquiring product information of each candidate product;

determining interest information of the target object in each candidate product based on the predicted portrait label and the product information of each candidate product;

and determining a target product from the candidate products according to the interest information of the target object to each candidate product.

In one embodiment, the determining interest information of the target object in the candidate products based on the predicted portrait tags and the product information of the candidate products includes:

for each candidate product, determining a product type of the candidate product;

determining a target Markov model corresponding to the candidate product based on the product type;

and inputting the product information of the candidate product and the predicted portrait label into a target Markov model corresponding to the candidate product to obtain the interest information of the target object to the candidate product.

In one embodiment, the method further includes a training process of a markov model corresponding to each product type, and the training process of the markov model corresponding to each product type includes:

for each product type, acquiring a sample product, a portrait label of a sample object and an interest label of the sample object for the sample product, wherein the sample product and the portrait label are the same as the product type;

based on a Markov decision process, taking various prediction results of the sample object on the sample product as states and taking a behavior of recommending the sample object to the product as an action;

and when the Markov model is transferred from one state to another state, determining feedback information aiming at the transfer process, and adjusting the model parameters of the Markov model to be trained according to the feedback information until a training termination condition is reached to obtain the Markov model after training as the Markov model corresponding to the product type.

In one embodiment, the determining a target product from the plurality of candidate products according to the interest information of the target object in each candidate product includes:

sequencing the candidate products according to the interest degrees represented by the interest information to obtain a candidate product sequence;

and determining products with interest degrees meeting preset conditions from the candidate product sequence as the target products.

In one embodiment, the determining a predicted image label for the target object based on the object feature information of the plurality of dimensions includes:

and inputting the object feature information of the multiple dimensions into a trained label prediction model to obtain a predicted image label for the target object.

In a second aspect, the application also provides a product recommendation device. The device comprises:

the acquisition module is used for acquiring object characteristic information of multiple dimensions of the target object;

a prediction module to determine a predicted image label for the target object based on the object feature information for the plurality of dimensions;

and the recommending module is used for determining a target product from a plurality of candidate products according to the predicted portrait label and pushing the target product to a client associated with the target object.

In one embodiment, the recommendation module further includes:

the information acquisition sub-module is used for acquiring product information of each candidate product;

an interest determination sub-module, configured to determine interest information of the target object for each candidate product based on the predicted portrait label and product information of the candidate product;

and the target determining submodule is used for determining a target product from the candidate products according to the interest information of the target object to each candidate product.

In one embodiment, the interest determination sub-module is further configured to determine, for each candidate product, a product type of the candidate product; determining a target Markov model corresponding to the candidate product based on the product type; and inputting the product information of the candidate product and the predicted portrait label into a target Markov model corresponding to the candidate product to obtain the interest information of the target object to the candidate product.

In one embodiment, the apparatus further comprises a model training module for obtaining, for each product type, a sample product of the same product type as the product type, a portrait label of a sample object, and an interest label of the sample object for the sample product; based on a Markov decision process, taking various prediction results of the sample object on the sample product as states and taking a behavior of recommending the sample object to the product as an action; and when the Markov model is transferred from one state to another state, determining feedback information aiming at the transfer process, and adjusting the model parameters of the Markov model to be trained according to the feedback information until a training termination condition is reached to obtain the Markov model after training as the Markov model corresponding to the product type.

In one embodiment, the target determination sub-module is further configured to rank the plurality of candidate products according to the interest degrees represented by the interest information to obtain a candidate product sequence; and determining products with interest degrees meeting preset conditions from the candidate product sequence as the target products.

In one embodiment, the prediction module is further configured to input the object feature information of the multiple dimensions into a trained label prediction model to obtain a predicted image label for the target object.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor implementing the following steps when executing the computer program:

determining a predicted portrait label for the target object based on the object feature information of the plurality of dimensions;

In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:

In a fifth aspect, the present application further provides a computer program product. The computer program product comprising a computer program which when executed by a processor performs the steps of:

According to the product recommendation method, the product recommendation device, the computer equipment, the storage medium and the computer program product, the predicted portrait label aiming at the target object is determined through the object feature information of multiple dimensions of the target object, and the prediction of the personal interest of the target object is realized, so that the target product determined from multiple candidate products can better meet the self requirement of the target object according to the predicted portrait label, the adaptation degree between the recommended product and the target object is improved, the accurate pushing of the product can be realized, and the utilization rate of the pushed information is improved.

Drawings

FIG. 1 is a flow diagram that illustrates a method for recommending products in one embodiment;

FIG. 2 is a schematic diagram of a Markov decision process in one embodiment;

FIG. 3 is a flow diagram that illustrates the processing of the tag prediction model in one embodiment;

FIG. 4 is a system architecture diagram that illustrates an automated mining of a user portrait label hierarchy, in one embodiment;

FIG. 5 is a schematic flow chart diagram illustrating a representation data query service in one embodiment;

FIG. 6 is a flow diagram that illustrates a target guest group screening service in one embodiment;

FIG. 7 is a flowchart illustrating a product recommendation method according to another embodiment;

FIG. 8 is a block diagram showing the construction of a product recommendation device in one embodiment;

FIG. 9 is a diagram of an internal structure of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

It should be further noted that the user information (including but not limited to user device information, user personal information, etc.) and 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 sufficiently authorized by each party.

In an embodiment, as shown in fig. 1, a product recommendation method is provided, and this embodiment is illustrated by applying the method to a terminal, and it may be understood that the method may also be applied to a server, and may also be applied to a system including the terminal and the server, and is implemented through interaction between the terminal and the server. The terminal can be but not limited to various personal computers, notebook computers, smart phones, tablet computers, internet of things equipment and portable wearable equipment, and the internet of things equipment can be smart sound boxes, smart televisions, smart air conditioners, smart vehicle-mounted equipment and the like. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, and the like. The server may be implemented as a stand-alone server or as a server cluster consisting of a plurality of servers. In this embodiment, the method includes the steps of:

step S110, object feature information of multiple dimensions of the target object is obtained.

The object characteristic information of multiple dimensions may include at least two of location characteristic information, business characteristic information, social characteristic information, evaluation characteristic information, risk characteristic information, financial characteristic information, preference characteristic information, marketing characteristic information, basic characteristic information, product characteristic information, behavior characteristic information, association characteristic information, and the like.

The position characteristic information can be information such as action tracks of individual target objects, company registration addresses of legal target objects, actual office addresses, logistics and the like.

The operation characteristic information can be the operation condition of an enterprise under the name of the individual target object, the enterprise operation condition of the legal target object, the water and electricity tax and the like.

The social characteristic information can be social information of the target object on various websites and application programs.

Wherein, the evaluation characteristic information can be a historical risk rating of the target object.

The risk characteristic information may be risk information of the target object, such as credit investigation, prohibition, early warning, public opinion, deterioration prediction, and the like.

The financial characteristic information may be information such as financial statements and indexes of the target object.

The preference characteristic information can be various types of preference information such as risk preference, channel preference and the like of the target object.

The marketing characteristic information may be information such as marketing activities performed on the target object and the depth of each channel contact.

The basic characteristic information can be demographic information, contact information and the like of the individual target object; shareholder information, high-management information, industry information, and the like of the corporate target object.

The product characteristic information can be information such as agreements signed by target objects at home and abroad, held products, and each collateral of account balance.

The behavior characteristic information may be fund transaction information, investment and financing, co-purchase behavior information and the like of the target object.

The associated characteristic information may be associated information of a fund association, a guarantee association, an investment association, a group association, and the like of the target object.

In specific implementation, it is considered that feature information of some dimensions cannot be directly obtained, and therefore, initial feature information of the dimensions of the target object in a preset time period can be obtained, and the initial feature information in the time period is further subjected to statistical processing to obtain object feature information of corresponding dimensions.

Step S120, based on the object feature information of the plurality of dimensions, determines a predicted image tag for the target object.

The predicted image label represents a prediction of a user behavior, a product interaction behavior, and the like.

In specific implementation, when object feature information of a target object under multiple dimensions is obtained, the object feature information of the multiple dimensions can be input into a label prediction model which is trained in advance, the label prediction model processes the object feature information of each dimension, and a predicted image label of the target object is output.

And step S130, determining a target product from the candidate products according to the predicted portrait label, and pushing the target product to a client associated with the target object.

In the concrete implementation, when a target product is determined, product information of each candidate product is required to be obtained, the product information of each candidate product and a prediction image label of a target object are predicted through a Markov model, interest information of the target object to each candidate product is obtained, the target product is determined from a plurality of candidate products according to the interest information of the target object to each candidate product, and then the target product is pushed to a client side related to the target object.

According to the product recommendation method, the prediction portrait label aiming at the target object is determined through the object feature information of multiple dimensions of the target object, and the prediction of the personal interest of the target object is realized, so that the target product determined from multiple candidate products can better meet the self requirement of the target object according to the prediction portrait label, the adaptation degree between the recommended product and the target object is improved, the personalized recommendation aiming at the target object is realized, and the recommendation effect of the product can be improved.

In an exemplary embodiment, the step S130 of determining the target product from the candidate products according to the predicted portrait label includes:

step S131, product information of each candidate product is obtained;

step S132, based on the predicted portrait label and the product information of each candidate product, determining the interest information of the target object to each candidate product;

and step S133, determining the target product from the candidate products according to the interest information of the target object to each candidate product.

The product information may include product names, product attributes, product description information, prices, and other information characterizing the product of the candidate products.

The interest information is an index that characterizes a degree of interest of the target object in the candidate product, for example, the interest information may be a purchase probability.

In specific implementation, before determining a target product, product information of each candidate product needs to be acquired, for each candidate product, interest information of the target object in the candidate product is determined based on the product information of the candidate product and a predicted pictorial label of the target object, and specifically, the interest information of the target object in the candidate product and the predicted pictorial label of the target object can be processed through an interest prediction model (for example, a markov model) trained in advance to obtain the interest information of the target object in the candidate product. After the interest information of the target object to each candidate product is obtained, a plurality of products with high interest degree of the target object can be selected from the candidate products to serve as the target products according to the interest information of the target object to each candidate product.

In the embodiment, the interest information of the target object to each candidate product is determined through the predicted portrait label of the target object and the product information of each candidate product, and the target product pushed to the target object is further determined based on the interest information of the target object to each candidate product, so that the determined target product can better accord with the interest of a user, is more adaptive to the user, and can improve the user experience.

In an exemplary embodiment, in step S132, determining interest information of the target object in each candidate product based on the predicted portrait label and the product information of each candidate product includes:

step S1321, aiming at each candidate product, determining the product type of the candidate product;

step S1322, determining a target Markov model corresponding to the candidate product based on the product type;

and step S1323, inputting the product information of the candidate product and the predicted portrait label into a target Markov model corresponding to the candidate product to obtain the interest information of the target object to the candidate product.

In the specific implementation, each product type has a corresponding markov model, before determining interest information of a target object to each candidate product, the product type of each candidate product needs to be determined first, the markov model corresponding to each candidate product is determined based on the product type, the product information of each candidate product and a prediction portrait label are further input into the markov model corresponding to the candidate product, and the interest information of the target object to the candidate product is obtained.

In the embodiment, in consideration of the influence of different product types on the interest information prediction result, different Markov models are set for different product types, and in the prediction of interest information, the candidate product is predicted through the Markov model corresponding to the candidate product, so that the confidence coefficient of the obtained interest information can be improved, and the adaptation degree of the candidate recommended target product and the target object is improved.

In an exemplary embodiment, the method further includes a training process of a markov model corresponding to each product type, and the training process of the markov model corresponding to each product type includes:

step S210, acquiring a sample product with the same product type, a portrait label of a sample object and an interest label of the sample object on the sample product aiming at each product type;

step S220, based on the Markov decision process, taking various prediction results of the sample object on the sample product as states and taking the behavior of recommending the sample object on the product as an action;

step S230, when the state is transferred to another state, determining feedback information aiming at the transfer process, adjusting model parameters of the Markov model to be trained according to the feedback information until a training termination condition is reached, and obtaining the Markov model after training as the Markov model corresponding to the product type.

In specific implementation, after a predicted portrait label of a target object is determined, in order to improve the capability of the predicted portrait label for supporting a scene, the application provides a product recommendation algorithm based on Markov decision reinforcement learning. The predicted portrait label is in scene E, with state space X, where each state X ∈ X is a description of the predicted portrait label in the scene environment. The input data of the Markov model is product information of a sample product and a portrait label of a sample object, and the output data is a mark of successful recommendation or failed recommendation.

Before training, firstly defining a scene recommendation state and a conversion process: and determining four states of potential objects, recommendation success, recommendation failure and object aversion according to the prediction result. Wherein, the potential object is an object which has a higher probability of purchasing the recommended product; the marketing success is the behavior of the target object to purchase the recommended product in the product recommendation period; marketing failure is the behavior that the target object does not purchase the recommended product within the product recommendation period; object aversion is the act of a target object complaint about a recommended product during a product recommendation cycle.

Defining the initial transition probability: and determining the probability of successful recommendation of the target object in the initial recommendation state.

Defining model training termination conditions: conditions for ending the state transition process, such as recommendation success and recommendation failure, are determined according to the scene recommendation target.

Defining an action: the method comprises a first action and a second action, wherein the first action is an action of recommending a product through multiple channels; the second action is an action that does not make a product recommendation to the target object.

As shown in fig. 2, for a product recommendation scenario in the markov decision process, the markov decision process in the product recommendation process is shown, and there are four states in the task: potential object, recommendation success, recommendation failure, and object aversion, two actions: recommended and not recommended. When transitioning from one state step to another, a reward positive value parameter is obtained if the state is promoting successful recommendation, and a negative value parameter is fed back if the state transitions to a potential object or directly results in object aversion. The state may then transition to a recommendation success state with or without recommendation. And feeding back the minimum value when the object recommendation fails. In the figure, arrows represent state transition, parameters a corresponding to the arrows represent actions causing the state transition, p represent transition probability, and r represents a returned feedback value.

In the embodiment, different Markov models are respectively trained for different types of products, so that the corresponding Markov models can be selected to predict the candidate products based on the product types of the candidate products, the confidence of the obtained interest information is improved, and the adaptation degree of the candidate recommended target products and the target object is improved.

In an exemplary embodiment, in step S133, determining the target product from the plurality of candidate products according to the interest information of the target object for each candidate product includes: sequencing a plurality of candidate products according to the interest degree represented by the interest information to obtain a candidate product sequence; and determining products with interest degrees meeting preset conditions from the candidate product sequence as target products.

The interest information is an index representing the interest degree of the target object in the candidate product, for example, the interest information may be a purchase probability.

In specific implementation, based on the interest information of the target object to each candidate product, each candidate product may be ranked in order from high to low or from low to high according to the interest degree represented by the interest information to obtain a candidate product sequence, and the top N candidate products with the highest interest degree or several candidate products with interest information greater than a preset value may be selected from the candidate product sequence as the target product.

For example, if the interest information is a purchase probability, and the purchase probabilities of the target object for 5 candidate products a, B, C, D, and E are 0.5, 0.7, 0.3, 0.6, and 0.2, respectively, the candidate products are ranked in order of the purchase probability from high to low to obtain a candidate product sequence B → D → a → C → E, and the products B and D with the purchase probability greater than 0.5 can be selected as the target products.

In this embodiment, the plurality of candidate products are ranked according to the interest degrees represented by the interest information, and the top N candidate products with the highest interest degree or the products with interest information larger than a preset value are determined from the obtained candidate product sequence and used as the target products, so that the products pushed to the target object can better meet the requirements of the target object, and the quality of the products pushed to the target object is improved.

In an exemplary embodiment, the determining the predicted image tag for the target object based on the object feature information of the plurality of dimensions in step S120 includes: and inputting the object characteristic information of multiple dimensions into the trained label prediction model to obtain a predicted image label for the target object.

The label prediction model may be a Deep learning model, and the Deep learning model may adopt a Deep Network (DNN), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), a Generative Adaptive Network (GAN), and the like.

Specifically, after object feature information of multiple dimensions is input into a trained label prediction model, the label prediction model determines the number of nodes of a hidden layer by adopting an AP (access point) clustering algorithm (Affinity propagation clustering), then a count vector associated with the embedded label is weighted to a network basis function center for training, and after learning the weight between the hidden layer and an output layer, the output layer generates a label vector. The label vector is a fixed-dimension vector composed of real numbers, and each label corresponds to one label vector.

Referring to fig. 3, a processing flow chart of the label prediction model is shown, and the number of hidden nodes is obtained through an AP clustering algorithm, an initial clustering center is constructed, a basis function center is obtained, clustering is performed by combining with each type of label technical vector, and finally a neural network basis function center is obtained, weights are calculated, and predicted portrait labels are generated.

In the embodiment, the predicted portrait label is determined through the object feature information of multiple dimensions of the target object, the influence of multiple factors on the prediction result is fully considered, and the accuracy of the obtained portrait label can be ensured.

In one embodiment, to facilitate understanding of embodiments of the present application by those skilled in the art, reference will now be made to the specific examples illustrated in the drawings. Referring to fig. 4, a system architecture diagram for automatically mining a user portrait label system is shown, which mainly includes a data layer unit 401, a feature library unit 402, a model training unit 403, a label prediction unit 404, and a scenario unit 405, where:

the data layer unit 401 is configured to provide underlying data for building a user portrait label system, and specifically includes: the data lake provides paste source data; the data warehouse provides main data and the information base provides knowledge data, and the knowledge data is source data of user characteristics.

The feature library unit 402 is configured to extract user feature information of the user in multiple dimensions from each data source of the data unit, where the user feature information includes location feature information, business feature information, social feature information, evaluation feature information, risk feature information, financial feature information, preference feature information, marketing feature information, basic feature information, product feature information, behavior feature information, and association feature information.

And a model training unit 403, which is a label mining process based on deep learning, inputs user feature information of each dimension of a user, utilizes the deep model to learn hidden representation, and then generates a predicted image label.

The tag prediction unit 404 is specifically conclusion information for predicting behaviors of the user, product purchasing behaviors, and the like.

And a scene unit 405, configured to provide a corresponding service for the user based on the predicted portrait label. The specific scenes comprise a trigger type product recommendation scene, a target user exploration scene, a resource flow direction analysis scene and the like. The triggered product recommendation scene is used for pushing products according to the predicted image labels of the users; the target user exploration scene is used for carrying out list screening according to information provided by the predicted portrait label in the accurate recommendation system to form a targeted recommendation list and provide a general data screening service for the outside; the resource flow direction analysis scene is used for rapidly and comprehensively knowing the risk condition of the user by means of the labeling, informatization and visualization attributes of the prediction portrait label, providing comprehensive credit decision support for the bank to implement user admission control, examination and approval and post-loan management, and reminding managers of value-added services such as recommendation, user saving, risk monitoring and the like for the user in charge at the first time of transaction occurrence.

In addition, the method also provides a portrait label query service which takes the user number as a key value to channels such as a branch office, an internet bank, a mobile phone bank and the like through an Application Programming Interface (API) and a distributed service Interface (DBI) as shown in fig. 5, supports all the channels, takes the intelligent panoramic user portrait label as a basis, and performs big data and artificial intelligence Application scene landing in the fields of product recommendation, risk, customer service and the like. And providing a target customer group screening service which takes the user characteristics as dimensions through a page service as shown in fig. 6, performing characteristic query in a characteristic library based on customized screening conditions, and screening out a target user list and target user characteristics.

A specific process of recommending products in a trigger-type product recommendation scenario is shown in fig. 7, and includes the following steps:

step S710, acquiring object characteristic information of multiple dimensions of a target object;

step S720, inputting object characteristic information of multiple dimensions into the trained label prediction model to obtain a predicted image label for the target object;

step S730, product information and product types of each candidate product are obtained;

step S740, determining a target Markov model corresponding to each candidate product based on the product type;

step S750, inputting the product information of the candidate product and the predicted portrait label into a target Markov model corresponding to the candidate product to obtain the interest information of the target object to the candidate product;

step S760, sequencing a plurality of candidate products according to the interest degree represented by the interest information to obtain a candidate product sequence;

step S770, determining products with interest degrees meeting preset conditions from the candidate product sequence as target products;

and step S780, pushing the target product to the client associated with the target object.

According to the method, on one hand, the target product is determined from the candidate products according to the predicted portrait label, so that the determined target product can better meet the self requirement of the target object, the adaptation degree between the recommended product and the target object is improved, the personalized recommendation for the target object is realized, and the recommendation effect of the product can be improved; on the other hand, the method for automatically analyzing, mining and generating the customer portrait label can automatically mine data aiming at the scene, namely randomly arranging and combining a plurality of labels, packaging a plurality of high-level label extraction characteristics, automatically learning a machine aiming at the scene by combining the low-level label, the high-level label and the mined characteristics, and obtaining a final conclusion: under a certain scene, which labels, which expert rules effect is most obvious, and what is what the value that each rule of every label brought efficiency promotion through the quantization, when promoting the portrait accuracy, reduce the cost of analyst's manual index construction and operation, effectively solve traditional analyst and build client portrait label and decide subjectivity big, one-sidedness is strong, in time update nature is not enough, inaccurate scheduling problem.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not limited to being performed in the exact order illustrated and, unless explicitly stated herein, may be performed in other orders. Moreover, at least a part of the steps in the flowcharts related to the embodiments described above may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the execution order of the steps or stages is not necessarily sequential, but may be performed alternately or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the present application further provides a product recommendation device for implementing the above-mentioned product recommendation method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the method, so the specific limitations in one or more embodiments of the product recommendation device provided below can be referred to the limitations of the product recommendation method in the above, and are not described herein again.

In one embodiment, as shown in fig. 8, there is provided a product recommendation device including: an obtaining module 810, a predicting module 820, and a recommending module 830, wherein:

an obtaining module 810, configured to obtain object feature information of multiple dimensions of a target object;

a prediction module 820 for determining a predicted image label for the target object based on object feature information of multiple dimensions;

and the recommending module 830 is configured to determine a target product from the candidate products according to the predicted portrait label, and push the target product to a client associated with the target object.

In one embodiment, the recommending module 830 further includes:

the information acquisition submodule is used for acquiring product information of each candidate product;

the interest determination sub-module is used for determining interest information of the target object on each candidate product based on the predicted portrait label and the product information of each candidate product;

and the target determining submodule is used for determining the target product from the candidate products according to the interest information of the target object to each candidate product.

In one embodiment, the device further comprises a model training module, which is used for acquiring a sample product with the same product type as the product type, a portrait label of a sample object and an interest label of the sample object on the sample product for each product type; based on a Markov decision process, taking various prediction results of a sample object on a sample product as states and taking a behavior of recommending the sample object on the product as an action; and when the Markov model is transferred from one state to another state, determining feedback information aiming at the transfer process, and adjusting the model parameters of the Markov model to be trained according to the feedback information until a training termination condition is reached to obtain the Markov model after training as the Markov model corresponding to the product type.

In one embodiment, the target determination sub-module is further configured to rank the plurality of candidate products according to the interest degree represented by the interest information to obtain a candidate product sequence; and determining products with interest degrees meeting preset conditions from the candidate product sequence as target products.

In one embodiment, the prediction module 820 is further configured to input object feature information of multiple dimensions into the trained label prediction model to obtain a predicted image label for the target object.

The modules in the product recommending device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 9. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program to run on the non-volatile storage medium. The communication interface of the computer device is used for communicating with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a product recommendation method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the configuration shown in fig. 9 is a block diagram of only a portion of the configuration associated with the present application, and is not intended to limit the computing device to which the present application may be applied, and that a particular computing device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:

In one embodiment, the processor when executing the computer program further performs the steps of: acquiring product information of each candidate product; determining interest information of the target object in each candidate product based on the predicted portrait label and the product information of each candidate product; and determining a target product from the candidate products according to the interest information of the target object to each candidate product.

In one embodiment, the processor when executing the computer program further performs the steps of: for each candidate product, determining a product type of the candidate product; determining a target Markov model corresponding to the candidate product based on the product type; and inputting the product information of the candidate product and the predicted portrait label into a target Markov model corresponding to the candidate product to obtain the interest information of the target object to the candidate product.

In one embodiment, the processor, when executing the computer program, further performs the steps of: for each product type, acquiring a sample product, a portrait label of a sample object and an interest label of the sample object for the sample product, wherein the sample product and the portrait label are the same as the product type; based on a Markov decision process, taking various prediction results of the sample object on the sample product as states and taking a behavior of performing product recommendation on the sample object as an action; and when the Markov model is transferred from one state to another state, determining feedback information aiming at the transfer process, and adjusting the model parameters of the Markov model to be trained according to the feedback information until a training termination condition is reached to obtain the Markov model after training as the Markov model corresponding to the product type.

In one embodiment, the processor, when executing the computer program, further performs the steps of: sequencing the candidate products according to the interest degrees represented by the interest information to obtain a candidate product sequence; and determining products with interest degrees meeting preset conditions from the candidate product sequence as the target products.

In one embodiment, the processor, when executing the computer program, further performs the steps of: and inputting the object characteristic information of the multiple dimensions into a trained label prediction model to obtain a predicted image label for the target object.

In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:

In one embodiment, the computer program when executed by the processor further performs the steps of: acquiring product information of each candidate product; determining interest information of the target object in each candidate product based on the predicted portrait label and the product information of each candidate product; and determining a target product from the candidate products according to the interest information of the target object to each candidate product.

In one embodiment, the computer program when executed by the processor further performs the steps of: for each candidate product, determining a product type of the candidate product; determining a target Markov model corresponding to the candidate product based on the product type; and inputting the product information of the candidate product and the predicted portrait label into a target Markov model corresponding to the candidate product to obtain the interest information of the target object to the candidate product.

In one embodiment, the computer program when executed by the processor further performs the steps of: for each product type, acquiring a sample product, a portrait label of a sample object and an interest label of the sample object for the sample product, wherein the sample product and the portrait label are the same as the product type; based on a Markov decision process, taking various prediction results of the sample object on the sample product as states and taking a behavior of recommending the sample object to the product as an action; and when the Markov model is transferred from one state to another state, determining feedback information aiming at the transfer process, and adjusting the model parameters of the Markov model to be trained according to the feedback information until a training termination condition is reached to obtain the Markov model after training as the Markov model corresponding to the product type.

In one embodiment, the computer program when executed by the processor further performs the steps of: sequencing the candidate products according to the interest degrees represented by the interest information to obtain a candidate product sequence; and determining products with interest degrees meeting preset conditions from the candidate product sequence as the target products.

In one embodiment, the computer program when executed by the processor further performs the steps of: and inputting the object characteristic information of the multiple dimensions into a trained label prediction model to obtain a predicted image label for the target object.

In one embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, performs the steps of:

In one embodiment, the computer program when executed by the processor further performs the steps of: sequencing the candidate products according to the interest degrees represented by the interest information to obtain a candidate product sequence; and determining a product with the interest degree meeting a preset condition from the candidate product sequence as the target product.

In one embodiment, the computer program when executed by the processor further performs the steps of: and inputting the object feature information of the multiple dimensions into a trained label prediction model to obtain a predicted image label for the target object.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include a Read-Only Memory (ROM), a magnetic tape, a floppy disk, a flash Memory, an optical Memory, a high-density embedded nonvolatile Memory, a resistive Random Access Memory (ReRAM), a Magnetic Random Access Memory (MRAM), a Ferroelectric Random Access Memory (FRAM), a Phase Change Memory (PCM), a graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others. The databases involved in the embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the various embodiments provided herein may be, without limitation, general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, or the like.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, and these are all within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for recommending products, the method comprising:

2. The method of claim 1, wherein determining a target product from a plurality of candidate products based on the predicted portrait label comprises:

acquiring product information of each candidate product;

3. The method of claim 2, wherein determining interest information of the target object in the respective candidate product based on the predicted portrait label and product information of the respective candidate product comprises:

4. The method of claim 3, further comprising a process of training the Markov model for each product type, the process of training the Markov model for each product type comprising:

5. The method of claim 2, wherein determining the target product from the plurality of candidate products according to the interest information of the target object in each candidate product comprises:

6. The method of claim 1, wherein determining the predicted image label for the target object based on the object feature information for the plurality of dimensions comprises:

7. A product recommendation device, the device comprising:

a determination module for determining a predicted image tag for the target object based on the object feature information of the plurality of dimensions;

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the product recommendation method of any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of a product recommendation method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, carries out the steps of the product recommendation method of any one of claims 1 to 6.