Commodity information recommendation method and apparatus, device, medium, and program product

Abstract

The present application relates to the field of data processing, and discloses a commodity information recommendation method and apparatus, a device, a medium, and a program product. The method comprises: training a recommendation model according to request propensity information of a sample user for a commodity, so as to obtain a commodity preference information set of the sample user as predicted by the recommendation model; iteratively training the recommendation model on the basis of the request propensity information, the commodity preference information set, and a large language model, until a performance parameter of the recommendation model meets a preset condition, and determining the recommendation model as a target recommendation model, the large language model being configured for extracting, from the commodity preference information set, sample information used for performing iterative training on the recommendation model; and using the target recommendation model to obtain commodity recommendation information according to the request propensity information of a target user for the commodity, and pushing the commodity recommendation information to the target user.

Description

Method, device, equipment, medium and program product for recommending commodity information Cross-reference to related applications This application claims priority to Chinese Patent Application No. 202411305219.2, filed September 18, 2024, entitled “Method, device, equipment, medium and program product for recommending commodity information,” the entire contents of which are incorporated herein by reference. TECHNICAL FIELD The present application belongs to the field of data processing, and particularly relates to a method, device, equipment, medium and program product for recommending commodity information. BACKGROUND With the continuous development of electronic information technology, users have increasingly high demands for information acquisition. In order to meet the demands of users, information service providers can analyze the behaviors and preferences of users to recommend personalized information to users and improve user experience. In the scenario of users applying for commodities, most users will only apply for a small number of commodity categories, and a large number of commodity categories will only be applied for by a small number of users. This power-law distribution between users and commodities makes it difficult for a recommendation model based on the behaviors and preferences of users to obtain collaborative signals between users and commodities, and to determine similar users who apply for similar commodities, so that the recommendation model is difficult to accurately recommend commodities to users. SUMMARY The embodiments of the present application provide a method, device, equipment, medium and program product for recommending commodity information, which can improve the accuracy of the recommendation model for recommending commodity information to users. In a first aspect, the embodiments of the present application provide a method for recommending commodity information, comprising: training a recommendation model according to application tendency information of sample users for commodities to obtain a set of preferred commodity information of sample users predicted by the recommendation model; iteratively training the recommendation model based on the application tendency information, the set of preferred commodity information and a large language model until a performance parameter of the recommendation model meets a preset condition, determining the recommendation model as a target recommendation model, and the large language model is used to extract sample information for iteratively training the recommendation model from the set of preferred commodity information; and obtaining recommended commodity information according to application tendency information of a measured user for commodities by using the target recommendation model and pushing the recommended commodity information to the measured user. In a second aspect, an embodiment of the present application provides a commodity information recommendation device, comprising: a prediction result acquisition module configured to train a recommendation model according to application tendency information of a sample user to commodities, to obtain a set of preferred commodity information of the sample user predicted by the recommendation model; an enhanced training module configured to iteratively train the recommendation model based on the application tendency information, the set of preferred commodity information, and a large language model, until a performance parameter of the recommendation model meets a preset condition, to determine the recommendation model as a target recommendation model, and the large language model is configured to extract sample information for iteratively training the recommendation model from the set of preferred commodity information; and an information pushing module configured to obtain recommended commodity information according to application tendency information of a measured user to commodities by using the target recommendation model, and to push the recommended commodity information to the measured user. In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory storing computer program instructions; and the processor implements the commodity information recommendation method of the first aspect when executing the computer program instructions. In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores computer program instructions, and the computer program instructions are executed by a processor to implement the commodity information recommendation method of the first aspect. In a fifth aspect, an embodiment of the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the commodity information recommendation method of the first aspect. The embodiments of the present application provide a commodity information recommendation method, device, equipment, medium, and program product. A recommendation model is trained according to application tendency information that can reflect an application tendency of a sample user to commodities, to obtain a set of preferred commodity information of the sample user predicted by the recommendation model. Sample information is further extracted from the set of preferred commodity information by using the inductive reasoning capability of a large language model based on the application tendency information to train the recommendation model, until a performance parameter of the recommendation model meets a preset condition, and the recommendation model is used as a target recommendation model to recommend commodity information for a measured user. In an iterative process of training the recommendation model by taking the output of the recommendation model as the input of the large language model and taking the output of the large language model as the sample information, the output of the large language model is taken as the sample information to constantly enhance the collaborative signal, to improve the problem of lacking the collaborative signal, to enhance the recommendation model, and to improve the accuracy of the recommendation model in recommending commodity information for a user. BRIEF DESCRIPTION OF DRAWINGS In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments of the present application. Those skilled in the art can also obtain other drawings according to these drawings without paying creative labor. FIG. 1 is a flowchart of a commodity information recommendation method provided by an embodiment of the present application; FIG. 2 is a flowchart of a method for recommending commodity information according to another embodiment of the present application; FIG. 3 is a schematic diagram of an example of a logic framework of a large language model enhancement recommendation algorithm based on a course learning paradigm according to an embodiment of the present application; FIG. 4 is a schematic diagram of an example of an embedded representation of user commodity interaction information according to an embodiment of the present application; FIG. 5 is a schematic diagram of a structure of a device for recommending commodity information according to an embodiment of the present application; FIG. 6 is a schematic diagram of a structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION The features and exemplary embodiments of various aspects of the present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are intended to explain, and not to limit, the present application. The present application can be implemented without some of the specific details of these specific embodiments. The following description of the embodiments is merely intended to provide a better understanding of the present application by showing examples of the present application. It should be noted that the acquisition, storage, use, processing, etc. of information, data in the embodiments of the present application are authorized by the user or relevant institution and comply with relevant regulations of national laws and regulations. With the continuous development of electronic information technology, users have increasingly high demands for information acquisition. In order to meet the needs of users, information service providers can analyze the behavior and preferences of users to recommend personalized information to users and improve user experience. In the scenario of user application for commodities, most users will only apply for a small number of commodity categories, and a large number of commodity categories will only be applied for by a small number of users. This power-law distribution between users and commodities makes the collaborative signal between users and commodities obtained by the recommendation model based on the behavior and preferences of users very sparse. Users who have not applied for commodities or commodities that have not been applied for do not have collaborative signals. Collaborative signals can reflect the similarity of the behavior of users applying for commodities. In the absence of collaborative signals, it is difficult to determine similar users applying for similar commodities, making it difficult for the recommendation model to accurately recommend commodities to users. The application provides a commodity information recommendation method, device, equipment, medium and program product, and a large model based on a course learning paradigm is used to enhance a recommendation model. In a model training process of the recommendation model, a large language model is used to extract commodity information that meets user preferences from commodity information output by the recommendation model as pseudo samples according to user and commodity interaction information or user portrait information, and the original samples and the pseudo samples are used to train the recommendation model, and the process is repeated until the recommendation model meets the requirements and is put into use. The pseudo samples can enhance the collaborative signal and enhance the recommendation model, thereby improving the accuracy of the recommendation model in recommending commodity information to users. The commodity information recommendation method, device, equipment, medium and program product provided by the application are described below. The application provides a commodity information recommendation method, device, equipment, medium and program product, and a large model based on a course learning paradigm is used to enhance a recommendation model. In a model training process of the recommendation model, a large language model is used to extract commodity information that meets user preferences from commodity information output by the recommendation model as pseudo samples according to user and commodity interaction information or user portrait information, and the original samples and the pseudo samples are used to train the recommendation model, and the process is repeated until the recommendation model meets the requirements and is put into use. The pseudo samples can enhance the collaborative signal and enhance the recommendation model, thereby improving the accuracy of the recommendation model in recommending commodity information to users. In step S101, the recommendation model is trained according to the sample user's application tendency information of the commodity to obtain a set of sample user's preferred commodity information predicted by the recommendation model. The sample user includes a user corresponding to a sample used in the model training process of the recommendation model. In the embodiment of the application, the commodity can include an entity commodity, a service, a business, etc. For example, the commodity can include common food, daily necessities, etc. The commodity can also include a credit card, a membership card, etc. The application of the commodity can be understood as purchase, handling, etc. For example, for an entity commodity, the application of the commodity is to purchase the commodity; for a service, a business, etc., the application of the commodity is to handle the service, handle the business. The application tendency information of the sample user to the commodity is used to represent the user's application tendency to the commodity, such as the user's tendency to apply for a certain commodity or a certain type of commodity, and the user's tendency not to apply for a certain commodity or a certain type of commodity. In some examples, the application tendency information can include, but is not limited to, user commodity interaction information and user portrait information. The user commodity interaction information represents the user's historical application commodity, that is, the user commodity interaction information can represent the commodity that interacts with the user for application, such as the commodity purchased by the user, the service and business handled by the user, etc. The user portrait information is a labeled information abstracted according to the user's social attributes, living habits, consumption behaviors, etc. According to the user portrait information, the user's preferences can be determined to a certain extent. The recommendation model is used to output commodity information recommended to a user. The recommendation model is trained by using the application tendency information, so that the recommendation model learns the tendency of the user to the commodity, and the recommendation model can predict the commodity information of the commodity to which the user tends to apply. According to the application tendency information of the sample user, the recommendation model can output a set of preferred commodity information of the sample user. The set of preferred commodity information includes commodity information of a commodity preferred by the user predicted by the recommendation model, and the number of commodity information in the set of preferred commodity information can be one or more than two, which is not limited herein. For example, if the original sample used to train the recommendation model, i.e., the application tendency information, is represented as wherein, is the original sample, is the positive sample, is the negative sample. The negative sample can be a sample constructed by using negative sampling. The set of preferred commodity information output by the recommendation model according to the application tendency information can be represented as wherein, is the set of preferred commodity information, is commodity information of a commodity considered by the recommendation model as the most preferred by the sample user, is commodity information of a commodity considered by the recommendation model as the second preferred by the sample user, is commodity information of a commodity considered by the recommendation model as the Cth preferred by the sample user, and C is a positive integer. That is, the set of preferred commodity information includes commodity information of the first C commodities arranged in descending order of preference of the sample user. In some examples, in the case that the sample user has a historical application commodity, i.e., the sample user has applied for a commodity, and has user commodity interaction information, the application tendency information includes the user commodity interaction information, and the recommendation model can include a model for recommending based on similarity of user behavior. For example, the recommendation model can include but is not limited to a collaborative filtering (CF) model, which predicts content that a user can be interested in by analyzing user behavior and preferences. The collaborative filtering model can specifically adopt a SimpleX model. In some examples, the application tendency information includes user portrait information, and the recommendation model includes a model for making a recommendation based on a probability prediction of the application of the commodity. For example, the recommendation model can include, but is not limited to, a Click-Through Rate Prediction (CTR) model. The CTR model predicts the probability of a certain content being selected by analyzing user behavior, historical interest data, and other related features. The CTR model can use, but is not limited to, the EulerNet model. In step S102, the recommendation model is iteratively trained based on the application tendency information, the set of preferred commodity information, and the large language model until the performance parameter of the recommendation model meets a preset condition, and the recommendation model is determined as a target recommendation model. The large language model is used to extract sample information for iteratively training the recommendation model from the preferred commodity information. The large language model can use the application tendency information as a basis to further extract commodity information from the set of preferred commodity information to construct pseudo samples for training the recommendation model using its summarization and reasoning capabilities. The steps of using the pseudo samples to retrain the recommendation model, the recommendation model outputting the preferred commodity information, and the large language model extracting pseudo samples from the preferred commodity information can be performed in a loop until the performance parameter of the recommendation model meets the preset condition. Retraining the recommendation model using the pseudo samples will correspondingly adjust the model parameters of the recommendation model to update the recommendation model. The preset condition includes a condition in which the recommendation model reaches a use standard. In some examples, the preset condition can include a difference between the performance parameter of the recommendation model after the current training and the performance parameter of the recommendation model after the last training being less than a preset threshold, i.e., the training of the recommendation model stops when the performance parameter of the trained recommendation model no longer continues to significantly improve. When the performance parameter of the recommendation model meets the preset condition, the current recommendation model is determined as the target recommendation model. The target recommendation model is a model that can be put into use. In step S103, the target recommendation model is used to obtain recommendation commodity information according to the application tendency information of the measured user to the commodity and push the recommendation commodity information to the measured user. The input information can be obtained according to the application tendency information of the measured user to the commodity, and the input information is input into the target recommendation model. The result information output by the target recommendation model is the recommendation commodity information. The recommendation commodity information includes information of the commodity recommended by the target recommendation model to the measured user. The recommendation commodity information can be pushed to the measured user. The way in which the recommendation commodity information is pushed to the measured user is not limited herein, for example, the recommendation commodity information can be pushed to the measured user through a short message, an application program, or the like. In some examples, the target recommendation model can include a model that makes recommendations based on the similarity of user behavior and a model that makes recommendations based on the probability prediction of commodity application. If the measured user has a history of applying for commodities, i.e., the measured user is an old user who has applied for commodities, the application tendency information of the measured user for the commodities includes user commodity interaction information, and the selected target recommendation model is the model that makes recommendations based on the similarity of user behavior. If the measured user does not have a history of applying for commodities, i.e., the measured user is a new user who has not applied for commodities, the application tendency information of the measured user for the commodities includes user portrait information, and the selected target recommendation model is the model that makes recommendations based on the probability prediction of commodity application. In some embodiments, the application tendency information of the sample user for the commodities can be divided into two parts, one part as a training sample for model training, and the other part as a verification sample for model verification. In the case where the performance parameters of the recommendation model meet the preset conditions, the verification sample can be used to verify the recommendation model to evaluate the training result of the model training, and if the training result meets the preset requirements, the recommendation model is determined as the target recommendation model for use. For ease of understanding, hereinafter, the user applying for commodities is taken as the user applying for a communication card, if the sample user is a user who has applied for a credit card, the application tendency information of the sample user can include the credit card benefit description information of the credit card once opened by the sample user, the recommendation model is trained according to the credit card benefit description information of the credit card once opened by the sample user, and the credit card benefit description information of the credit card preferred by the sample user predicted by the recommendation model is obtained. Based on the credit card benefit description information of the credit card once opened by the sample user, the information extracted from the credit card benefit description information of the credit card preferred by the sample user predicted by the recommendation model is used as sample information to train the recommendation model again, until the performance parameters of the recommendation model meet the standards, and the recommendation model is used as the target recommendation model. The target recommendation model can push credit card recommendation information to the measured user according to the application tendency information of the measured user who has once applied for a credit card, so as to recommend a credit card to the measured user. Similarly, if the sample user is a user who has not applied for a credit card, the application tendency information of the sample user can include user portrait information of the sample user, and the model training process of the recommendation model is similar to the model training process of the recommendation model related to the sample user who has applied for a credit card, which will not be described herein. The target recommendation model obtained by training can push credit card recommendation information to the measured user who has not applied for a credit card, so as to recommend a credit card to the measured user. The recommendation method of commodity information in the embodiments of the present application can recommend a set of commodities to a given user set A set of commodities And user commodity interaction information ε +In the case of implicit feedback, the product information is recommended, and the product information that the user may also be interested in is mined test . In this embodiment of the present application, denotes a Cartesian product, denotes the content other than the user-product interaction information ε + in the possible interaction relationship between the user and the product, i.e., the content that the embodiment of the present application hopes to mine. In the embodiment of the present application, the recommendation model is trained according to the application tendency information that can reflect the application tendency of the sample user to the product, and a set of preferred product information of the sample user predicted by the recommendation model is obtained. Based on the application tendency information, the inductive reasoning ability of the large language model is used to further extract information from the set of preferred product information as sample information to train the recommendation model until the performance parameter of the recommendation model meets the preset condition, and the recommendation model is used as a target recommendation model to recommend product information for the user under test. In the iteration process of training the recommendation model with the output of the recommendation model as the input of the large language model and the output of the large language model as the sample information, the output of the large language model is used as the sample information to continuously enhance the collaborative signal, so as to improve the problem of lacking collaborative signal, enhance the recommendation model, and improve the accuracy of the recommendation model in recommending product information for the user. Moreover, by iteratively training the recommendation model with the output of the large language model as the sample information step by step until the performance of the recommendation model no longer improves significantly, the collaborative signal is enhanced while the noise introduced into the model training is reduced, so that the enhancement of the collaborative signal and the reduction of the noise introduction can achieve a balance, maximizing the performance gain of the model training of the recommendation model. In order to verify the effectiveness of the product information recommendation method in the embodiment of the present application, three different ways of credit card information recommendation are performed with credit card application as an example, and the effects of the three different ways are compared. Way one: the product information recommendation method in the embodiment of the present application is used to push the credit card information of 10 credit cards output by the target recommendation model to the user under test. Way two: a recommendation strategy is obtained by professional personnel according to experience, 10 card-issuing institutions with the highest credit card application amount are selected, and the credit card information of the credit cards mainly promoted by the 10 card-issuing institutions is pushed to the user under test according to the card approval success rate of each card-issuing institution. Way three: 10 credit card information of credit cards are randomly selected from all credit cards and pushed to the user under test. After the credit card information is pushed to the user under test according to the above three ways, the statistical data is shown in Table One. Table One The old user is a user who has a history of applying for a commodity, that is, a user who has applied for a credit card. The new user is a user who does not have a history of applying for a commodity, that is, a user who has not applied for a credit card. The exposure quantity represents the number of credit card information pushed to the user. The quantity of incoming materials represents the number of users applying for a credit card. The number of users applying for a recommended card is the number of users applying for a credit card corresponding to the pushed credit card information. In mode one, the ratio of the number of users applying for a recommended card corresponding to the old user to the quantity of incoming materials is 46.51%, and the ratio of the number of users applying for a recommended card corresponding to the new user to the quantity of incoming materials is 56.53%. In mode two, the ratio of the number of users applying for a recommended card corresponding to the old user to the quantity of incoming materials is 37.57%, and the ratio of the number of users applying for a recommended card corresponding to the new user to the quantity of incoming materials is 43.20%. From the above, it can be seen that the accuracy of recommending credit card information to users by using mode one of the commodity information recommendation method in the present application is higher than that of mode two. In the scenario of old users, the recommendation accuracy of mode one is improved by 23.8% compared with the recommendation accuracy of mode two; in the scenario of new users, the recommendation accuracy of mode one is improved by 30.86% compared with the recommendation accuracy of mode two. It can be seen that the commodity information recommendation method provided in the present application can improve the recommendation accuracy of the recommendation model. In some embodiments, the process of extracting sample information from the set of preferred commodity information using a large language model and training the recommendation model using the extracted sample information can be performed cyclically to gradually improve the performance of the recommendation model. FIG. 2 is a flowchart of a commodity information recommendation method provided by another embodiment of the present application. The difference between FIG. 2 and FIG. 1 is that step S102 in FIG. 1 can be specifically refined as steps S1021 to S1023 in FIG. 2. In step S1021, the preferred commodity information and the repelled commodity information of the sample user are extracted from the set of preferred commodity information using a large language model according to the application tendency information and the set of preferred commodity information. The application tendency information can be used as the basis for reasoning by the large language model, and the set of preferred commodity information of the sample user predicted by the recommendation model can be used as a candidate set. The preferred commodity information and the repelled commodity information of the sample user are extracted from the candidate set by the large language model. The preferred commodity information of the sample user can represent the commodity that the large language model reasons to be of interest to the sample user in the set of preferred commodity information. In some examples, the preferred commodity information of the sample user can represent one or more commodities that the large language model reasons to be of most interest to the sample user in the set of preferred commodity information. The repelled commodity information of the sample user can represent the commodity that the large language model reasons to be of no interest to the sample user in the set of preferred commodity information. In some examples, the preferred commodity information of the sample user can represent one or more commodities that the large language model reasons to be of least interest to the sample user in the set of preferred commodity information. In some examples, the application tendency information of the sample user includes positive samples and negative samples, the positive samples represent the goods or the category of the goods to which the sample user tends to apply, and the negative samples represent the goods or the category of the goods to which the sample user does not tend to apply. In the process of reasoning the preferred good information and the repelled good information by the large language model based on the application tendency information, the application tendency information as the basis is the positive sample, that is, according to the positive sample and the preferred good information set, the preferred good information and the repelled good information are extracted from the preferred good information set by the large language model. In some examples, the large language model can be prompted to input the required preferred good information and repelled good information by setting a prompt. Specifically, a first prompt can be constructed according to the application tendency information and the preferred good information set, the first prompt including extraction requirement information, the application tendency information, and the preferred good information set; the first prompt is input into the large language model to obtain the preferred good information and the repelled good information output by the large language model. The extraction requirement information in the first prompt is used to give the requirement for reasoning by the large language model, the application tendency information in the first prompt can be used as the basis for reasoning by the large language model, and the preferred good information set in the first prompt can be used as the candidate set for extracting the preferred good information and the repelled good information by the large language model. For example, taking the credit card as the good and the user having a historical application of the credit card as an example, the first prompt input into the large language model and the output of the large language model can be as follows: Among them, "Please select one credit card that the user is most likely to like and one credit card that the user is most likely to dislike from the candidate set based on the user's historical application record of the credit card. When making the decision, you should analyze and reason according to the credit card benefit information I provide. When outputting the result, only output the serial number in the candidate set [] and do not give any analysis explanation or other information. The serial numbers of the most favorite card and the most disliked card are separated by a space." is the extraction requirement information in the first prompt. The content of the historical application record is the application tendency information in the first prompt, specifically the user good interaction information. The content of the candidate set is the preferred good information set in the first prompt. The large language model reasons based on the first prompt, and the output preferred good information is "495", indicating that the credit card indicated by

[0430] in the candidate set is the credit card that the user is most interested in.

[0495] The indicated credit card is the credit card that the user is most interested in according to the reasoning of the large language model; the output repelled good information of the large language model based on the first prompt is "430", indicating that the credit card indicated by

[0430] in the candidate set is the credit card that the user is least interested in. After obtaining the user preference information and the user repelled information, the analysis result given by the large language model, that is, the reply, can be obtained by inputting the prompt "why give this result". For example, taking a credit card as a commodity, and the user does not have a historical application commodity, the first prompt word input into the large language model and the output of the large language model can be as follows: Among them, "Please select one credit card that the user is most likely to like and one credit card that the user is least likely to like from the candidate set based on the user portrait. When making decisions, you should analyze and reason based on the user portrait information I provided. When outputting the results, only output the serial numbers in the [] in the candidate set, and do not give any analysis explanation or other information. The serial numbers of the most preferred card and the least preferred card are separated by a space." is the extraction requirement information in the first prompt word. The content of the user portrait is the application tendency information in the first prompt word, which is the user portrait information. The content of the candidate set is the preferred commodity information set in the first prompt word. The large language model reasons based on the first prompt word, and the output preferred commodity information is "2049", indicating that

[2049] The indicated credit card is the credit card that the user is most interested in according to the reasoning of the large language model; the large language model reasons based on the first prompt word, and the output rejection commodity information is "1897", indicating that

[1897] The indicated credit card is the credit card that the user is least interested in according to the reasoning of the large language model. After obtaining the user preference information and the user rejection information, the analysis result given by the large language model, i.e., the reply, can be obtained by inputting the prompt word "why give this result". In step S1022, the recommendation model is trained using the preferred commodity information, the rejection commodity information, and the application tendency information to obtain the performance parameters of the trained recommendation model. The application tendency information can be regarded as an original sample, and the preferred commodity information and the rejection commodity information output by the large language model can be regarded as newly added samples. The preferred commodity information can be regarded as a newly added positive sample, and the rejection commodity information can be regarded as a newly added negative sample. The recommendation model is trained using the preferred commodity information, the rejection commodity information, and the application tendency information as samples to adjust the model parameters of the recommendation model and obtain the performance parameters of the trained recommendation model to determine the performance of the recommendation model. In step S1023, it is determined whether the performance parameters meet the preset conditions. If the performance parameters meet the preset conditions, step S1024 is performed; if the performance parameters do not meet the preset conditions, step S1021 is returned. In step S1024, the recommendation model is determined as the target recommendation model. If the performance parameter meets the preset condition, it indicates that the recommendation model has reached the use requirement and can be put into use. If the performance parameter does not meet the preset condition, it indicates that the recommendation model has not reached the use requirement and needs to be further improved. The step S1021 is executed again, and the set of preferred commodity information in the step S1021 is the set of preferred commodity information output by the trained recommendation model according to the preferred commodity information, the excluded commodity information and the application tendency information. The steps of extracting the preferred commodity information and the excluded commodity information by the large language model, training the recommendation model, and obtaining the set of preferred commodity information output by the trained recommendation model are repeated until the performance parameter of the recommendation model meets the preset condition, and the recommendation model is determined as the target recommendation model. That is, if the performance parameter does not meet the preset condition, the recommendation model can be trained again by using the application tendency information, the preferred commodity information and the excluded commodity information extracted from the set of preferred commodity information output by the trained recommendation model by the large language model, until the performance parameter of the recommendation model meets the preset condition, and the recommendation model is determined as the target recommendation model. FIG. 3 is a schematic diagram of an example of a logic framework of a large language model enhanced recommendation algorithm based on a curriculum learning paradigm provided by an embodiment of the present application. As shown in FIG. 3, the original user commodity interaction information 21 can be used to train the recommendation model 22, and the recommendation model 22 predicts a set of preferred commodity information 23, which can be regarded as predicted user commodity interaction information. The original user commodity interaction information 21 and the set of preferred commodity information 23 are input into the large language model 24, and the large language model 24 extracts preferred commodity information and excluded commodity information 25 from the set of preferred commodity information 23. The preferred commodity information and the excluded commodity information 25 can be regarded as user commodity interaction information selected by the large language model 24, and the selected user commodity interaction information can be regarded as a sample expanded by the large language model 24. The preferred commodity information and the excluded commodity information 25 can be used as a sample to retrain the recommendation model. By means of the powerful instruction reserve and understanding and analysis capability of the large language model, the original application tendency information is expanded, the large language model faces natural language, the large language model can be placed in the back end, and the candidate set input into the large language model is reduced through the recommendation model, the calculation amount of the large language model is reduced, the pseudo-label generated by the large language model is used to enhance the recommendation algorithm of the recommendation model, and the pseudo-label is the preference commodity information and the repulsion commodity information output by the large language model as the sample in the above embodiment. The curriculum learning method used in the embodiment of the application is a machine learning training strategy, which trains the model by simulating the easy-to-difficult and step-by-step way in human education. The preference commodity information set predicted by the recommendation model is the information of the commodities that the recommendation model considers interesting to the user, and the repulsion commodity information extracted from the preference commodity information set by the large model is a difficult sample that is difficult for the recommendation model to distinguish. The recommendation model can be trained according to the easily distinguishable data, i.e., the application tendency information, and gradually transferred to training the recommendation model by using the difficult data, i.e., the difficult sample, so as to improve the discrimination ability of the recommendation model for the commodities interesting to the user. In some embodiments, the application tendency information includes user commodity interaction information, and the user has a historical application commodity. In practice, the description information of the historical application commodity has problems such as long narrative, disordered format, and low information density. If the description information of the historical application commodity is directly used to train the recommendation model, it is difficult to obtain ideal results. The refining and induction capability of the large language model can be used to refine the description information of the historical application commodity. For example, if the commodity is a credit card, the benefit information of the credit card can contain a large amount of detailed information such as numbers and times, but for credit card application, the user is mainly concerned about the category to which the benefit belongs, for example, the benefit information of the credit card “gift video application A, video application B, video application C membership 1 year (3 selected 1)” can be summarized as “video website membership”. Specifically, a second prompt word can be generated, the second prompt word includes refining requirement information and description information of the historical application commodity; the second prompt word is input into the large language model, and user commodity interaction information refined by the large language model from the description information is obtained. The second prompt word can guide the large language model to generate high-quality summaries according to the description information of the historical application commodity. For example, taking the commodity as a credit card and the description information of the historical application commodity as the benefit information of the credit card, the second prompt word can be as follows: Among them, the benefit details are the description information of the historical application commodity. The content in the reply is the user commodity interaction information refined by the large language model, and the user commodity interaction information is the information obtained by refining the large language model. The application tendency information can be text information, and the text information can be converted into a vector for facilitating model training. Specifically, the application tendency information can be converted into a first feature vector by using a sentence vector extraction model; the first feature vector is subjected to dimension reduction processing to obtain a second feature vector; and the second feature vector is used to train the recommendation model. The application tendency information can include user commodity interaction information. The sentence vector extraction model can be used to convert text information into a vector, and the sentence vector extraction model can include but is not limited to a Sentence Transformer model. The first feature vector is a feature vector of the application tendency information obtained by using the sentence vector extraction model. However, the first feature vector is a high-dimensional vector in dense representation, and therefore needs to be subjected to dimension reduction processing, which can be realized by using a dimension reduction algorithm, and the dimension reduction algorithm can include but is not limited to a UMAP (Uniform Manifold Approximation and Projection) algorithm. The second feature vector is a feature vector obtained after dimension reduction processing, and is a low-dimensional embedding vector. The second feature vector can be used as a feature vector of the application tendency information. FIG. 4 is a schematic diagram of an example of an embedded representation of user commodity interaction information provided by an embodiment of the present application. As shown in FIG. 4, the original credit card benefit description text 31 is input into a large language model 32, and the large language model 32 can output refined user commodity interaction information, i.e., refined credit card benefit description text 33. The refined credit card benefit description text is subjected to a sentence vector extraction model 34 to obtain a first feature vector, and the first feature vector is subjected to dimension reduction processing 35 to obtain a second feature vector, i.e., an embedded representation 36 of the credit card benefit description text. The second aspect of the present application provides a recommendation device for commodity information. FIG. 5 is a structural schematic diagram of a recommendation device for commodity information according to an embodiment of the present application. As shown in FIG. 5, the recommendation device for commodity information 400 can include a prediction result acquisition module 401, an enhanced training module 402, and an information pushing module 403. The prediction result acquisition module 401 can be used to train a recommendation model according to application tendency information of a sample user for a commodity to obtain a set of preferred commodity information of the sample user predicted by the recommendation model. The enhanced training module 402 can be used to iteratively train the recommendation model based on the application tendency information, the set of preferred commodity information, and a large language model until a performance parameter of the recommendation model satisfies a preset condition, and determine the recommendation model as a target recommendation model. The large language model is used to extract sample information for iteratively training the recommendation model from the set of preferred commodity information. The information pushing module 403 can be used to obtain recommendation commodity information according to application tendency information of a measured user for a commodity by using the target recommendation model and push the recommendation commodity information to the measured user. In some embodiments, the enhanced training module 402 can be specifically configured to: according to the application tendency information and the set of preferred commodity information, extract the preferred commodity information and the repelled commodity information of the sample user from the set of preferred commodity information by using the large language model; train the recommendation model by using the preferred commodity information, the repelled commodity information and the application tendency information to obtain a performance parameter of the trained recommendation model; if the performance parameter meets a preset condition, determine the recommendation model as a target recommendation model; if the performance parameter does not meet the preset condition, retrain the recommendation model by using the application tendency information, the preferred commodity information extracted from the set of preferred commodity information output by the trained recommendation model and the repelled commodity information until the performance parameter of the recommendation model meets the preset condition, and determine the recommendation model as the target recommendation model. In some examples, the application tendency information includes positive samples and negative samples. The enhanced training module 402 can be specifically configured to: according to the positive samples and the set of preferred commodity information, extract the preferred commodity information and the repelled commodity information from the set of preferred commodity information by using the large language model. In some examples, the enhanced training module 402 can be specifically configured to: according to the application tendency information and the set of preferred commodity information, construct a first prompt word, the first prompt word including extraction requirement information, the application tendency information and the set of preferred commodity information; input the first prompt word into the large language model to obtain the preferred commodity information and the repelled commodity information output by the large language model. In some embodiments, in the case that the sample user has a historical application commodity, the application tendency information includes user commodity interaction information, the user commodity interaction information representing the historical application commodity of the user, and the recommendation model includes a model for recommending based on the similarity of user behavior. In the case that the sample user does not have a historical application commodity, the application tendency information includes user portrait information, and the recommendation model includes a model for recommending based on the probability prediction of commodity application. In some embodiments, the application tendency information includes user commodity interaction information. The recommendation device 400 of commodity information can further include a refining module. The refining module can be configured to: generate a second prompt word, the second prompt word including refining requirement information and description information of the historical application commodity; input the second prompt word into the large language model to obtain the user commodity interaction information refined from the description information by the large language model. In some embodiments, the prediction result obtaining module 401 can be specifically configured to: convert the application tendency information into a first feature vector by using a sentence vector extraction model; perform dimension reduction processing on the first feature vector to obtain a second feature vector; and train the recommendation model by using the second feature vector. It should be noted that the commodity information recommendation device 400 is a device corresponding to the commodity information recommendation method described above, and all implementation manners in the method embodiments are applicable to the device embodiments, and the same technical effects can be achieved, and thus will not be described herein. The third aspect of the present application further provides an electronic device. FIG. 6 is a structural schematic diagram of an electronic device according to an embodiment of the present application. As shown in FIG. 6, the electronic device 500 includes a memory 501, a processor 502, and a computer program stored in the memory 501 and executable on the processor 502. In some examples, the processor 502 can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement one or more embodiments of the present application. The memory 501 can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions and when the software is executed (e.g., by one or more processors), it is operable to perform operations described with reference to the commodity information recommendation method according to embodiments of the present application. The processor 502 runs a computer program corresponding to the executable program code stored in the memory 501 by reading the executable program code, to implement the commodity information recommendation method in the above embodiments. In some examples, the electronic device 500 can further include a communication interface 503 and a bus 504. As shown in FIG. 6, the memory 501, the processor 502, and the communication interface 503 are connected through the bus 504 and complete communication among each other. The communication interface 503 is mainly used to realize communication between various modules, devices, units, and / or equipment in embodiments of the present application. Input devices and / or output devices can also be accessed through the communication interface 503. Bus 504 includes a hardware, software, or both that couples components of electronic device 500 to each other. As an example and not by way of limitation, bus 504 can include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand (IB) interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or some other suitable bus or interconnect, or a combination of two or more of these. Where appropriate, bus 504 can include one or more buses. Although this application describes and shows a particular bus, this application contemplates any suitable bus or interconnect. The fourth aspect of the present application provides a computer readable storage medium, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method for recommending commodity information in the above embodiments and achieve the same technical effects. To avoid repetition, details are not described here. The computer readable storage medium can include a non-transitory computer readable storage medium, such as a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, etc., which is not limited here. The fifth aspect of the present application provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the method for recommending commodity information in the above embodiments and achieve the same technical effects. To avoid repetition, details are not described here. It should be clarified that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. For the device embodiments, equipment embodiments, computer-readable storage medium embodiments, and computer program product embodiments, the relevant parts can be referred to the description section of the method embodiments. This application is not limited to the specific steps and structures described above and shown in the figures. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application. Furthermore, for the sake of brevity, detailed descriptions of known methods and techniques are omitted here. The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the above embodiments are exemplary and not restrictive. Different technical features appearing in different embodiments can be combined to achieve beneficial effects. Based on a study of the drawings, specification, and claims, those skilled in the art should be able to understand and implement other variations of the disclosed embodiments. In the claims, the term "comprising" does not exclude other means or steps; the quantifier "a" does not exclude a plurality; the terms "first" and "second" are used to identify names and not to indicate any particular order. No reference numerals in the claims should be construed as limiting the scope of protection. The functionality of multiple parts appearing in the claims can be implemented by a single hardware or software module. The appearance of certain technical features in different dependent claims does not mean that these technical features cannot be combined to achieve beneficial effects.

Claims

1. A method for recommending product information, comprising: training a recommendation model according to application tendency information of a sample user to a product to obtain a set of preferred product information of the sample user predicted by the recommendation model; iteratively training the recommendation model based on the application tendency information, the set of preferred product information, and a large language model until a performance parameter of the recommendation model meets a preset condition, determining the recommendation model as a target recommendation model, and the large language model is used to extract sample information for iteratively training the recommendation model from the set of preferred product information; obtaining recommended product information according to the application tendency information of a tested user to a product by using the target recommendation model and pushing the recommended product information to the tested user.

2. The method of claim 1, wherein, The iteratively training the recommendation model based on the application tendency information, the set of preferred product information, and the large language model until the performance parameter of the recommendation model meets the preset condition, and determining the recommendation model as the target recommendation model, comprises: extracting preferred product information and repulsive product information of a sample user from the set of preferred product information by using the large language model according to the application tendency information and the set of preferred product information; training the recommendation model by using the preferred product information, the repulsive product information, and the application tendency information to obtain a performance parameter of the trained recommendation model; if the performance parameter meets the preset condition, determining the recommendation model as the target recommendation model; if the performance parameter does not meet the preset condition, re-training the recommendation model by using the application tendency information, the preferred product information extracted from the set of preferred product information output by the trained recommendation model, and the repulsive product information until the performance parameter of the recommendation model meets the preset condition, and determining the recommendation model as the target recommendation model.

3. The method of claim 2, wherein, The application tendency information comprises positive samples and negative samples; The extracting the preferred product information and the repulsive product information of the sample user from the set of preferred product information by using the large language model according to the application tendency information and the set of preferred product information, comprises: extracting the preferred product information and the repulsive product information from the set of preferred product information by using the large language model according to the positive samples and the set of preferred product information.

4. The method of claim 2, wherein, The extracting the preferred product information and the repulsive product information of the sample user from the set of preferred product information by using the large language model according to the application tendency information and the set of preferred product information, comprises: constructing a first prompt word according to the application tendency information and the set of preferred product information, the first prompt word comprising extraction requirement information, the application tendency information, and the set of preferred product information; inputting the first prompt word into the large language model to obtain the preferred product information and the repulsive product information output by the large language model.

5. The method of claim 1, wherein, In a case where the sample user has historical application commodities, the application tendency information comprises user commodity interaction information, the user commodity interaction information characterizing historical application commodities of the user, and the recommendation model comprises a model for making recommendations based on similarity of user behaviors; In a case where the sample user does not have historical application commodities, the application tendency information comprises user portrait information, and the recommendation model comprises a model for making recommendations based on probability prediction of commodity application.

6. The method of claim 5, wherein, The application tendency information comprises the user commodity interaction information. The method further comprises: generating a second prompt word, the second prompt word comprising refined requirement information and description information of the historical application commodity; inputting the second prompt word into the large language model to obtain the user commodity interaction information refined from the description information by the large language model.

7. The method of claim 1, wherein, The training of the recommendation model according to the application tendency information of the sample user to commodities comprises: converting the application tendency information into a first feature vector by using a sentence vector extraction model; performing dimension reduction processing on the first feature vector to obtain a second feature vector; training the recommendation model by using the second feature vector.

8. A recommendation device for commodity information, comprising: a prediction result acquisition module configured to train a recommendation model according to application tendency information of a sample user to commodities to obtain a set of preferred commodity information of the sample user predicted by the recommendation model; an enhanced training module configured to iteratively train the recommendation model based on the application tendency information, the set of preferred commodity information, and a large language model until a performance parameter of the recommendation model meets a preset condition, determine the recommendation model as a target recommendation model, and use the large language model to extract sample information for iteratively training the recommendation model from the set of preferred commodity information; an information pushing module configured to obtain recommended commodity information according to the application tendency information of a test user to commodities by using the target recommendation model and push the recommended commodity information to the test user.

9. An electronic device comprising: a processor and a memory having computer program instructions stored thereon; the processor executes the computer program instructions to implement the recommendation method for commodity information according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, the computer program instructions being executed by a processor to implement the recommendation method for commodity information according to any one of claims 1 to 7.

11. A computer program product comprising a computer program, the computer program being executed by a processor to implement the recommendation method for commodity information according to any one of claims 1 to 7.

Images