Method for training commodity recommendation model, commodity recommendation method and related device

Abstract

The embodiment of the application provides a kind of training method of commodity recommendation model, commodity recommendation method and related device, specifically include: obtaining real commodity recommendation session;Based on the real commodity recommendation session generates incremental commodity recommendation session, and generates the commodity recommendation logic corresponding to the incremental commodity recommendation session;Wherein, the semantic similarity between the incremental commodity recommendation session and the real commodity recommendation session meets expectation;The commodity recommendation logic is used to elucidate the recommendation reason for the commodity in the incremental commodity recommendation session according to logical derivation mode;Based on the real commodity recommendation session, the incremental commodity recommendation session and the commodity recommendation logic, train commodity recommendation model. Implement the embodiment of the application can improve the generalization ability of commodity recommendation model in the case where the amount of data used for model training is limited.

Description

Technical Field

[0001] The embodiments described in this application relate to the field of artificial intelligence technology, and in particular to a training method for a product recommendation model, a product recommendation method, and related apparatus. Background Technology

[0002] To meet diverse user needs, e-commerce platforms offer product recommendation services. When users are unsure which product to choose, the service can be triggered to recommend products that meet their needs.

[0003] In traditional product recommendation methods, when a recommendation request is detected, a pre-trained recommendation model is invoked to generate product recommendations. The generalization ability of the recommendation model is crucial to its accuracy; generalization ability refers to whether the model can accurately recommend products to users who have not learned from existing samples. To enhance the generalization ability of the recommendation model, it is typically necessary to train it using a large amount of real-world product recommendation data.

[0004] However, when the amount of real product recommendation data is limited, the trained recommendation model suffers from insufficient generalization ability. Summary of the Invention

[0005] In view of this, multiple embodiments of this application aim to provide a training method and related apparatus for a product recommendation model, which can improve the generalization ability of the product recommendation model when the amount of data used for model training is limited.

[0006] One embodiment of this application provides a method for training a product recommendation model. The method includes: acquiring a real product recommendation session; generating an incremental product recommendation session based on the real product recommendation session, and generating product recommendation logic corresponding to the incremental product recommendation session; wherein the semantic similarity between the incremental product recommendation session and the real product recommendation session meets expectations; the product recommendation logic is used to explain the reasons for recommending products in the incremental product recommendation session according to logical deduction; and training a product recommendation model based on the real product recommendation session, the incremental product recommendation session, and the product recommendation logic.

[0007] One embodiment of this application also provides a product recommendation method, the method comprising: upon receiving consultation information, invoking a product recommendation model according to one embodiment of this application, so that the product recommendation model generates product recommendation information and product recommendation logic corresponding to the consultation information; wherein, the product recommendation logic is used to explain the reasons for recommending products in the product recommendation information according to logical deduction.

[0008] One embodiment of this application also provides a training apparatus for a product recommendation model. The apparatus includes: a session acquisition module for acquiring real product recommendation sessions; an incremental data generation module for generating incremental product recommendation sessions based on the real product recommendation sessions, and generating product recommendation logic corresponding to the incremental product recommendation sessions; wherein the semantic similarity between the incremental product recommendation sessions and the real product recommendation sessions meets expectations; the product recommendation logic is used to explain the reasons for recommending products in the incremental product recommendation sessions according to logical deduction; and a model training module for training a product recommendation model based on the real product recommendation sessions, the incremental product recommendation sessions, and the product recommendation logic.

[0009] One embodiment of this application also provides a product recommendation device, the device comprising: a product recommendation module, configured to, upon receiving consultation information, invoke the product recommendation model in the training device, so that the product recommendation model generates product recommendation information and product recommendation logic corresponding to the consultation information; wherein, the product recommendation logic is configured to explain the reasons for recommending products in the product recommendation information according to logical deduction.

[0010] One embodiment of this application also provides a computer device, the computer device including a memory and a processor, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the method as described above.

[0011] One embodiment of this application also provides a computer-readable storage medium storing at least one computer program that, when executed by a processor, can implement the method described above.

[0012] One embodiment of this application also provides a computer program product for implementing the method as described above.

[0013] In several embodiments provided in this application, incremental product recommendation sessions with semantically similar meanings can be generated based on real product recommendation sessions, along with corresponding product recommendation logic. Since the product recommendation logic explains the reasons for recommending products in the incremental product recommendation session, using real product recommendation sessions, incremental product recommendation sessions, and product recommendation logic as the basis for model training allows the product recommendation model to be sufficiently trained and learn the product recommendation logic even with a limited number of real product recommendation sessions, thereby improving the generalization ability of the product recommendation model. Attached Figure Description

[0014] Figure 1This is a schematic diagram of a system architecture for implementing a training method for a product recommendation model, provided as one embodiment of this application.

[0015] Figure 2 A flowchart illustrating a method for training a product recommendation model according to one embodiment of this application.

[0016] Figure 3 A schematic diagram of a module for training a product recommendation model provided in one embodiment of this application.

[0017] Figure 4 A schematic diagram of a product recommendation device provided in one embodiment of this application.

[0018] Figure 5 A schematic diagram of a computer device provided for one embodiment of this application. Detailed Implementation

[0019] The information to be retrieved in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0020] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0021] E-commerce platforms provide users with the ability to select goods online. However, in some cases, users have purchasing needs but lack a specific target. To meet diverse user needs, e-commerce platforms offer product recommendation functions. These functions can take the form of intelligent customer service or shopping assistants, recommending products that meet the user's needs based on their input (e.g., which office chair is best for a small workspace) (e.g., a simple swivel chair from brand XX). The core of the product recommendation function is the recommendation model. The generalization ability of the recommendation model depends on the amount of training data; the larger the data volume, the easier it is to train a recommendation model with strong generalization ability. A recommendation model with strong generalization ability can also use learned recommendation logic to accurately understand the semantics of the inquiry and recommend products that match the inquiry when faced with unfamiliar information. However, when the amount of real-world product recommendation data is limited, the trained recommendation model may suffer from insufficient generalization ability.

[0022] The methods used by related technologies to improve the generalization ability of models lie in increasing the amount of training data. To obtain sufficient samples, these technologies employ at least one of the following methods to collect data: intelligent customer service repeatedly inquiring about user needs, requesting information from other platforms about user requests, and obtaining historical user search records, shopping records, or browsing records. Clearly, these methods require significant computational or communication resources to obtain enough samples to train the model and achieve the expected generalization ability.

[0023] Therefore, it is necessary to provide a training method for a product recommendation model. This method can generate semantically similar incremental product recommendation sessions based on real product recommendation sessions, along with corresponding product recommendation logic. Since the product recommendation logic explains the reasons for recommending products in the incremental product recommendation sessions, using real product recommendation sessions, incremental product recommendation sessions, and product recommendation logic as the basis for model training allows the product recommendation model to be sufficiently trained and learn the product recommendation logic even with a limited number of real product recommendation sessions, thereby improving the generalization ability of the product recommendation model.

[0024] Please see Figure 1 In several embodiments provided in this application, the training method for the product recommendation model can be applied to a training device for the product recommendation model. The training device for the product recommendation model can be an electronic device with certain computing power and network access capabilities. This electronic device can be a desktop computer, laptop computer, tablet computer, or a server. This electronic device can connect to the server via a network. The server can be a distributed server, including multiple processors, memory, network communication modules, etc., working together to achieve various functions. Alternatively, the server can also be a server cluster formed by several servers, possessing higher computing and data processing capabilities. With the development of science and technology, the server can also be implemented using new technological means, such as a new type of "server" based on quantum computing. Of course, in some embodiments, the training device for the product recommendation model can also be a program module running in an electronic device.

[0025] Specifically, the server receives real product recommendation sessions sent by electronic devices, generates incremental product recommendation sessions from these sessions, and generates product recommendation logic corresponding to these incremental sessions. Based on the real product recommendation sessions, the incremental sessions, and the product recommendation logic, a product recommendation model is trained. Furthermore, when an electronic device receives inquiry information, it sends the inquiry information to the server, causing the server to invoke the product recommendation model, which then generates product recommendation information and logic corresponding to the inquiry information.

[0026] In some implementations, the electronic device / server is used to acquire real product recommendation sessions, generate incremental product recommendation sessions based on these sessions, and generate product recommendation logic corresponding to the incremental sessions. Based on the real product recommendation sessions, the incremental sessions, and the product recommendation logic, a product recommendation model is trained. Furthermore, the electronic device / server can invoke the product recommendation model upon receiving inquiry information, enabling the model to generate product recommendation information and logic corresponding to the inquiry information.

[0027] One embodiment of this application provides an example of an application scenario for a training method for a product recommendation model. This training method can be applied to scenarios where intelligent customer service recommends products.

[0028] In a conversation where real product recommendations are obtained: {User: "What mop cleans the floor best?"; Smart Customer Service: "Have you ever used a floor scrubber?"; User: "I've used the AA brand, but the water tank is too small, and I have to refill it several times after each mop, which is very inconvenient"; Smart Customer Service: "We recommend the BB brand floor scrubber, which has a large 1L water tank, or the CC brand floor scrubber, which has an external backpack water tank in addition to its own water tank, allowing it to carry a total of 2L of water."}

[0029] Furthermore, incremental product recommendation sessions can be generated based on real product recommendation sessions {User: "Recommend a floor scrubber. I have experience using floor scrubbers and need one with a large water capacity"; Intelligent Customer Service: "We recommend purchasing the BB brand floor scrubber with a 1L water tank, and the CC brand floor scrubber with both a built-in water tank and an external backpack water tank, which can carry 2L of water"}. Additionally, product recommendation logic corresponding to the incremental product recommendation sessions is generated to demonstrate the logical relationship between the user's input query "Recommend a floor scrubber. I have experience using floor scrubbers and need one with a large water capacity" and the final output product recommendation information "We recommend purchasing the BB brand floor scrubber with a 1L water tank, and the CC brand floor scrubber with both a built-in water tank and an external backpack water tank, which can carry 2L of water".

[0030] By training a product recommendation model based on real product recommendation sessions, incremental product recommendation sessions, and product recommendation logic, a product recommendation model that meets the expected generalization ability can be obtained.

[0031] The product recommendation model is deployed in the intelligent customer service system. When the system receives a user's inquiry, "Which vacuum cleaner is good and has many replacement heads?", the intelligent customer service system calls the product recommendation model. The model analyzes the user's need for a vacuum cleaner with multiple replacement heads based on the learned product recommendation logic. Based on this, it retrieves the matching product "SS brand multi-head vacuum cleaner" from the product description information and pushes the "SS brand multi-head vacuum cleaner" to the user to meet their needs.

[0032] Please see Figure 2 One embodiment of this application provides a method for training a product recommendation model. This method can be applied to a training device for a product recommendation model. The method may include the following steps.

[0033] Step S110: Obtain a real product recommendation session.

[0034] Step S120: Generate an incremental product recommendation session based on the real product recommendation session, and generate product recommendation logic corresponding to the incremental product recommendation session; wherein, the semantic similarity between the incremental product recommendation session and the real product recommendation session meets expectations; the product recommendation logic is used to explain the reasons for recommending products in the incremental product recommendation session according to logical deduction.

[0035] Step S130: Train a product recommendation model based on the real product recommendation session, the incremental product recommendation session, and the product recommendation logic.

[0036] In this embodiment, a real product recommendation session refers to the actual interaction record between the user and the intelligent customer service / human customer service. A real product recommendation session includes consultation information characterizing the user's needs (e.g., "I have a stuffy nose and headache recently," "What kind of mop cleans best?") and product recommendation information to meet those needs (e.g., "Recommend using medicine A and medicine B together," "Recommend trying a new floor scrubber"). In some embodiments, a real product recommendation session may also include communication information for uncovering user needs (e.g., "Considering it's springtime and pollen levels are high, do you have a history of pollen allergies causing a stuffy nose and headache?" "What model of mop do you currently use? Have you used a floor scrubber?"). The training device for the product recommendation model can obtain real product recommendation sessions from at least one e-commerce platform via a data interface or database connection. These real product recommendation sessions can serve as the basis for model training and as the basis for generating incremental product recommendation sessions.

[0037] In some implementations, the training device for the product recommendation model can further preprocess the real product recommendation sessions after acquiring them and before generating incremental product recommendation sessions based on them. This preprocessing aims to standardize the format of each real product recommendation session or remove noise from the real product recommendation sessions. Noise refers to information considered invalid in the real product recommendation sessions (e.g., "Thank you for your affirmation," "Are you still there?").

[0038] In this embodiment, the training device for the product recommendation model can invoke a pre-trained large language model based on real product recommendation conversations. This allows the pre-trained large language model to rewrite information from the real product recommendation conversation (e.g., "Ten years of rhinitis, requires long-term treatment") to generate incremental product recommendation conversations with expected semantic similarity to the real conversations (e.g., containing "Long-term rhinitis, are there any suitable products for continuous use"). Alternatively, the pre-trained large language model can perform semantic understanding of the real product recommendation conversations and generate incremental product recommendation conversations similar to the real conversations based on product knowledge learned online. The incremental product recommendation conversations are hypothetical, but their form is the same as the real conversations. That is, the incremental conversations can also contain any of the following: consultation information representing user needs, product recommendation information to meet user needs, or communication information to uncover user needs.

[0039] In some implementations, the training device for the product recommendation model can also call a semantic similarity model to verify the semantic similarity between the real product recommendation session and the incremental product recommendation session, so as to ensure that the semantic similarity between the real product recommendation session and the incremental product recommendation session meets the expectations. The expectations here may include a preset similarity (e.g., 0.8). When the semantic similarity is greater than or equal to the preset similarity, it can be considered that the semantic similarity meets the expectations. Here, semantic similarity can be understood as the similarity between the full text of the real product recommendation session and the incremental product recommendation session, or it can be understood as the thematic similarity between the real product recommendation session and the incremental product recommendation session. The thematic similarity is used to reflect the core topic discussed in the session.

[0040] In some implementations, the training device for the product recommendation model can generate multiple incremental product recommendation sessions based on a real product recommendation session to increase the number of samples used for model training. Alternatively, it can combine multiple real product recommendation sessions to generate an incremental product recommendation session that integrates user needs from multiple real product recommendation sessions to enrich the diversity of samples used for model training.

[0041] In this embodiment, to improve the usability of incremental product recommendation sessions as model training samples, the training device for the product recommendation model can also call a pre-trained logical reasoning model based on the incremental product recommendation sessions, so that the logical reasoning model generates product recommendation logic corresponding to the incremental product recommendation sessions. The product recommendation logic can be used to characterize the process by which the large model derives product recommendation information from the consultation information in the incremental product recommendation sessions. Optionally, the product recommendation logic can be in the form of text, which can be displayed in the user interface when triggered. The product recommendation logic explains the reasons for recommending products in the incremental product recommendation sessions according to logical deduction. Optionally, the logical deduction method includes a user needs analysis process (e.g., a user has chronic rhinitis and may need long-term treatment), a product knowledge retrieval process (e.g., finding a multi-course product set suitable for long-term rhinitis treatment), and a recommendation reason generation process (e.g., recommending product A because it includes a 3-month treatment course, suitable for long-term use). The product recommendation logic executes the above processes in the above order to obtain the recommendation reasons.

[0042] In this embodiment, the product recommendation logic serves as the model training sample. The model can learn how to extract user needs from incremental product recommendation sessions and how to use product knowledge to search for products in the product library that meet those needs. When real product recommendation sessions are limited, incremental product recommendation sessions can be generated as model training samples to increase the training load. To ensure the model possesses the expected generalization ability, the product recommendation logic corresponding to the incremental product recommendation sessions is also used as the model training sample. This allows the model to not only receive sufficient training but also learn the product recommendation logic from the training. Compared to related technologies that typically improve model generalization ability by training models with massive amounts of labeled data, this application provides corresponding product recommendation logic and incremental product recommendation sessions as model training samples based on real product recommendation sessions. This not only supplements the training load but also ensures that a product recommendation model with the expected generalization ability is obtained.

[0043] In some implementations, a pre-trained logical reasoning model can be invoked based on a real product recommendation session, so that the logical reasoning model generates product recommendation logic corresponding to the real product recommendation session, which serves as a training sample for the model to further improve the model's generalization ability.

[0044] In this embodiment, the trained product recommendation model can be deployed in an intelligent customer service scenario. When a user inputs a consultation message (e.g., which type of contact lenses is better for dry eyes), the product recommendation model can be invoked based on this consultation message. Since the product recommendation model has learned the product recommendation logic, it can infer the user's needs corresponding to the consultation message and the products that meet those needs based on the product recommendation logic, and generate product recommendation information (e.g., recommend using XX brand contact lenses, which have high oxygen permeability and relatively low water content, and will not aggravate dry eye symptoms) and provide it to the user.

[0045] In this embodiment, incremental product recommendation sessions with semantically similar meanings can be generated based on real product recommendation sessions, along with corresponding product recommendation logic. Since the product recommendation logic explains the reasons for recommending products in the incremental product recommendation sessions, using real product recommendation sessions, incremental product recommendation sessions, and product recommendation logic as the basis for model training allows the product recommendation model to be sufficiently trained and learn the product recommendation logic even with a limited number of real product recommendation sessions. This ensures that the product recommendation model possesses the expected generalization ability.

[0046] In some implementations, the training device for the product recommendation model can extract product description information from product images; wherein, the product images include at least one of the following: product display images, product detail page images, product application scenario images, and product usage effect images; and the product recommendation model is trained based on the product recommendation logic, the product description information, the real product recommendation session, and the incremental product recommendation session.

[0047] In this embodiment, to improve the recommendation accuracy of the product recommendation model, the training device of the product recommendation model can extract product description information as input for model training. This allows the product recommendation model to base its recommendations on comprehensive product description information, thus improving the accuracy of product recommendations. Specifically, e-commerce platforms store a massive number of product images. Each product corresponds to at least one product image. These product images can be product display images (e.g., pharmaceutical packaging, actual images of medical devices) used to show the standardized appearance of the product, product detail page images (e.g., drug ingredient lists, screenshots of instruction manuals) used to represent the product's functions, product application scenario images (e.g., illustrations of users using nebulizers to treat rhinitis), or product usage effect images (e.g., comparisons of skin condition before and after using skincare products).

[0048] In this embodiment, the training device for the product recommendation model can preprocess the product images (e.g., denoising, adjusting resolution, etc.), and then call an optical character recognition model to extract text information from the product images as product description information (e.g., "suitable for chronic rhinitis"), or call a multimodal model to extract visual features of the images (e.g., the visual style of drug packaging, user behavior features in the scene image) as product description information. The product description information can characterize the product from one or more dimensions (e.g., efficacy, course of treatment, applicable population).

[0049] In some implementations, the training device for the product recommendation model can associate product description information with product SKUs to form a product knowledge base for the product recommendation model to call. For example, the product knowledge base includes: {Product A: Ingredients (cetirizine hydrochloride), Applicable population (patients with chronic rhinitis), Treatment course (3 months)}.

[0050] In this embodiment, the training device for the product recommendation model can input product recommendation logic, product description information, real product recommendation sessions, and incremental product recommendation sessions into the product recommendation model, so that the product recommendation model can learn the product recommendation logic, query products from the product description information that can meet the user's needs, and push the products to the user in the form of product recommendation information.

[0051] In some implementations, the training device for the product recommendation model can perform text recognition on a product image to obtain a text recognition result; based on the text recognition result, a multimodal evaluation model is invoked so that the multimodal evaluation model generates an evaluation result that characterizes the relationship between the text recognition result and the product image; if the evaluation result indicates that the product image contains the text recognition result, the text recognition result is identified as product description information.

[0052] In this embodiment, to ensure the accuracy of the product description information input to the model and to combat model illusion, the training device for the product recommendation model can use an OCR algorithm to extract text from the product image and call a pre-trained large model to perform semantic parsing on non-textual information (such as charts and symbols) in the product image, thereby obtaining text recognition results. Then, based on the text recognition results and the product image, a multimodal evaluation model is invoked to generate an evaluation result representing the relationship between the two. The evaluation result is used to determine whether the text recognition result truly reflects the content of the product image, and based on the evaluation result, high-confidence product description information can be selected from the text recognition results.

[0053] In some implementations, the training device for the product recommendation model can call a multimodal evaluation model based on the text recognition result, so as to use the multimodal evaluation model to map the text recognition result into text semantic features and the product image into image semantic features, generate a cross vector space representing the feature matching relationship between the text semantic features and the image semantic features, and combine the fusion result of the text semantic features, the image semantic features, and the cross vector space to generate an evaluation result of the relationship between the text recognition result and the product image.

[0054] In this embodiment, in order to obtain an accurate evaluation of the relationship between the text recognition result and the product image, the training device of the product recommendation model can call the multimodal evaluation model based on the text recognition result, so that the multimodal evaluation model maps / encodes the text recognition result into text semantic features, also known as text semantic space, through the feature mapping unit. For example, the text recognition result "treatment course 3 months" is mapped into the vector [0.2, 0.5, 0.8], and the product image is mapped into image semantic features, also known as image semantic space, through the feature mapping unit. For example, the product image is mapped / encoded into [0.3, 0.6, 0.7].

[0055] In this embodiment, the multimodal evaluation model can also utilize the cross-attention mechanism to calculate the feature matching relationship between text semantic features and image semantic features to generate a cross vector space. The cross vector space is used to characterize the correlation between the text recognition result and various features of the product image. For example, if the "treatment course of 3 months" in the text recognition result is highly correlated with the "treatment course description" area in the product image, then the cross vector space is used to highlight the strong correlation between the two in the form of feature vectors.

[0056] In this embodiment, the multimodal evaluation model can also obtain the fusion results of text semantic features, image semantic features, and cross-vector space. The fusion results are then used by a linear regression layer to generate an evaluation result that assesses the relationship between the text recognition result and the product image. The linear regression layer includes a self-attention layer to highlight key features, a pooling layer to compress feature dimensions while preserving core semantics, and a normalization layer to standardize the feature distribution. The evaluation result can be represented as a confidence score between 0 and 1, with a higher confidence score indicating a higher correlation between the text recognition result and the product image.

[0057] In some cases, when the evaluation result is greater than a preset value (e.g., 0.8), the product image is considered to contain text recognition results.

[0058] In some implementations, the training device for the product recommendation model can invoke a quality evaluation model based on the product recommendation logic and the incremental product recommendation session, so that the quality evaluation model generates a quality evaluation result based on the relevance between the context in the incremental product recommendation session and the product recommendation logic; wherein, the quality evaluation result is used to characterize the reasonableness of the product recommendation logic; and the quality evaluation result of the product recommendation logic used to train the product recommendation model meets expectations.

[0059] In this embodiment, to ensure the accuracy of the product recommendation logic used as input for model training, the training device for the product recommendation model can invoke a quality evaluation model. This model calculates the text similarity between the incremental product recommendation session and the product recommendation logic, verifies the coverage of the product recommendation logic with key elements representing user needs in the incremental product recommendation session, and checks the consistency between the product and its description information within the product recommendation logic. It then integrates text similarity, coverage, and consistency to obtain a quality evaluation result representing the product recommendation logic. The quality evaluation result can be expressed numerically; a higher value indicates a higher degree of rationality in the product recommendation logic. Furthermore, product recommendation logic that meets the expected quality evaluation result can be selected as input for model training to improve model training accuracy.

[0060] In some implementations, the training apparatus for the product recommendation model can train an initial model based on pre-labeled sample data and prompts to indicate the model's tasks, in order to obtain the product recommendation model to be trained.

[0061] In this embodiment, to warm up the model, an initial model can be driven to learn the tasks to be completed, thought processes, and response styles based on pre-labeled sample data and prompts indicating the model's tasks, thereby obtaining a product recommendation model. The pre-labeled sample data can come from real product recommendation conversations and includes consultation information, corresponding product recommendation information, and the recommendation logic between them. The initial model is an artificial intelligence model with natural language understanding capabilities, which can be understood as a base model. By driving the initial model through parameter iteration using pre-labeled sample data, a product recommendation model to be trained can be obtained. This model already possesses preliminary product recommendation capabilities. Training this model based on product recommendation logic, incremental product recommendation conversations, and real product recommendation conversations yields a product recommendation model with the expected generalization ability.

[0062] In some implementations, the training device for the product recommendation model can invoke the product recommendation model based on sample session information, so that the product recommendation model generates target product recommendation information and target product recommendation logic corresponding to the sample session information; based on the semantic similarity between the target product recommendation logic and the standard recommendation logic, the product recommendation model is driven to dynamically optimize the recommendation logic parameters; based on the consistency between the target product recommendation information and the standard recommended products, the product recommendation model is driven to dynamically optimize the product recommendation parameters.

[0063] In this embodiment, to improve model training accuracy, the product recommendation model can also undergo supervised training based on positive feedback. The sample session information can be a real product recommendation session or other product recommendation sessions; this embodiment does not limit this. The sample session information includes standard recommended products used as the basis for model tuning, and the sample session information corresponds to standard recommendation logic used as the basis for model tuning. Specifically, the training device for the product recommendation model can call the product recommendation model based on the sample session information, so that the product recommendation model generates target product recommendation information and target product recommendation logic corresponding to the sample session information. The product recommendation model includes product recommendation parameters related to product recommendation and recommendation logic parameters participating in logical reasoning. A regularized reward model (ORM) can be used to verify the consistency between the target product recommendation information and the standard recommended products. When they are consistent, reward information can be output to incentivize the product recommendation model to dynamically tune the product recommendation parameters. Furthermore, a process reward model (PRM) can be used to determine the similarity between the standard recommendation logic and the target product recommendation logic. When the similarity meets expectations, reward information can be output to incentivize the product recommendation model to dynamically tune the recommendation logic parameters.

[0064] In this embodiment, the collaboration between the Regular Reward Model (ORM) and the Process Reward Model (PRM) can make the product recommendation logic of the product recommendation model more reasonable and can make the matching degree between the products recommended by the product recommendation model and the user's needs higher.

[0065] The embodiments of this application also provide a product recommendation method, which includes: upon receiving consultation information, invoking a product recommendation model from one embodiment of this application, so that the product recommendation model generates product recommendation information and product recommendation logic corresponding to the consultation information; wherein, the product recommendation logic is used to explain the reasons for recommending products in the product recommendation information according to logical deduction.

[0066] The specific functions and effects of the training method for the product recommendation model in this embodiment can be explained by referring to other embodiments of this application, and will not be repeated here.

[0067] Please see Figure 3 This application also provides a training apparatus for a product recommendation model. The training apparatus for the product recommendation model may include: a session acquisition module for acquiring real product recommendation sessions; an incremental data generation module for generating incremental product recommendation sessions based on the real product recommendation sessions, and generating product recommendation logic corresponding to the incremental product recommendation sessions; wherein the semantic similarity between the incremental product recommendation sessions and the real product recommendation sessions meets expectations; the product recommendation logic is used to explain the reasons for recommending products in the incremental product recommendation sessions according to logical deduction; and a model training module for training a product recommendation model based on the real product recommendation sessions, the incremental product recommendation sessions, and the product recommendation logic.

[0068] In this embodiment, the specific functions and effects of the training device for the product recommendation model can be explained by referring to other embodiments of this application, and will not be repeated here.

[0069] Please see Figure 4 One embodiment of this application also provides a product recommendation device, the device comprising: a product recommendation module, configured to, upon receiving consultation information, invoke the product recommendation model in the training device, so that the product recommendation model generates product recommendation information and product recommendation logic corresponding to the consultation information; wherein, the product recommendation logic is configured to explain the reasons for recommending products in the product recommendation information according to logical deduction.

[0070] In this embodiment, the specific functions and effects of the training device for the product recommendation model can be explained by referring to other embodiments of this application, and will not be repeated here.

[0071] Please see Figure 5 This application also provides a computer device comprising: a memory and a processor, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the method described above.

[0072] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to implement the method as described above.

[0073] This application also provides a computer program product containing instructions that, when executed by a processor, implement the method as described above.

[0074] The user information or user account information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, etc.) involved in various embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws and regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0075] It is understood that the specific examples in this document are only intended to help those skilled in the art better understand the embodiments of this application, and are not intended to limit the scope of the invention.

[0076] It is understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0077] It is understood that the various implementation methods described in this application can be implemented individually or in combination, and the implementation methods in this application are not limited in this respect.

[0078] Unless otherwise stated, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0079] It is understood that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0080] It is understood that the memory in the embodiments of this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Specifically, non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0081] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the information to be retrieved. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0082] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the aforementioned method implementations, and will not be repeated here.

[0083] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0084] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0085] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0086] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the information to be retrieved in this application, essentially or in terms of its contribution to the prior art, or a portion of the information to be retrieved, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The above description is merely a specific embodiment of this application, but the scope of protection of this invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A training method for a product recommendation model, characterized in that, The method includes: Obtain real product recommendation sessions; An incremental product recommendation session is generated based on the real product recommendation session, and a product recommendation logic corresponding to the incremental product recommendation session is generated; wherein, the semantic similarity between the incremental product recommendation session and the real product recommendation session meets expectations; the product recommendation logic is used to explain the reasons for recommending products in the incremental product recommendation session according to a logical deduction method; the incremental product recommendation session is a fictional product recommendation session, including consultation information to represent user needs, product recommendation information to meet user needs, and communication information to mine user needs; the logical deduction method includes a user needs analysis process, a product knowledge retrieval process, and a recommendation reason generation process; A product recommendation model is trained based on the real product recommendation session, the incremental product recommendation session, and the product recommendation logic.

2. The method according to claim 1, characterized in that, The steps for training a product recommendation model based on the real product recommendation session, the incremental product recommendation session, and the product recommendation logic include: Extract product description information from product images; wherein, the product images include at least one of the following: product display images, product details page images, product application scenario images, and product usage effect images; A product recommendation model is trained based on the product recommendation logic, the product description information, the real product recommendation session, and the incremental product recommendation session.

3. The method according to claim 2, characterized in that, The steps for extracting product description information from product images include: Perform text recognition on product images to obtain text recognition results; Based on the text recognition result, a multimodal evaluation model is invoked so that the multimodal evaluation model generates an evaluation result that characterizes the relationship between the text recognition result and the product image; If the evaluation result indicates that the product image contains the text recognition result, the text recognition result is identified as product description information.

4. The method according to claim 3, characterized in that, The step of calling a multimodal evaluation model based on the text recognition result, so that the multimodal evaluation model generates an evaluation result characterizing the relationship between the text recognition result and the product image, includes: Based on the text recognition result, a multimodal evaluation model is invoked to map the text recognition result into text semantic features and the product image into image semantic features. A cross vector space representing the feature matching relationship between the text semantic features and the image semantic features is generated. Combining the fusion result of the text semantic features, the image semantic features, and the cross vector space, an evaluation result of the relationship between the text recognition result and the product image is generated.

5. The method according to claim 1, characterized in that, The method further includes: Based on the product recommendation logic and the incremental product recommendation session, a quality evaluation model is invoked, so that the quality evaluation model generates a quality evaluation result based on the relevance between the context in the incremental product recommendation session and the product recommendation logic; wherein, the quality evaluation result is used to characterize the reasonableness of the product recommendation logic; the quality evaluation result of the product recommendation logic used to train the product recommendation model meets expectations.

6. The method according to claim 1, characterized in that, The method further includes: The product recommendation model is invoked based on the sample session information, so that the product recommendation model generates target product recommendation information and target product recommendation logic corresponding to the sample session information; Based on the semantic similarity between the target product recommendation logic and the standard recommendation logic, the product recommendation model is driven to dynamically optimize the recommendation logic parameters. Based on the consistency between the target product recommendation information and the standard recommended products, the product recommendation model is driven to dynamically optimize the product recommendation parameters.

7. A product recommendation method, characterized in that, The method includes: Upon receiving consultation information, the product recommendation model obtained by the training method described in any one of claims 1 to 6 is invoked, so that the product recommendation model generates product recommendation information and product recommendation logic corresponding to the consultation information; wherein, the product recommendation logic is used to explain the reasons for recommending the products in the product recommendation information in a logical deduction manner.

8. A training device for a product recommendation model, characterized in that, The device includes: The session acquisition module is used to acquire real product recommendation sessions; An incremental data generation module is used to generate incremental product recommendation sessions based on the real product recommendation sessions, and to generate product recommendation logic corresponding to the incremental product recommendation sessions; wherein, the semantic similarity between the incremental product recommendation sessions and the real product recommendation sessions meets expectations; the product recommendation logic is used to explain the reasons for recommending products in the incremental product recommendation sessions according to logical deduction; the incremental product recommendation sessions are hypothetical product recommendation sessions, including consultation information to represent user needs, product recommendation information to meet user needs, and communication information to mine user needs; the logical deduction method includes a user needs analysis process, a product knowledge retrieval process, and a recommendation reason generation process; The model training module is used to train a product recommendation model based on the real product recommendation session, the incremental product recommendation session, and the product recommendation logic.

9. A product recommendation device, characterized in that, The device includes: The product recommendation module is used to, upon receiving consultation information, invoke the product recommendation model obtained by the training method described in any one of claims 1 to 6, so that the product recommendation model generates product recommendation information and product recommendation logic corresponding to the consultation information; wherein, the product recommendation logic is used to explain the reasons for recommending the products in the product recommendation information according to logical deduction.

10. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the method as described in any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, is capable of implementing the method as described in any one of claims 1 to 7.

12. A computer program product, characterized in that, The computer program product is used to implement the method as described in any one of claims 1 to 7.

Images