A product recommendation method, device and storage medium

Abstract

The embodiment of the present invention provides a product recommendation method, wherein the method comprises: upon receiving a product application request sent by a target user, obtaining an initial product set according to a preset model; filtering a target product from the initial product set according to the tag characteristics of the target user; and recommending the target product to the target user. The product recommendation method provided by the embodiment of the present invention can effectively recommend products according to user tag information.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a product recommendation method, device and storage medium. Background Art

[0002] In the marketing scenario of credit products, the traditional model is that the account manager makes matching recommendations based on the customer's basic information, loan purpose, and the characteristics of the available loan products.

[0003] With the development of online systems and data, as well as the continuous maturity of big data technology, the system has labeled users with characteristic tags by recording their basic information and tracking their browsing behavior. The recommendation method based on user tags has also become the mainstream marketing method in recent years. Tags, as a kind of categorizable and independent characteristic value attribute, are widely used in the abstract classification and generalization of a certain feature of a certain group or object. Currently, the commonly collected tags are divided into two categories: user attribute tags and user behavior tags. However, both user attribute tags and user behavior tags are only an objective record of the user's behavior. Although they can reflect the user's habits to a certain extent, they still lack guiding value when analyzing user behavior in real time. Summary of the invention

[0004] The embodiments of the present invention provide a product recommendation method, device and storage medium, which can more effectively recommend products based on user tag information.

[0005] An embodiment of the present invention provides a product recommendation method, which includes: upon receiving a product application request sent by a target user, obtaining an initial product set according to a preset model; screening out a target product from the initial product set according to the label characteristics of the target user; and recommending the target product to the target user.

[0006] Optionally, filtering out the target product from the initial product set according to the label characteristics of the target user includes: filtering out the target product from the initial product set according to the attribute label, product preference label and repayment ability label of the target user.

[0007] Optionally, before obtaining the initial product set according to the preset model, the method also includes: obtaining feature information of the preset product, and using the feature information as the feature of the preset model, wherein the feature information includes at least one of the following: product loan amount, product loan term; obtaining product usage information of preset users, and using the product usage information as a training parameter of the preset model, wherein the product usage information includes at least one of the following: product calculation information, product application information, user repayment information on schedule, user repayment collection information; and using the product usage information to train the preset model.

[0008] Optionally, after obtaining the product usage information of a preset user, the method further includes: when the product usage information includes product application information, marking the product usage information as first data; when the product usage information includes product calculation information, marking the product usage information as second data.

[0009] Optionally, after marking the product usage information as a second parameter, the method further includes: inputting the second data into a first model to obtain a classification probability; and when the classification probability is greater than a preset threshold, updating the label of the second data to the first data.

[0010] Optionally, using the product usage information to train the preset model includes: inputting the first data into a second model for training to obtain a first parameter; and using the second data to process the first parameter to implement training of the preset model.

[0011] Optionally, obtaining the product usage information of a preset user includes: obtaining the product usage information sent by the preset user, and synchronizing the product usage information to an asynchronous message queue.

[0012] An embodiment of the present invention provides a product recommendation device, comprising an acquisition module, a screening module and a recommendation module, wherein the acquisition module is used to acquire an initial product set according to a preset model when receiving a product application request sent by a target user; the screening module is used to screen out a target product from the initial product set according to the label characteristics of the target user; and the recommendation module is used to recommend the target product to the target user.

[0013] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any one of the methods in the embodiments of the present invention is implemented.

[0014] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the processor implements any one of the methods in the embodiments of the present invention by executing a program in the memory.

[0015] The product recommendation method provided by the embodiment of the present invention can more effectively recommend products based on user tag information. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a product recommendation method according to an embodiment of the present invention;

[0017] Figure 2 is a schematic diagram of a data collection method according to an embodiment of the present invention;

[0018] Figure 3 is a flow chart of a method for processing user data labels based on a KNN model according to an embodiment of the present invention;

[0019] Figure 4 is a flow chart of a model training method according to an embodiment of the present invention;

[0020] Figure 5 is a framework diagram of a product recommendation device according to an embodiment of the present invention;

[0021] Figure 6 The present invention is a flowchart of a product strategy recommendation method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present application more clear, the embodiments of the present application will be described in detail with reference to the accompanying drawings. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other arbitrarily without conflict.

[0023] In addition, in the embodiments of the present invention, words such as "optional" or "exemplary" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "optional" or "exemplary" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "optional" or "exemplary" is intended to present related concepts in a specific way.

[0024] The embodiment of the present invention provides a product recommendation method. Figure 1 is a flow chart of a product recommendation method according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0025] S102, when receiving a product application request sent by a target user, obtaining an initial product set according to a preset model;

[0026] S104, filtering out target products from the initial product set according to the tag characteristics of the target user;

[0027] S106: recommend the target product to the target user.

[0028] The preset model is obtained by pre-training, and a quasi-predictive behavior model that approximates expert understanding is constructed. When the target user sends a product application request in the future, the preset model can recommend products based on the target user's label characteristics.

[0029] Optionally, the target product is screened out from the initial product set according to the label characteristics of the target user, including: screening out the target product from the initial product set according to the attribute label, product preference label and repayment ability label of the target user.

[0030] Among them, user attribute tags may include user feature tags such as user age, user occupation, user gender, etc. In addition, factors such as user product preferences, user loan product repayment ability, and quasi-predictive behavior evaluation can be combined with loan product characteristics to perform matching and screening processing, and calculate the current best product recommendation strategy set to improve the suitability of subsequent recommended products.

[0031] Optionally, before obtaining the initial product set according to the preset model, the method also includes: obtaining feature information of the preset product, and using the feature information as a feature of the preset model, wherein the feature information includes at least one of the following: product loan amount, product loan term; obtaining product usage information of preset users, and using the product usage information as a training parameter of the preset model, wherein the product usage information includes at least one of the following: product calculation information, product application information, user repayment information on schedule, user repayment collection information; and using the product usage information to train the preset model.

[0032] Before model training, user data needs to be collected first. Figure 2 FIG. 1 is a schematic diagram of a data collection method according to an embodiment of the present invention. Figure 2As shown, a link is embedded in the front-end loan product details page. When the user clicks it, the current product label count increases by one. When the user initiates a loan trial calculation, the credit system calculates the trial calculation result and the repayment plan, and synchronizes a trial calculation information to the asynchronous message queue. When the user initiates a loan application, the credit business system synchronizes to the asynchronous message queue. After the user initiates the repayment on schedule, the credit background system receives the user's repayment record and synchronizes the current repayment message to the asynchronous message queue. If the account manager initiates repayment collection for the user, the credit background system synchronizes the collection message to the asynchronous message queue. Here, by adding a user behavior prediction scenario, that is, supplementing the tracking of the trial calculation record in the personal loan calculator module, and adding tracking and marking of the relationship between the user's repayment execution record and the collection record, the data source is further enriched.

[0033] After the analysis system receives the asynchronous message, it first converts and stores the attributes in the message element into model tags that have been set in the system. This part of the data is temporarily stored as the original data.

[0034] For each repayment data of the loan agreement, if there is a collection record before the current repayment, a "passive" label will be added to the current user, enriching the label type and increasing the dimension of repayment execution ability rating.

[0035] Optionally, after obtaining the product usage information of a preset user, it also includes: when the product usage information includes product application information, marking the product usage information as first data; when the product usage information does not include product application information, marking the product usage information as second data.

[0036] Optionally, after marking the product usage information as a second parameter, the method further includes: inputting the second data into a first model to obtain a classification probability; and when the classification probability is greater than a preset threshold, updating the label of the second data to the first data.

[0037] Optionally, the preset model is trained using the product usage information, including: inputting the first data into the second model for training to obtain a first parameter; and using the second data to process the first parameter to achieve training of the preset model.

[0038] Before making product recommendations for target users, the preset model needs to be trained based on the collected user data. The method for training the model is as follows:

[0039] First, abstract product attributes into categories and specific labels under categories, such as product cycle labels: short-term, medium-term, medium-long term, long-term; credit limit labels: small amount, large amount, etc. Introduce pre-established evaluable labels, including user attribute labels, user credit product preference labels, and user selected product corresponding repayment execution ability rating labels. The user attribute label group includes user feature labels such as user age, user occupation, and user gender. Then, the loan amount, term, planned monthly repayment and other indicators in the loan calculator are used as features for the training model. Since there are users who have not actually applied for a loan, the data of these users can be used as an unlabeled learning feature group (i.e., the second data mentioned above); users who have actually applied for a loan can be used as a labeled learning feature group (i.e., the first data mentioned above). Figure 3 This is a flow chart of a method for user data label processing based on a KNN model according to an embodiment of the present invention. First, an initial K-nearest neighbor model (KNN) is trained based on a labeled data group. A relatively high initial classification probability threshold Th is set, and then the unlabeled data is passed through a KNN classifier (i.e., the first model mentioned above) to obtain a result label and classification probability, and the classification probability is compared with Th. If it exceeds Th, the label of the current data group is updated.

[0040] Second, the model is trained using user data.

[0041] Figure 4 is a flow chart of a model training method according to an embodiment of the present invention. The labeled data is brought into the fully supervised learning generation model to estimate the prior probability, mean and covariance of the labeled data, that is, P(C i ), μ i And Σ, and then get the probability value P(c i |x).

[0042] Formula 1:

[0043]

[0044] Among them, n represents the current calculation subscript, m represents the number of sample types, and c i represents category i, and x represents training samples.

[0045] Formula 2:

[0046]

[0047]

[0048] Among them, N represents the total number of samples, Ni represents the number of samples belonging to category Ci, and X u represents unlabeled data, X rRepresents samples labeled as category Ci.

[0049] Then the semi-supervised model is introduced to initialize the prior probability, mean and covariance of the labeled data, that is, P(C i ), μ i and Σ, initialize θ={P(C 1 ),……,μ 1 ,……,Σ}.

[0050] Calculate the posterior probability of unlabeled data: P θ (c i |x u ), and then use the posterior probability to update the prior probability and mean of Formula 2, and then calculate P(C i ) and then bring it into P θ (c i |x u ) calculation until the parameters converge.

[0051] By continuously training the semi-supervised model, the aforementioned preset model is finally obtained.

[0052] Optionally, obtaining the product usage information of a preset user includes: obtaining the product usage information sent by the preset user, and synchronizing the product usage information to an asynchronous message queue.

[0053] The user's trial calculation, browsing, application and other behavioral data are collected through asynchronous message queues and separated from the serial processing logic of the business system, thereby improving the efficiency of process processing.

[0054] The embodiment of the present invention also provides a product recommendation device. Figure 5 is a framework diagram of a product recommendation device according to an embodiment of the present invention. Figure 5 As shown, the device includes an acquisition module 52, a screening module 54 and a recommendation module 56, wherein:

[0055] The acquisition module 52 is used to acquire an initial product set according to a preset model when receiving a product application request sent by a target user;

[0056] The screening module 54 is used to screen out target products from the initial product set according to the tag characteristics of the target user;

[0057] The recommendation module 56 is used to recommend the target product to the target user.

[0058] Optionally, the screening module 54 is specifically configured to screen out the target product from the initial product set according to the attribute label, product preference label and repayment ability label of the target user.

[0059] Optionally, the device also includes a first acquisition module, a second acquisition module and a training module, wherein the first acquisition module is used to acquire feature information of a preset product, and use the feature information as a feature of the preset model, wherein the feature information includes at least one of the following: product loan amount, product loan term; the second acquisition module is used to acquire product usage information of a preset user, and use the product usage information as a training parameter of the preset model, wherein the product usage information includes at least one of the following: product calculation information, product application information, user repayment information on schedule, user repayment collection information; the training module is used to train the preset model using the product usage information.

[0060] Optionally, the device also includes a marking module, which is used to mark the product usage information as first data when the product usage information includes product application information; and mark the product usage information as second data when the product usage information includes product calculation information.

[0061] Optionally, the device also includes an updating module, which is used to input the second data into the first model to obtain a classification probability; when the classification probability is greater than a preset threshold, the label of the second data is updated to the first data.

[0062] Optionally, the training module is specifically used to input the first data into the second model for training to obtain the first parameter; and use the second data to process the first parameter to achieve training of the preset model.

[0063] Optionally, the second acquisition module is specifically used to acquire product usage information sent by a preset user, and synchronize the product usage information to an asynchronous message queue.

[0064] The following specific scenarios illustrate the embodiments of the present invention.

[0065] Marketing activity scenario in online loan system. Generate original model label data by collecting user browsing history, loan application product types, repayment records and collection records. First, train the original data received from the asynchronous message queue in the offline analysis system, and then classify some unlabeled data through the KNN nearest neighbor model, with the threshold set to 0.75; then bring the labeled data into the fully supervised learning generation model to obtain the expected value, covariance and other data of each label, as well as the probability value of the feature data belonging to a specific label; continuously train through the semi-supervised model, complete the calculation, and correct the model.

[0066] Figure 6 FIG. 1 is a flow chart of a product strategy recommendation method according to an embodiment of the present invention. Figure 6As shown in the figure, when the user enters the loan calculator page for trial calculation, the user input parameters are extracted and brought into the above semi-supervised model, and the product label group is obtained according to the calculated probability. According to the product label, the list of all products with label features PL0 is selected. Then, the product set PL1 is initially screened out according to the user attribute label, and PL0 and PL1 are taken as the union to obtain the product recommendation strategy set PL2. Again, according to the user's credit product preference, the list of available products PL3 is obtained, and PL2 and PL3 are taken as the union to obtain PL4. According to the repayment execution ability assessment results, the recommended product PL5 is screened out, and the intersection with PL4 is obtained to obtain PL6.

[0067] An embodiment of the present invention further provides an electronic device, which includes a processor and a memory; the number of processors in the electronic device may be one or more, and the memory, as a computer-readable storage medium, may be used to store computer executable programs. The processor executes various functional applications and data processing of the electronic device by running software programs and instructions stored in the memory, that is, implementing the method in any of the above embodiments.

[0068] An embodiment of the present application also provides a storage medium containing computer executable instructions, which implement the method in any of the above embodiments when executed by a computer processor.

[0069] Optionally, the processor executes the instruction to implement a product recommendation method, the method comprising:

[0070] S1, upon receiving a product application request from a target user, obtaining an initial product set according to a preset model;

[0071] S2, filtering out target products from the initial product set according to the tag characteristics of the target user;

[0072] S3, recommending the target product to the target user.

[0073] The above are merely exemplary embodiments of the present application and are not intended to limit the protection scope of the present application.

[0074] In general, various embodiments of the present application can be implemented in hardware or dedicated circuits, software, logic or any combination thereof. For example, some aspects can be implemented in hardware, while other aspects can be implemented in firmware or software that can be executed by a controller, microprocessor or other computing device, although the present application is not limited thereto.

[0075] Embodiments of the present application may be implemented by executing computer program instructions by a data processor of a mobile device, for example in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0076] The block diagram of any logic flow in the accompanying drawings of the present application can represent program steps, or can represent interconnected logic circuits, modules and functions, or can represent a combination of program steps and logic circuits, modules and functions. The computer program can be stored on a memory. The memory can have any type suitable for the local technical environment and can be implemented using any suitable data storage technology, such as but not limited to read-only memory (ROM), random access memory (RAM), optical memory device and system (digital versatile disc DVD or CD disc), etc. Computer-readable media may include non-transient storage media. The data processor can be any type suitable for the local technical environment, such as but not limited to a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (FGPAs) and a processor based on a multi-core processor architecture.

[0077] By way of exemplary and non-limiting examples, a detailed description of exemplary embodiments of the present application has been provided above. However, various modifications and adjustments to the above embodiments will be apparent to those skilled in the art when considered in conjunction with the accompanying drawings and claims, but will not depart from the scope of the present invention. Therefore, the proper scope of the present invention will be determined according to the claims.

Claims

1. A product recommendation method, wherein: include: In the case where the product usage information includes product application information, marking the product usage information as first data; The first data is labeled data; In the case where the product usage information includes product trial calculation information, the product usage information is marked as second data; the second data is unlabeled data; Inputting the second data into the first model to obtain classification probability; When the classification probability is greater than a preset threshold, updating the label of the second data to the first data; Upon receiving a product application request from a target user, obtaining an initial product set according to a preset model; Filtering target products from the initial product set according to the tag characteristics of the target user; recommending the target product to the target user; The training process of the preset model includes: Abstract product attributes into categories and specific labels under categories; Introducing pre-defined evaluable labels, including user attribute labels, user credit product preference labels, and repayment execution ability rating labels corresponding to user selected products; Bring the labeled data into the fully supervised learning generation model to estimate the prior probability, mean and covariance of the labeled data, that is, P(c i ), μ i And Σ, and then get the probability value P(c i |x); Formula 1: Among them, n represents the current calculation subscript, m represents the number of sample types, and c i represents category i, and x represents training samples; Formula 2: Among them, N represents the total number of samples, N i Represents category c i The number of samples, x u represents unlabeled data, x r Represents the label as category c i Samples of Initialize the prior probability, mean and covariance of the labeled data, that is, P(c i ), μ i and Σ, initialize θ={P(c1),…,μ1,…,Σ}; Calculate the posterior probability P of unlabeled data θ (c i |x u ), and then use the posterior probability to update the prior probability and mean of Formula 2, and then calculate P(c i ) and then bring it into P θ (c i |x u ) calculation until the parameters converge.

2. The method of claim 1, wherein: The step of selecting a target product from the initial product set according to the tag characteristics of the target user includes: The target product is screened out from the initial product set according to the attribute label, product preference label and repayment ability label of the target user.

3. The method of claim 1, wherein: Before getting the initial set of products based on the preset model, it also includes: Acquire characteristic information of a preset product, and use the characteristic information as a characteristic of the preset model, wherein the characteristic information includes at least one of the following: product loan amount and product loan term; Acquire product usage information of a preset user, and use the product usage information as a training parameter of the preset model, wherein the product usage information includes at least one of the following: product trial calculation information, product application information, user repayment information on schedule, and user repayment collection information; The preset model is trained using the product usage information.

4. The method of claim 1, wherein: The using the product usage information to train the preset model includes: Inputting the first data into a second model for training to obtain first parameters; The first parameter is processed using the second data to implement training of the preset model.

5. The method of claim 3, wherein: The obtaining of product usage information of a preset user includes: Acquire product usage information sent by a preset user, and synchronize the product usage information to an asynchronous message queue.

6. A product recommendation device, comprising an acquisition module, a screening module and a recommendation module, wherein: The acquisition module is used to acquire an initial product set according to a preset model when receiving a product application request sent by a target user; The screening module is used to screen out target products from the initial product set according to the tag characteristics of the target user; The recommendation module is used to recommend the target product to the target user; Also includes training modules; The training module is used to train the preset model, and the training process of the preset model includes: Abstract product attributes into categories and specific labels under categories; Introducing pre-defined evaluable labels, including user attribute labels, user credit product preference labels, and repayment execution ability rating labels corresponding to user selected products; Bring the labeled data into the fully supervised learning generation model to estimate the prior probability, mean and covariance of the labeled data, that is, P(c i ), μ i And Σ, and then get the probability value P(c i |x); Formula 1: Among them, n represents the current calculation subscript, m represents the number of sample types, and c i represents category i, and x represents training samples; Formula 2: Among them, N represents the total number of samples, N i Represents category c i The number of samples, x u represents unlabeled data, x r Represents the label as category c i Samples of Initialize the prior probability, mean and covariance of the labeled data, that is, P(c i ), μ i and Σ, initialize θ={P(c1),…,μ1,…,Σ}; Calculate the posterior probability P of unlabeled data θ (c i |x u ), and then use the posterior probability to update the prior probability and mean of Formula 2, and then calculate P(c i ) and then bring it into P θ (c i |x u ) calculation until the parameters converge; It also includes a marking module, which is used to mark the product usage information as first data when the product usage information includes product application information, and the first data is labeled data; when the product usage information includes product trial calculation information, the product usage information is marked as second data, and the second data is self-labeled data; It also includes an updating module, which is used to input the second data into the first model to obtain the classification probability; when the classification probability is greater than a preset threshold, the label of the second data is updated to the first data.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

8. An electronic device comprising a memory and a processor, characterized in that: The processor implements the method according to any one of claims 1 to 5 by executing the computer program in the memory.

Images