Credit welfare platform commodity recommendation method based on large language model

Abstract

The invention relates to the technical field of e-commerce, in particular to an integral welfare platform commodity recommendation method based on a large language model, and the method comprises the following steps: 1, collecting data; step 2; step 3, model training; 4, performing personalized recommendation; the device comprises a data collection module, a data preprocessing module, a model training module, a recommendation algorithm module and a recommendation generation module. The commodity recommendation method aims at performing more accurate and personalized commodity recommendation on the user by introducing a large language model and a mixed recommendation algorithm and combining behavior data, basic information data and credit data of the user on a point welfare platform and other platforms, so that the user experience and the platform service quality are improved. According to the method, accurate analysis and commodity recommendation of user point use habits are realized by utilizing collaborative filtering, content-based recommendation and a mixed recommendation algorithm, and meanwhile, the intelligence and adaptability of a recommendation system are enhanced by combining multi-platform credit data.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and in particular to a method for recommending goods on a points welfare platform based on a large language model. Background Art

[0002] In the modern e-commerce environment, point welfare platforms have been widely used as a means to encourage user participation and consumption. However, existing point welfare platforms mainly rely on users' browsing, searching and purchasing history data to recommend products. This method has certain limitations in data utilization and cannot fully reflect users' actual needs and preferences.

[0003] For example, a method and system for recommending points products based on user portraits (publication number: CN117710049A, publication date: 2024-03-15), the method includes: obtaining user behavior data in the mall; building a user portrait based on the behavior data; obtaining all points product data and recently visited product data; and recommending points products to users based on the user portrait, the points product data and the recently visited product data.

[0004] This method can achieve accurate recommendation of point products based on user portraits and product characteristics. It comprehensively considers multiple attributes of users and products, and provides personalized point product recommendation services to different users for the same product based on differences in user portraits and behavioral habits. At the same time, weighted sorting algorithms and deep learning algorithms are also used to improve recommendation accuracy and achieve online calculation and rapid response. However, this method usually ignores the unique usage habits of users on the point welfare platform, especially the way and preference of users for free points. This results in the inability of existing recommendation systems to fully realize their potential when facing point welfare platforms, and user satisfaction and recommendation effects still need to be improved. Summary of the invention

[0005] In view of the technical defects existing in the background technology, the present invention proposes a product recommendation method for a points welfare platform based on a large language model, which solves the above technical problems and meets the actual needs. The specific technical solution is as follows: A method for recommending products on a points welfare platform based on a large language model comprises the following steps: Step 1: Data collection: Obtain multi-dimensional data of users and product data of the points welfare platform; Step 2: Data preprocessing: Perform data cleaning, data standardization, feature extraction and dimensionality reduction on the multidimensional data and product data obtained in step 1 to generate structured data input, which is divided into a training set, a validation set and a test set; Step 3: Model training: Use the large language model (LLM) and recommendation algorithm to train the model and build a recommendation model; Step 3.1: Select a suitable pre-trained large language model, use the training set pre-processed in step 2 to train the pre-trained large language model, and introduce LoRA technology for migration fine-tuning to obtain the fine-tuned large language model; Step 3.2: Use the fine-tuned large language model from step 3.1 to obtain user profiles and key features; Step 3.3: Select a suitable similarity calculation method, initialize the user similarity matrix, use the training set preprocessed in step 2 and the user portrait in step 3.2 to train the collaborative filtering algorithm model, calculate the similarity between users, obtain the result of similarity calculation, and make personalized recommendations based on user similarity; Step 3.4: Construct the feature vector space of the products, use the training set preprocessed in step 2 to train the content-based recommendation algorithm model combined with the key features in step 3.2, calculate the similarity between products, and make personalized recommendations based on the attribute characteristics of the products; Step 3.5: Combine the collaborative filtering algorithm model and the content-based recommendation algorithm model to build a hybrid recommendation algorithm and recommendation model; Step 3.6: Use the validation set described in step 2 to validate and evaluate the performance of the recommendation model, adjust and optimize the model based on the validation results, until the performance of the recommendation model on the validation set reaches a satisfactory level, and finally use the test set to evaluate the recommendation model; Step 4: Personalized recommendation: Based on the product data in step 1 and the output of the recommendation model in step 3, a personalized product recommendation list for the user is generated; It also includes a data collection module, a data preprocessing module, a model training module, a recommendation algorithm module and a recommendation generation module; The data collection module is responsible for collecting user behavior data, basic information data, multi-platform credit data and points usage habit data; The data preprocessing module is used to clean, standardize, extract features and reduce the dimension of the collected data; The model training module uses a large language model and a recommendation algorithm to train the preprocessed data and build a recommendation model; The recommendation algorithm module: implements collaborative filtering, content-based recommendation and hybrid recommendation algorithms to make personalized recommendations for users; The recommendation generation module generates a personalized product recommendation list for the user based on the output of the recommendation model and displays it to the user.

[0006] As a further technical solution of the present invention, the multidimensional data is the data of users on the points welfare platform and related platforms, and the multidimensional data includes user behavior data, user basic information data, user multi-platform credit data and points usage habit data; the product data is the attribute characteristics of the products on the points welfare platform.

[0007] As a further technical solution of the present invention, the user behavior data is the browsing, searching and purchasing behavior data of users on the platform; the user basic information data is the basic information of the user's age, gender, occupation and income; the user multi-platform credit data is the data of the user's credit score and consumption record on other platforms; the points usage habit data is the user's frequency of use of free points or habit data of redeeming product categories; wherein the weight of the points usage habit data in the multidimensional data is higher than that of other data.

[0008] As a further technical solution of the present invention, the step 2 comprises the following steps: Step 2.1: Data cleaning: Clean the multidimensional data and product data collected in step 1 to remove noise, errors, duplications and irrelevant data, including null value processing, illegal value processing, inconsistent data processing and duplicate data processing, and obtain the cleaned multidimensional data; Step 2.2: Data standardization: Standardize the cleaned multidimensional data to obtain the standardized original data; Step 2.3: Feature extraction: Extract useful features from the raw data in step 2.2, such as the user's purchase preferences and browsing history; Step 2.4: Dimensionality reduction: Use methods such as principal component analysis (PCA) to reduce data dimensions, retain key information, and reduce computational complexity.

[0009] As a further technical solution of the present invention, the recommendation algorithm of step 3 includes a user-based collaborative filtering algorithm, a content-based recommendation algorithm, and a hybrid recommendation system combining collaborative filtering and content-based recommendation algorithms.

[0010] As a further technical solution of the present invention, step 4 comprises the following steps: Step 4.1: Obtain the output result of the recommendation model described in step 3; Step 4.2: Generate a personalized product recommendation list for the user based on the user's multi-platform credit data and points usage habits, combined with the output of the recommendation model.

[0011] The beneficial effects of the present invention are: The present invention aims to make more accurate and personalized product recommendations to users by introducing a large language model and a hybrid recommendation algorithm, combining the user's behavioral data, basic information data and credit data on the points welfare platform and other platforms, thereby improving the user experience and platform service quality. The present invention uses collaborative filtering, content-based recommendation and hybrid recommendation algorithms to achieve accurate analysis of user point usage habits and product recommendations, while combining multi-platform credit data to enhance the intelligence and adaptability of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a process framework diagram of a product recommendation method for a points welfare platform based on a large language model. DETAILED DESCRIPTION

[0013] The following is combined with Figure 1 The implementation modes of the present invention are described with relevant embodiments, but the implementation modes of the present invention are not limited to the following embodiments, and the present invention relates to relevant necessary components in this technical field, which should be regarded as the known technology in this technical field and can be known and mastered by the technicians in this technical field.

[0014] With the development of big data and artificial intelligence technology, more and more recommendation systems have begun to introduce users' multi-platform credit data and behavioral data to improve the accuracy of recommendations. However, these systems usually ignore the unique usage habits of users on point welfare platforms, especially the way and preference of users for free points. This results in the existing recommendation systems being unable to fully realize their potential when facing point welfare platforms, and user satisfaction and recommendation effects still need to be improved.

[0015] In addition, although traditional recommendation algorithms, such as collaborative filtering and content-based recommendation, have solved the problem of personalized recommendation to a certain extent, they still have problems such as high computational complexity and insufficient recommendation accuracy when processing large-scale, multi-dimensional data. Therefore, there is an urgent need for a comprehensive recommendation system that combines multiple data sources and advanced algorithms to better meet user needs and improve the accuracy of recommendations and the user stickiness of the platform.

[0016] Although there is not necessarily a necessary connection between point usage habits and other user-related data, there is a certain correlation between them. Users' point usage habits usually reflect their consumption preferences, consumption capacity, and loyalty to the platform. These habits combined with other user-related data (such as behavioral data, basic information data, credit data, etc.) can provide a more comprehensive and accurate user portrait for product recommendations on the point welfare platform, thereby improving the accuracy of the recommendation system and user satisfaction. Furthermore, three aspects can reflect the cause and effect of point usage habits: ① Consumption preference: Points usage habits can reflect the user's preference for specific goods or services. For example, if a user frequently uses points to redeem hotel coupons, the recommendation system can infer that the user may have a high interest in travel and accommodation, and thus recommend related goods or services; ② Consumption capacity: Points usage habits can indirectly reflect the user's consumption capacity. For example, if a user frequently uses points to redeem high-end goods or services, the recommendation system can infer that the user has a high consumption capacity and recommend higher-value or higher-end goods or services. ③ Loyalty: Points usage habits can reflect the user’s loyalty to the platform. For example, if a user actively uses points to redeem goods or services, the recommendation system can infer that the user has a high loyalty to the platform, thereby recommending more diverse goods or services to enhance user stickiness.

[0017] The invention provides a method for recommending products on a points welfare platform based on a large language model. Figure 1 As shown, the recommendation method of the present invention solves the problems of insufficient data utilization, low recommendation accuracy and low user satisfaction in the existing recommendation system. The recommendation method of the present invention comprises the following steps: Step 1: Data collection: Obtain multi-dimensional data of users and product data of the points welfare platform; Step 2: Data preprocessing: Perform data cleaning, data standardization, feature extraction and dimensionality reduction on the multidimensional data and product data obtained in step 1 to generate structured data input, which is divided into training set, validation set and test set; Step 3: Model training: Use the large language model (LLM) and recommendation algorithm to train the model and build a recommendation model; Step 4: Personalized recommendation: Based on the product data in step 1 and the output of the recommendation model in step 3, a personalized product recommendation list is generated for the user.

[0018] The recommendation device of the present invention comprises a data collection module, a data preprocessing module, a model training module, a recommendation algorithm module and a recommendation generation module; Data collection module: responsible for collecting user behavior data, basic information data, multi-platform credit data and points usage habit data; Data preprocessing module: cleans, standardizes, extracts features and reduces dimension of collected data; Model training module: Use the large language model and recommendation algorithm to train the preprocessed data and build a recommendation model; Recommendation algorithm module: implements collaborative filtering, content-based recommendation and hybrid recommendation algorithms to provide personalized recommendations to users; Recommendation generation module: Generates a personalized product recommendation list for the user based on the output of the recommendation model and displays it to the user.

[0019] The present invention uses advanced intelligent recommendation algorithms to improve the accuracy of product recommendations by comprehensively utilizing the user's behavior data, basic information data, and multi-platform credit data on the points welfare platform. It also provides personalized product recommendation services based on the user's point usage habits, meets the user's personalized needs, and improves the user's satisfaction and loyalty to the platform.

[0020] Introduce large language models and hybrid recommendation algorithms to fully explore and utilize multi-dimensional and multi-platform data resources to improve the intelligence and refinement of the recommendation system. Through flexible algorithm design and model training, the recommendation system can quickly adapt to the needs of different user groups and usage scenarios, enhancing the scalability and versatility of the system. Through intelligent data analysis and recommendation strategies, provide accurate, timely and efficient product recommendation services to enhance the overall experience of users on the points welfare platform.

[0021] The following are examples of the present invention.

[0022] Example 1

[0023] A device for recommending products on a points welfare platform based on a large language model, comprising a data collection module, a data preprocessing module, a model training module, a recommendation algorithm module and a recommendation generation module; The data collection module is responsible for collecting user behavior data, basic information data, multi-platform credit data and points usage habit data; The data preprocessing module is used to clean, standardize, extract features and reduce the dimension of the collected data; The model training module uses a large language model and a recommendation algorithm to train the preprocessed data and build a recommendation model; The recommendation algorithm module: implements collaborative filtering, content-based recommendation and hybrid recommendation algorithms to make personalized recommendations for users; The recommendation generation module generates a personalized product recommendation list for the user based on the output of the recommendation model and displays it to the user.

[0024] A product recommendation method for a points welfare platform based on a large language model, such as Figure 1 As shown, the following steps are included: Step 1: Data collection: Obtain multi-dimensional data of users and product data of the points welfare platform; among them, multi-dimensional data refers to the data of users on the points welfare platform and related platforms, and multi-dimensional data includes user behavior data, user basic information data, user multi-platform credit data and points usage habit data; product data refers to the attribute characteristics of products on the points welfare platform.

[0025] User behavior data refers to the browsing, searching and purchasing behavior of users on the platform; User basic information data includes basic information such as user age, gender, occupation and income; User multi-platform credit data refers to the user's credit score and consumption records on other platforms; The points usage habit data is the user's habit data on the frequency of using free points or the types of redeemed goods, etc. Among them, the points usage habit data has a higher weight.

[0026] Step 2: Data preprocessing: Perform data cleaning, data standardization, feature extraction and dimensionality reduction on the multidimensional data and product data obtained in step 1 to generate structured data input, which is divided into training set, validation set and test set; Step 2.1: Data cleaning: Clean the multidimensional data and product data collected in step 1 to remove noise, errors, duplications and irrelevant data, including null value processing, illegal value processing, inconsistent data processing and duplicate data processing, and obtain the cleaned multidimensional data; Step 2.2: Data standardization: Standardize the cleaned multidimensional data to obtain the standardized original data; the standardization formula is: , is the original eigenvalue, is the feature mean, is the characteristic standard deviation.

[0027] Step 2.3: Feature extraction: Extract useful features from the raw data in step 2.2, such as the user's purchase preferences and browsing history; Step 2.4: Dimensionality reduction: Use methods such as principal component analysis (PCA) to reduce data dimensions, retain key information, and reduce computational complexity.

[0028] The original data set is converted into a set of linearly uncorrelated principal components through orthogonal transformation. The principal components are the features extracted from the original data set. The feature extraction process of PCA is to reduce the dimension of the data from n to k and sort the axis list in descending order of divergence.

[0029] First, based on the construction of the covariance matrix, we need to calculate the discreteness and covariance of the initial features. Assume that X has a data set consisting of m data, each of which is composed of n-dimensional column vectors. Arrange X into a matrix of n rows and m columns: , Then centralize: .

[0030] Afterwards, the coordinate transformation is required , w is an orthogonal basis, where W T W is a unit vector. If we reduce the n-dimensional space to the k-dimensional space, some coordinates in the coordinate system will be deleted, and the new coordinates will be , the projection of X in k dimension is , the original data is: , where w is the standard normal basis matrix, which can be rearranged to The largest eigenvector is taken out and normalized to form the matrix W. Each sample in the sample set is transformed to obtain the output sample set Z′.

[0031] Step 2.5: Divide the data after dimensionality reduction in step 2.4 into training set, validation set and test set.

[0032] Step 3: Model training: Use a large language model (LLM) and recommendation algorithm to train the model and build a recommendation model; a large language model is a type of deep learning model with a large number of parameters. In the field of natural language processing (NLP), they learn language patterns, grammar, and semantics by processing large amounts of text data, thereby understanding and generating human language.

[0033] Recommendation algorithms include user-based collaborative filtering algorithms, content-based recommendation algorithms, and hybrid recommendation systems that combine collaborative filtering and content-based recommendation algorithms.

[0034] Step 3.1: Select a suitable pre-trained large language model. The large language model in this embodiment is preferably GPT-3, GPT-4, PaLM, Galactica, and LLaMA. Use the training set pre-processed in step 2 to train the pre-trained large language model, and introduce LoRA technology for migration fine-tuning to obtain the fine-tuned large language model; In this step, the deep learning model introduces LoRA technology for migration fine-tuning. LoRA adds a side network to all fully connected layers in the pre-trained large model. This side network uses the product of two rank decomposition matrices to approximate the update value of the fully connected layer weights in domain adaptation.

[0035] Specifically, LoRA adds two structures A and B next to the pre-trained model structure. The parameters of these two structures are initialized to Gaussian distribution and 0, respectively, so the additional parameters are 0 at the beginning of training.

[0036] The input dimension of A and the output dimension of B are the same as the input and output dimensions of the original model, while the output dimension of A and the input dimension of B are much smaller than the input and output dimensions of the original model. During training, only the parameters of A and B are updated, and the pre-trained model parameters are fixed. During inference, the idea of ​​reparametrization can be used to merge AB and W, so that no additional calculations are introduced during inference. Moreover, for different downstream tasks, only A and B need to be retrained based on the pre-trained model, which can also speed up the training pace of large models.

[0037] Step 3.2: Use the fine-tuned large language model from step 3.1 to obtain user profiles and key features; Step 3.3: Select a suitable similarity calculation method, initialize the user similarity matrix, use the training set preprocessed in step 2 and the user portrait in step 3.2 to train the collaborative filtering algorithm model, calculate the similarity between users, obtain the result of similarity calculation, and make personalized recommendations based on user similarity; User-based collaborative filtering (user-based nearest neighbor recommendation). This algorithm needs to take the rating data and the identity number of the user to be recommended as input information, and then use the preference calculation formula to find similar users, that is, neighbors, of the current user to be recommended; finally, use the neighbor users to predict the current user's predicted rating for the unseen item p. The assumptions of this algorithm include that within a certain period of time, the interests and hobbies of the current user and the neighbor users will not change much, and if the interests of the two users are similar at the beginning, they should also be similar at the end. It turns out that this assumption is true.

[0038] The user-based collaborative filtering algorithm makes recommendations by calculating the similarity between users. The specific formula is as follows: ,in, represents the predicted rating of user u for product i, and sim(u,v) represents the similarity between user u and user v, which can be calculated by Pearson correlation coefficient or cosine similarity.

[0039] Pearson correlation coefficient formula: ; Cosine similarity formula: .

[0040] Step 3.4: Construct the feature vector space of the products, use the training set preprocessed in step 2 to train the content-based recommendation algorithm model combined with the key features in step 3.2, calculate the similarity between products, and make personalized recommendations based on the attribute characteristics of the products; The algorithm idea of ​​content-based recommendation is to compare the similarity between the user's preferred items and the items that have not been seen by reasonably using the item content to decide whether to recommend the item. This algorithm is rooted in information retrieval technology and rarely or does not require the evaluation of the items. First, it is necessary to extract the content features of the items, and then establish the user's interest model. The interest model is learned based on the content features of the items that the user has preferred in the past; finally, the user's interest model is compared with the similarity of the items to be recommended.

[0041] The specific formula is as follows: ;in, represents the predicted rating of user u for product i, represents the weight of user u on feature j, Represents the value of product i on feature j.

[0042] Step 3.5: Combine the collaborative filtering algorithm model and the content-based recommendation algorithm model to build a hybrid recommendation algorithm and recommendation model; the hybrid recommendation system combines collaborative filtering and content-based recommendation methods. The specific formula is as follows: ,in, and denote collaborative filtering and content-based recommendation scores, respectively. and is the weight coefficient, satisfying + =1.

[0043] Step 3.6: Use the validation set in step 2 to verify and evaluate the performance of the recommendation model. Adjust and optimize the model based on the validation results until the performance of the recommendation model on the validation set reaches a satisfactory level. Finally, use the test set to evaluate the recommendation model.

[0044] This embodiment uses flexible algorithm design and model training to enable the recommendation system to quickly adapt to the needs of different user groups and usage scenarios. The hybrid recommendation algorithm, combined with collaborative filtering and content-based recommendation methods, can maintain good recommendation effects in different data scenarios and enhance the scalability and versatility of the system.

[0045] Step 4: Personalized recommendation: Based on the product data in step 1 and the output of the recommendation model in step 3, a personalized product recommendation list is generated for the user.

[0046] Step 4.1: Get the output result of the recommendation model in step 3; Step 4.2: Generate a personalized product recommendation list for the user based on the user's multi-platform credit data and points usage habits, combined with the output of the recommendation model. Provide personalized product recommendation services based on the user's usage habits for free points, so that the recommendation results are more in line with the user's actual needs. By analyzing the user's point usage frequency and redemption product categories, recommend products that meet the user's preferences, and improve user satisfaction and platform stickiness.

[0047] The present invention improves the accuracy of product recommendations by integrating the user's behavior data, basic information data and multi-platform credit data on the points welfare platform and adopting advanced intelligent recommendation algorithms. Based on collaborative filtering, content-based recommendation and hybrid recommendation algorithms, it can better capture the user's interests and preferences and provide more accurate recommendation results.

[0048] Based on the user's usage habits of free points, we provide personalized product recommendation services to make the recommendation results more in line with the user's actual needs. By analyzing the user's point usage frequency and the redemption product categories, we recommend products that meet the user's preferences to improve user satisfaction and platform stickiness.

[0049] Introducing large language models and intelligent recommendation algorithms, fully tapping and utilizing multi-dimensional, multi-platform data resources, and improving the intelligence and refinement of the recommendation system. Through data cleaning, standardization, and feature extraction, the quality and consistency of data are guaranteed, providing a reliable data foundation for the recommendation algorithm.

[0050] User points data can be applied to a variety of advanced algorithms to improve the product recommendation effect of the points welfare platform. Different types of algorithms differ in how they use points data. For example, collaborative filtering algorithms analyze the similarities between users to find other users with similar point usage habits as the target user and recommend products that these users like; item-based collaborative filtering analyzes the frequency and method of redemption of different products to find other products similar to those redeemed by the target user and recommend them to the user; content-based recommendation algorithms use the historical records of user point redemptions to analyze the user's interests and preferences, and then recommend new products that are similar to the user's historical redemption products; hybrid recommendation algorithms combine collaborative filtering and content-based recommendation algorithms, using user points data and other user-related data to recommend new products. Provide more accurate recommendations; deep learning algorithms use deep learning models such as neural networks to extract features and recognize patterns from user points data and other user-related data, thereby achieving more complex user behavior prediction and product recommendations; reinforcement learning algorithms regard point redemption as an "action" and continuously adjust recommendation strategies through reinforcement learning algorithms to maximize user point redemption behavior; association rule mining algorithms analyze user point redemption history to mine association rules between products, such as "user A redeemed product X and then product Y", and then make recommendations; clustering algorithms cluster users according to their point usage habits and recommend the most suitable products for each cluster.

[0051] For example, user A often uses points to redeem electronic products. The recommendation system can recommend the latest electronic products to user A based on this habit. User B tends to choose daily necessities when redeeming points. The recommendation system can recommend cost-effective daily necessities to user B based on this habit.

[0052] There are some differences between the collection of user points data and the collection of other user-related data, mainly in terms of the nature, purpose, privacy protection and data collection methods of the data. When collecting and using data, it is necessary to comprehensively consider various factors and take corresponding measures to ensure the accuracy and security of the data and maximize the value of the data. This type of data mainly focuses on the user's point acquisition and redemption behavior on the platform, such as the way to obtain points, the frequency of redemption, and the type of redemption goods. These data directly reflect the user's consumption habits and preferences. Combined with the introduction of information such as basic information of the user (such as age, gender, occupation), behavioral data (such as browsing history, search history, purchase history), credit data (such as credit score, consumption record), etc., it can describe user behavior from multiple dimensions and establish a reasonable calculation basis.

[0053] Through flexible algorithm design and model training, the recommendation system can quickly adapt to the needs of different user groups and usage scenarios. The hybrid recommendation algorithm, combined with collaborative filtering and content-based recommendation methods, can maintain good recommendation effects in different data scenarios and enhance the scalability and versatility of the system.

[0054] Through intelligent data analysis and recommendation strategies, we provide accurate, timely and efficient product recommendation services to enhance the overall experience of users on the points welfare platform. The system can update user data and recommendation results in real time, ensuring the timeliness and accuracy of recommendations, so that users can get satisfactory recommendations every time they interact on the platform.

[0055] Improving the effectiveness of the recommendation system can not only improve user experience and satisfaction, but also increase the time users spend on the platform and the frequency of consumption, thereby increasing the overall value of the platform. Through accurate recommendations, users' search costs can be reduced, their dependence on the platform can be increased, and the platform's activity and revenue growth can be driven.

[0056] In the process of data collection and processing, we pay attention to user privacy protection and data security, adopt strict data encryption and anonymization measures to ensure the security and privacy of user data and enhance users' trust in the platform. This invention significantly improves the product recommendation accuracy and user satisfaction of the points welfare platform by comprehensively utilizing multi-dimensional data and advanced intelligent recommendation algorithms, provides strong data support and technical guarantee for platform operation, and improves user experience and platform value.

[0057] In terms of data usage, this method uses user points data mainly to optimize point welfare strategies, increase point redemption rates, and make more accurate product recommendations by analyzing users' point usage habits. With reference to algorithms such as Tik Tok video push, it can also be further applied to building user portraits and performing user segmentation. By analyzing user behavior and preferences, it can achieve personalized recommendations and precision marketing. Therefore, from another perspective, in addition to having high direct application value, it also has high commercial value.

[0058] By analogy, JD Finance's Points Mall uses a hybrid recommendation algorithm that combines the user's shopping history, browsing behavior, and points redemption records on the JD platform to recommend products that meet the user's preferences. For example, if the user frequently buys electronic products and has used points to redeem mobile phone cases, the recommendation system may recommend the latest mobile phone models or other electronic product accessories; China Mobile's points mall also uses a hybrid recommendation algorithm, combining users' mobile phone usage habits, call bill consumption records and points redemption history to recommend mobile phone packages, data packages, digital products, etc. For example, if a user frequently uses his mobile phone to surf the Internet and has redeemed data packages with points, the recommendation system may recommend a higher data package or Wi-Fi device.

[0059] The above is only a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for recommending products on a points welfare platform based on a large language model, characterized in that: The following steps are involved: Step 1: Data collection: Obtain multi-dimensional data of users and product data of the points welfare platform; Step 2: Data preprocessing: Perform data cleaning, data standardization, feature extraction and dimensionality reduction on the multidimensional data and product data obtained in step 1 to generate structured data input, which is divided into a training set, a validation set and a test set; Step 3: Model training: Use the large language model (LLM) and recommendation algorithm to train the model and build a recommendation model; Step 3.1: Select a suitable pre-trained large language model, use the training set pre-processed in step 2 to train the pre-trained large language model, and introduce LoRA technology for migration fine-tuning to obtain the fine-tuned large language model; Step 3.2: Use the fine-tuned large language model from step 3.1 to obtain user profiles and key features; Step 3.3: Select a suitable similarity calculation method, initialize the user similarity matrix, use the training set preprocessed in step 2 and the user portrait in step 3.2 to train the collaborative filtering algorithm model, calculate the similarity between users, obtain the result of similarity calculation, and make personalized recommendations based on user similarity; Step 3.4: Construct the feature vector space of the products, use the training set preprocessed in step 2 to train the content-based recommendation algorithm model combined with the key features in step 3.2, calculate the similarity between products, and make personalized recommendations based on the attribute characteristics of the products; Step 3.5: Combine the collaborative filtering algorithm model and the content-based recommendation algorithm model to build a hybrid recommendation algorithm and recommendation model; Step 3.6: Use the validation set described in step 2 to validate and evaluate the performance of the recommendation model, adjust and optimize the model based on the validation results, until the performance of the recommendation model on the validation set reaches a satisfactory level, and finally use the test set to evaluate the recommendation model; Step 4: Personalized recommendation: Based on the product data in step 1 and the output of the recommendation model in step 3, a personalized product recommendation list for the user is generated; It also includes a data collection module, a data preprocessing module, a model training module, a recommendation algorithm module and a recommendation generation module; The data collection module is responsible for collecting user behavior data, basic information data, multi-platform credit data and points usage habit data; The data preprocessing module is used to clean, standardize, extract features and reduce the dimension of the collected data; The model training module uses a large language model and a recommendation algorithm to train the preprocessed data and build a recommendation model; The recommendation algorithm module: implements collaborative filtering, content-based recommendation and hybrid recommendation algorithms to make personalized recommendations for users; The recommendation generation module generates a personalized product recommendation list for the user based on the output of the recommendation model and displays it to the user.

2. The method for recommending products on a points welfare platform based on a large language model according to claim 1, characterized in that: The multi-dimensional data refers to the data of users on the points welfare platform and related platforms, including user behavior data, user basic information data, user multi-platform credit data and points usage habit data; the commodity data refers to the attribute characteristics of commodities on the points welfare platform.

3. The method for recommending goods on a points welfare platform based on a large language model according to claim 1, characterized in that: The user behavior data refers to the browsing, searching and purchasing behavior data of the user on the platform; the user basic information data refers to the basic information of the user's age, gender, occupation and income; the user multi-platform credit data refers to the credit score and consumption record data of the user on other platforms; The points usage habit data is the user's usage frequency of free points or habit data of redeeming commodity categories; wherein the weight of the points usage habit data in the multidimensional data is higher than that of other data.

4. The method for recommending goods on a points welfare platform based on a large language model according to claim 1, characterized in that: The step 2 comprises the following steps: Step 2.1: Data cleaning: Clean the multidimensional data and product data collected in step 1 to remove noise, errors, duplications and irrelevant data, including null value processing, illegal value processing, inconsistent data processing and duplicate data processing, and obtain the cleaned multidimensional data; Step 2.2: Data standardization: Standardize the cleaned multidimensional data to obtain the standardized original data; Step 2.3: Feature extraction: Extract useful features from the raw data in step 2.2, such as the user's purchase preferences and browsing history; Step 2.4: Dimensionality reduction: Use methods such as principal component analysis (PCA) to reduce data dimensions, retain key information, and reduce computational complexity.

5. The method for recommending goods on a points welfare platform based on a large language model according to claim 1, characterized in that: The recommendation algorithm of step 3 includes a user-based collaborative filtering algorithm, a content-based recommendation algorithm, and a hybrid recommendation system combining collaborative filtering and content-based recommendation algorithms.

6. The method for recommending products on a points welfare platform based on a large language model according to claim 1, characterized in that: The step 4 comprises the following steps: Step 4.1: Obtain the output result of the recommendation model described in step 3; Step 4.2: Generate a personalized product recommendation list for the user based on the user's multi-platform credit data and points usage habits, combined with the output of the recommendation model.

Images