An interpretable friend recommendation method based on large language model label classification

Abstract

The application discloses an interpretable friend recommendation method based on large language model label classification, comprising the following steps: obtaining text attribute information and attention information of a user; constructing a text attribute word library of the user, re-segmenting and updating the text attribute of the user based on the text attribute word library, and obtaining an updated text attribute information set; taking the updated text attribute information set as input of a large language model Q&A question and answer mode and performing feature extraction and classification; calculating the similarity, attention and recommendation degree between users; obtaining a recommended candidate recommendation list CRL based on the recommendation degree; and optimizing the candidate recommendation list CRL by using a multi-objective evolutionary optimization framework to obtain an optimal recommendation list. The application aims to realize the interpretability of the recommendation result between users, and finally obtains a balanced solution set satisfying the user demand to solve the personalized friend recommendation problem.

Description

Technical Field

[0001] This invention relates to the field of friend recommendation, and more specifically to an interpretable friend recommendation method based on large language model label classification. Background Technology

[0002] A recommender system (RS) is essentially an information filtering system designed to recommend products, services, or information to users based on their preferences and behaviors. Algorithmic recommendations aim to save users time, improve user satisfaction, and build user trust. However, users not only want recommendations but also want to understand the reasons behind them. Explainable recommender systems differ from traditional ones by providing users with reasons for the recommendations, clearly explaining why these items are recommended, making the recommendations more acceptable to users. Explainable recommender systems not only increase the transparency of the system but also enhance user trust and acceptance, thereby improving user satisfaction with the recommended products.

[0003] Depending on the specific application scenario, recommendation systems employ various interpretation methods. For example, there are user- and item-based interpretations. User-based interpretations analyze the behavioral characteristics of target users to identify user groups with similar interests and experiences, then recommend potentially interesting content based on these group preferences. Item-based interpretations allow users to understand that recommended items are similar to other items they have previously liked. Feature-based interpretations match the features of candidate items that the user is interested in with the user's preferences. By using features as an explanation, users can intuitively understand why content is recommended, such as generating scholar recommendations based on research interests, academic fields, and schools. Text-based interpretations generate textual information to help users understand each recommendation. Visual interpretations leverage the intuitiveness of visual images for interpretable recommendations.

[0004] In recent years, with the explosive growth of internet information, tags (keywords specified by users to describe items) have become increasingly important in the field of internet information retrieval, and tag recommendation has therefore attracted widespread attention from researchers. First, tags can be freely associated by users, reflecting their personalized preferences. Second, tags express semantic relationships between items, which helps in assessing the quality of potential items. Third, by utilizing the co-occurrence of tags, both user communities and item clusters can be built, further helping target individuals find relevant and interesting items. Therefore, tags offer a promising method for solving some thorny problems in recommender systems, such as the cold start problem.

[0005] Numerous studies have explored the application of tags in recommender systems. Wu et al. proposed a generative model (Tag2Word) that generates words based on tag-word distribution and the tags themselves. He et al. addressed the cold-start problem by proposing a recommendation algorithm combining tag data and Naive Bayes classification. Tag data represents user attribute features, a Naive Bayes classifier matches new and returning users, and similar user groups infer new users' interests. Chen et al. proposed a graph neural network-enhanced personalized tag recommendation model that integrates graph neural networks into a pairwise interaction decomposition model, considering two types of interaction graphs generated by tag assignment (user-tag interaction graph and item-tag interaction graph). For each interaction graph, a graph neural network captures the encoded cooperation signals, and these signals are incorporated into the learning of entity representations by transmitting and assembling representations of entity neighbors along the interaction graph. Zou et al. proposed an attention-based tag-aware recommendation model that captures multiple latent tag-based features of users and items. This model uses an embedding method to generate dense tag-based feature vectors for each user and each item. To compress these vectors into fixed-length feature vectors, an attention pooling layer is constructed, which automatically assigns different weights based on the importance of different features. The feature vectors of users and items are concatenated and used as input to a multi-layer fully connected network to learn non-linear, high-level interaction features.

[0006] Although tag recommendation algorithms vary, based on the three main characteristics of tag recommendation tasks—the diversity of item content, the relevance between tags, and the differences in user preferences—tag recommendation methods can be divided into three categories: content-based tag recommendation methods; tag relevance-based recommendation methods; and user preference-based recommendation methods. Similarly, current recommendation systems also face several challenges. First, they are highly complex. Interpretable recommendation systems need to consider multiple factors such as user preferences, behavioral characteristics, and recommendation rationale, making system design and implementation complex. Second, there is the trade-off between interpretability and performance. Prioritizing interpretability may compromise system performance and accuracy, requiring a balance between the two. Third, there is the difficulty of user comprehension. Users may struggle to understand the explanations or reasons provided by the recommendation system, leading to decreased trust in the recommendations. Fourth, there is data sparsity. Faced with data sparsity, interpretable recommendation systems may struggle to accurately explain the reasons or basis for recommendation results. Summary of the Invention

[0007] The purpose of this invention is to provide an interpretable friend recommendation method based on large language model label classification, which aims to recommend other friends with similar characteristics to each user across multiple platforms. The recommendation results should not only meet the requirements of accuracy and diversity, but also interpretability, so as to explain the reasons or basis for the recommendation results, thereby further realizing personalized friend recommendations for users.

[0008] In one aspect, the present invention proposes an interpretable friend recommendation method based on large language model label classification. According to an embodiment of the present invention, the method includes the following steps:

[0009] (1) Obtain the user's text attribute information and follow information;

[0010] (2) Construct a text attribute dictionary for users, re-segment and update user text attributes based on the text attribute dictionary, and obtain the updated set of text attribute information;

[0011] (3) The updated set of text attribute information is used as the input of the large language model Q&A question answering mode and features are extracted and classified to construct the user's basic information and obtain the user's classification information;

[0012] (4) Calculate the similarity, attention, and recommendation rates between users;

[0013] (5) Based on the recommendation degree obtained in step (4), a matrix ReSim with dimension N×N is obtained. Each column represents the recommendation degree between the current user and other users. The recommendation degree values ​​are sorted in descending order and the top K are taken to obtain the candidate recommendation list CRL recommended for N users, with dimension K×N.

[0014] (6) The candidate recommendation list CRL is optimized using a multi-objective evolutionary optimization framework to obtain the optimal recommendation list.

[0015] Furthermore, the interpretable friend recommendation method based on large language model label classification according to the above embodiments of the present invention may also have the following additional technical features:

[0016] In some embodiments of the present invention, step (1) is specifically as follows: Let U be the set of all text attribute information for each user, where U = {u1, u2, ... u...} i ,…u N Let V represent all the information followed by each user, where V = {v1, v2, ... v}. i ,…,v N}

[0017] In some embodiments of the present invention, step (2) is specifically as follows: assuming there are N users and digital serialization is performed, the text attribute information set U = {u1, u2, ... u} for the N users is... i ,…u N The problem of over-segmentation exists. A new text attribute lexicon, Lex, is constructed, and the text attribute information set U = {u1, u2, ... u...} of N users is analyzed. i ,…u N The process involves re-segmenting the text, removing duplicate information from user attributes, deleting stop words, and unifying case, ultimately resulting in an updated set of text attribute information U' = {u'1, u'2, ..., u'}. i ,…u' N}

[0018] In some embodiments of the present invention, in step (3), the classification is specifically divided into seven categories of labels, and the classification information is denoted as Lab={lab1,lab2,…,lab i ,…,lab N}, lab i The set of labels for the i-th user.

[0019] In some embodiments of the present invention, step (4) is specifically as follows:

[0020] (401) Calculate the similarity J between users

[0021] The similarity between users with the same attribute tags is calculated using the Jaccard similarity coefficient:

[0022]

[0023] In the formula, J(A,B) represents two user label sets, A and B. A and lab B The similarity between them, lab A ∩lab B The number of elements in the intersection of user A's tag set and user B's tag set is denoted by lab. A ∪lab B J represents the number of elements in the union of the tag set of user A and the tag set of user B. The larger the J value, the more similar the attribute tags of user A and user B are.

[0024] (402) Calculate the attention level Att among users

[0025] Based on all the attention information of N users, V = {v1, V2, ... v} i ,…,v NTo build a social network, the formula for user attention is defined as follows:

[0026]

[0027] In the formula, Att(A,B) represents the potential attention level of user A to user B, and user B is one of the users that user A follows; AN(A,B) represents the number of users that user A follows who follow B; max(AN(A)) represents the maximum number of users that user A follows who follow B. The larger the value of Att, the greater the probability that user A follows user B.

[0028] (403) Calculate the recommendation rate between users using ReSim

[0029] The recommendation score of user A to user B is defined as follows:

[0030]

[0031] In the formula, the larger the ReSim value, the greater the probability that user A will recommend user B.

[0032] In some embodiments of the present invention, step (6) includes the following steps: selecting the accuracy and diversity of the recommendation list as a multi-objective function, initializing the population, and population evolution.

[0033] In some embodiments of the present invention, the selection of the accuracy and diversity of the recommendation list as a multi-objective function is as follows:

[0034] Let PopSize be the population size, maxGen be the total number of iterations, Gen be the current iteration number set to 0, and k be the number of friend users recommended to a single user;

[0035] Regarding accuracy, similarity values ​​are used for combined calculations. For user A, the top k users are selected from the candidate recommendation list CRL for similarity calculation, as shown below:

[0036]

[0037] In the formula, k is the length of the recommendation list R, and ReSim(A,R(i)) represents the recommendation similarity between user A and the i-th user in the friend user list R recommended for user A;

[0038] The diversity of the recommendation list is measured by the number of tag attributes. For user A, the number of attribute tags for each user in the friend user recommendation list R is calculated, and the number of tags is used as a reflection of diversity, as shown below:

[0039]

[0040] In the formula, k is the length of the recommendation list R, and |lab(R(j))| represents the number of tags of the i-th user in the recommendation list R recommended by user A.

[0041] In some embodiments of the present invention, the initialization of the population is specifically as follows: an initial population P is generated using a random initialization strategy, that is, a decision matrix of row k columns of PopSize is randomly generated, and a target matrix of row M columns of PopSize is calculated.

[0042] In some embodiments of the present invention, the population evolution includes the following steps:

[0043] a. Individual choice

[0044] Population P is randomly divided into two equal subpopulations, P1 and P2. The relative superiority of individuals between the two subpopulations is calculated using the density estimate Fit based on migration, as defined below:

[0045]

[0046] In the formula, Fit(p) i P) represents individual p i In the migration-based density estimate of population P, p i and p j Let d(p) be the i-th and j-th individuals in population P, and n represent the population size of P; when d(p) i ,p j When d(p) is less than 0, i ,p j If d(p) is 0, otherwise d(p) i ,p j ) for individual p i and individual p j The Euclidean distance between them in the objective function f; f m Let f be the m-th objective function, and M represent the number of objective functions f. In this paper, there are two objective functions f, namely Acc and Cov. It is assumed that subpopulation P1 is better than P2.

[0047] b. Generation of offspring

[0048] Based on subpopulations P1 and P2, crossover and mutation operators are used to generate two new populations P'1 and P'2, and these two new populations are merged to obtain the final population P'.

[0049] c. Environmental selection

[0050] The population P∪P' is calculated and sorted using a non-dominated sorting algorithm and a transition-based density estimation method to obtain a candidate population with multiple frontiers. The individuals with the highest PopSize are selected and retained for the next generation, i.e., updated to population P. Finally, the iteration number Gen = Gen + 1 is set, and the process is repeated in step a until the iteration number Gen > maxGen.

[0051] d. Select the first individual from the sorted population P as the optimal recommendation list for the user and output it.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] 1. This invention proposes an interpretable friend recommendation method based on large language model tag classification, providing an effective solution for personalized recommendations in the absence of historical data. By utilizing a large language model to extract and represent text terms, and combining user attention information with the personalized friend recommendation capabilities of a multi-objective evolutionary algorithm, this algorithm can improve recommendation performance under cold start conditions. Simultaneously, the capabilities of the large language model can generate reasonable recommendation explanations, increasing the transparency and user trust of the recommendation system, and providing users with a better recommendation experience.

[0054] 2. Current recommendation algorithms rely too heavily on general metrics such as popularity and trending, resulting in a lack of personalization and novelty in the recommendations. Furthermore, the lack of explanation and justification for these recommendations makes it difficult for users to understand the rationale behind them, reducing user trust in the recommendation system. This invention proposes an explainable friend recommendation method based on large language model tag classification. It uses a large language model to classify users' textual attribute information into tags, and then employs a multi-objective evolutionary framework to enhance the diversity among users. Tag classification effectively explains the rationale behind the recommendations, tag-based recommendation degree calculation improves the accuracy of the recommendations, and the multi-objective evolutionary framework increases the diversity of the recommendations. Attached Figure Description

[0055] Figure 1 This is a flowchart of an interpretable friend recommendation method based on large language model tag classification in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the content classification part of the large language model proposed in this embodiment of the invention;

[0057] Figure 3 This is a schematic diagram of the multi-objective evolutionary recommendation part proposed in an embodiment of the present invention;

[0058] Figure 4 This is a friend recommendation interface designed in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0060] An interpretable friend recommendation method based on large language model label classification, the process of which is as follows: Figure 1 As shown, the specific steps are as follows:

[0061] Step 1: Obtain the user's text attribute information and follow information.

[0062] Let U be the set of all text attribute information for each user, where U = {u1, u2, ..., u} i ,…u N}, u i Let V represent the text attribute information of the i-th user, and N be the total number of users; let V be all the follow information of each user, where V = {v1, v2, ... v...} i ,…,v N}, v i This represents the information followed by the i-th user;

[0063] Step 2: Generate text attribute information U = {u1, u2, ... u} for all users. i ,…u N The process involves analyzing and identifying over-segmented attribute information, and then using this information to construct a user text attribute lexicon. Based on this lexicon, the user text attribute information is re-segmented and updated to obtain an updated set of user text attribute information. Note: Over-segmentation refers to the unnecessary splitting of a word into multiple parts in natural language processing, making the originally grammatically and semantically correct word incomplete or inaccurate.

[0064] Suppose there are N users and the data is serialized. The set of text attribute information for each of the N users is U = {u1, u2, ..., u...} i ,…u N The problem of over-segmentation occurs; for example, "Tianjin University of Technology" is segmented into "Tianjin" and "University of Technology"; "ping-pong paddle" is segmented into "ping-pong" and "paddle", etc. Therefore, based on these issues, this method first constructs a new lexicon, Lex, which contains the over-segmented words in the dataset. Then, it analyzes the text attribute information set U = {u1, u2, ... u...} of N users. i ,…u NThe process of re-segmenting text, removing duplicate information from user attributes, deleting stop words, and unifying capitalization is performed to obtain the updated text attribute information set U' = {u'1, u'2, ..., u'}. i ,…u' N}

[0065] Step 3: Use the updated set of text attribute information as input to the Q&A question-answering mode of the large language model, perform feature extraction and classification, construct the user's basic information, and obtain the user's classification information.

[0066] By asking a question to the large language model, such as "Based on the provided text attribute terms, please divide them into 7 categories," the large language model will then answer with the specific 7 categories. For example, if we provide the large language model with user information from a scholar website and ask, "Based on the provided text attribute terms, please divide them into 7 categories," the large language model would answer, "They can be divided into the following 7 categories: Name, Work Unit or School, Research Interests, Degree or Title, Academic Organization, Competition Awards, and Other." Based on the 7 categories answered by the large language model, we construct the basic information of the users and obtain the key information (tags) of N users: Lab = {lab1, lab2, ..., lab...} i ,…,lab N}

[0067] For example Figure 2 As shown, users are serialized into names {User1, User2, ..., Useri, ..., UserN}, where Useri is the name of the i-th user. Then, a Q&A pattern of Large Language Model (LLM) is applied to the text attribute information of each user, dividing all text attribute information into 7 categories, and labeling the N users after classification with Lab = {lab1, lab2, ..., lab...} i ,…,lab N The table indicates that lab i Let lab represent the set of tags for the i-th user. i = {Label1,Label2,…,Labelj,…,Labeln}, where Labelj is the j-th label of the i-th user, and the value of n is determined by the length of the text attribute information of the i-th user.

[0068] Step 4: Calculate the similarity, attention, and recommendation rates between users.

[0069] Step 4.1: Obtain the similarity J between users.

[0070] The similarity between users with the same attribute tags is calculated using the Jaccard similarity coefficient, as defined below:

[0071]

[0072] In equation (1), J(A,B) represents two user label sets, A and B. A and lab B The similarity between them, lab A ∩lab B Wherein represents the number of elements in the intersection of user A's tag set and user B's tag set, lab A ∪lab B J represents the number of elements in the union of the tag sets of user A and user B. The larger the J value, the more similar the attribute tags of user A and user B are. For example, if user A has 3 tags and user B has 3 tags, and user A and user B have 2 overlapping tags, then J(A,B) = 2 / (3+3-2) = 0.5.

[0073] Step 4.1: Obtain the level of attention (Att) between users.

[0074] Based on all the follow information of N users, V = {v1, v2, ..., v...} N To build a social network, the formula for user attention can represent the attention information of users in the social network. The defined attention level is as follows:

[0075]

[0076] In equation (2), Att(A,B) represents the potential attention level of user A towards user B, where user B is one of the users followed by the users followed by user A. AN(A,B) represents the number of users followed by user A who are also following user B, and max(AN(A)) represents the maximum number of users followed by the users followed by user A. The larger the value of Att, the greater the probability that user A is following user B. For example, if user A follows 3 users who are also following user B, then AN(A,B) = 3. Since the number of users followed by user B is the maximum, then max(AN(A)) = 3, and Att(A,B) = 3 / 3 = 1.

[0077] For example Figure 2As shown, the third user, User3, has three following entries: v3 = {User0, User1, User65}. User0 follows User73, User1 follows User0, User59, User63, User70, User73, and User76, and User65 follows User0 and User73. Since User3 has already followed User0, Att(User3, User0) = 0. Statistics show that User59, User63, User70, and User76 were followed once, and User73 was followed three times. Therefore, Att(User3, User59) = 1 / 3, Att(User3, User63) = 1 / 3, Att(User3, User70) = 1 / 3, Att(User3, User76) = 1 / 3, and Att(User3, User73) = 3 / 3 = 1.

[0078] Step 4.2: Obtain the recommendation rate among users (ReSim).

[0079] The Recommendation List (CRL) for each user is based on their recommendation score. The recommendation score is calculated by weighting the similarity between users with similar attributes and their engagement within the same social network; the sum of these weights is 1. Therefore, the specific definition of the recommendation score for recommending user B to user A is as follows:

[0080]

[0081] In equation (3), the larger the ReSim value, the greater the probability of recommending user B to user A. Based on the examples given in steps 1.3 and 1.4, ReSim(A,B)=(1+0.5) / 2=0.75.

[0082] Step 5: Calculate the recommendation score among N users based on equation (3), resulting in an N×N matrix ReSim, where each column represents the recommendation score between the current user and other users. Sort the recommendation scores in descending order and take the top K=100 to obtain a candidate friend recommendation list CRL for N users, with a list length of K.

[0083] Step 6: Optimize the candidate recommendation list CRL using a multi-objective evolutionary optimization framework, and design two objective functions to satisfy the accuracy and diversity of user recommendation results; let PopSize be the population size, maxGen be the total number of iterations, Gen be the current iteration number set to 0, and k be the number of friend users recommended to a single user.

[0084] Step 6.1: Set the objective function

[0085] We choose accuracy and diversity of the recommendation list as the multi-objective function. For accuracy, we use similarity values ​​for combined calculation. For user A, we select the top k users from the candidate recommendation list CRL and calculate their similarity. The specifics are as follows:

[0086]

[0087] In equation (4), k is the length of the recommendation list R, and ReSim(A,R(i)) represents the recommendation similarity between user A and the i-th user in the friend user list R recommended for user A.

[0088] The diversity function for the recommendation list is measured by the number of attribute tags. For user A, we calculate the number of attribute tags for each user in the friend recommendation list R, using the number of tags as a reflection of diversity. The details are as follows:

[0089]

[0090] In equation (5), k is the length of the recommendation list R, and |lab(R(j))| represents the number of tags of the i-th user in the list R recommended by user A.

[0091] Step 6.2, Population Evolution

[0092] Step 6.2.1, Individual Selection

[0093] Population P is randomly divided into two equal subpopulations, P1 and P2. The relative superiority of individuals between the two subpopulations is calculated using the density estimate Fit based on migration, as defined below:

[0094]

[0095] In equation (6), Fit(p) i P) represents individual p i In the migration-based density estimate of population P, p i and p j Let d(p) be the i-th and j-th individuals in population P, and N represent the population size of P; when d(p) i ,p j When d(p) is less than 0, i ,p j If d(p) is 0, otherwise d(p) i ,p j ) for individual p i and individual p j The Euclidean distance between them in the objective function f; f mLet f be the m-th objective function, and M represent the number of objective functions f. In this paper, there are two objective functions f, namely Acc and Cov. It is assumed that subpopulation P1 is better than P2.

[0096] Step 6.2.2, Generation of Offspring

[0097] Based on subpopulations P1 and P2, crossover and mutation operators are used to generate two new populations P'1 and P'2, and these two new populations are merged to obtain the final population P'.

[0098] Step 6.2.3, Environment Selection

[0099] The population P∪P' is calculated and sorted using a non-dominated sorting algorithm and a transition-based density estimation method to obtain a candidate population with multiple frontiers. The individuals with the highest PopSize are selected and retained for the next generation, which is then updated to the population P. Finally, the iteration count Gen = Gen + 1 is set, and the process is repeated in step 6.2.1 until the iteration count Gen > maxGen.

[0100] For example Figure 3 As shown, assume a population P is generated for user User1 and divided into subpopulations P1 and P2. Crossover and mutation occur between subpopulations P1 and P2 to obtain two new populations P'1 and P'2. These two new populations are then merged to obtain the final population P'. The next generation population P is then selected through environmental selection. When the number of iterations Gen > maxGen, the optimal individual in population P is output, which is also User1's recommendation list, with a length of k = 5.

[0101] Step 7: Select the first recommended individual from the sorted population P as the optimal individual for the user and output it.

[0102] For example Figure 4 The image shows the designed user recommendation interface. Assuming we input the question "Recommend other users to user10," the interface will output five recommended friends for user10 and analyze them. This invention, by designing a user recommendation interface that displays the basis for the recommendation results, not only increases user trust in the recommendations but also makes them more acceptable to users. Furthermore, displaying recommendation results through the user interface facilitates the collection of user feedback. Users can directly see detailed information about recommended items on the interface, making it easier to interact with the system, provide feedback, and make the system more intelligent and personalized.

[0103] The above description is merely an example and illustration of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the present invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. An interpretable friend recommendation method based on large language model label classification, characterized in that, Includes the following steps: (1) Obtain the user's text attribute information and follow information; (2) Construct a text attribute dictionary for users, re-segment and update the text attributes of users based on the text attribute dictionary, and obtain the updated set of text attribute information; (3) The updated set of text attribute information is used as the input of the large language model Q&A question answering mode and features are extracted and classified to construct the user's basic information and obtain the user's classification information; (4) Calculate the similarity, attention, and recommendation rates among users; (5) Based on the recommendation score obtained in step (4), a dimension of is obtained. matrix Each column represents the recommendation score between the current user and other users, sorted in descending order based on the recommendation score and taking the top results. Thus obtain Candidate recommendation list recommended by users And the dimension is ; (6) Use a multi-objective evolutionary optimization framework to optimize the candidate recommendation list Optimize to obtain the best recommendation list; Specifically, step (4) is as follows: (401) Calculate the similarity J between users The similarity between users with the same attribute tags is calculated using the Jaccard similarity coefficient: In the formula, express and Two sets of user tags and Similarity between them Indicates user The set of tags and users The number of elements in the intersection of the tag sets. Indicates user The set of tags and users The number of elements in the union of the tag sets. The larger the value, the better the user With users The more similar the attribute tags between them; (402) Calculate the level of attention between users based on All the information followed by this user To build a social network, the formula for user attention is used to represent the information users are following in the social network. The defined attention level is as follows: In the formula, Indicates user For users Potential attention, users User One of the users followed by the user they follow; Indicates user Followed users follow again The number of users, Indicates user The maximum number of users that a followed user can follow. The larger the value, the better the user Follow users The higher the probability; (403) Calculate the recommendation rate between users user Recommended users The recommendation level is defined as follows: In the formula, The larger the value, the better the user Recommended users The higher the probability; Step (6) includes the following steps: selecting the accuracy and diversity of the recommendation list as a multi-objective function, initializing the population, and population evolution; The selection of the accuracy and diversity of the recommendation list as a multi-objective function is as follows: make For population size, This represents the total number of iterations. Set the current iteration number to 0. The number of friends recommended to a single user; Regarding accuracy, similarity values ​​are used for combined calculations; for users... From the candidate recommendation list Before being selected The similarity calculation is performed on each user, as detailed below: In the formula, For the recommended list Length, Indicates user With users Recommended friend user list The Middle Recommendation similarity between users; The diversity of the recommendation list is measured by the number of tag attributes, for users. Calculate the friend user recommendation list The number of attribute tags for each user, used as a reflection of diversity, is shown below: In the formula, For the recommended list Length, Represented as user Recommended list The Middle Number of tags per user; The initial population is generated as follows: a random initialization strategy is used to generate the initial population. That is, randomly generated OK The decision matrix of the column is calculated. OK The target matrix of the column; The population evolution includes the following steps: a. Individual choice Randomly select the population Divide into two subpopulations equally and And through density estimates based on transfer. The superiority or inferiority of individuals between two subpopulations is calculated using the following definition: In the formula, For individuals In population The density estimate based on transfer, and For population The The and the first Individual, Indicates population Quantity; when When less than 0, =0, otherwise For individuals and individuals Between in the objective function The Euclidean distance value; For the first One objective function, Describe the objective function Number; assuming subpopulation Compare excellent; b. Generation of offspring Based on subpopulation and Two new populations are generated using crossover and mutation operators. and The two new populations are then merged to obtain the final population. ; c. Environmental selection Population calculations were performed using a non-dominated sorting algorithm and a transition-based density estimation method. The populations are then sorted to obtain candidate populations with multiple frontiers; the top-ranked populations are selected. Individuals are preserved to the next generation, thus renewing the population. Finally, the number of iterations was reduced. Then proceed to step a, and repeat the process until the specified number of iterations is reached. until; d. From the sorted population Select the first individual as the user's optimal recommendation list and output it.

2. The interpretable friend recommendation method based on large language model label classification according to claim 1, characterized in that: The specific steps (1) are as follows: Let... For each user, there is a collection of all text attribute information. ;set up For all the information that each user is interested in, .

3. The interpretable friend recommendation method based on large language model label classification according to claim 2, characterized in that: The specific steps (2) are as follows, assuming there are Individual users and perform digital serialization, targeting A collection of text attribute information for each user Over-segmentation issues have arisen; therefore, a new text attribute lexicon, Lex, is constructed and... A collection of text attribute information for each user The process involves re-segmenting the text, removing duplicate information from user attributes, deleting stop words, and standardizing case, ultimately resulting in an updated set of text attribute information. .

4. The interpretable friend recommendation method based on large language model label classification according to claim 3, characterized in that: In step (3), the classification is specifically divided into seven categories of labels, and the classification information is recorded as follows: , The first A set of tags for each user.

Images