A Method for Constructing a Personalized Recommendation System Based on Multiple Individual Variations

Abstract

This invention discloses a method for constructing a personalized recommendation system based on multiple entity variants, relating to computer software technology. The method involves inputting a user relationship network topology graph—an adjacency matrix—into a multi-target community detection algorithm model based on multiple entity variants to obtain relationship network communities. Similarity processing is then performed on these communities to obtain the desired nearest-neighbor users, and personalized data from these nearest-neighbor users is recommended to the target user. This invention combines multiple entity variants with a multi-target community detection algorithm, solving the problem of single-user recommendation and improving the diversity and efficiency of recommendations.

Description

Technical Field

[0001] This invention relates to computer software technology, and more specifically, to a method for constructing a personalized recommendation system based on multiple individual variants. Background Technology

[0002] With the continuous development of network technology, the amount of information is also growing rapidly. How to extract the information users want from this vast amount of data has become a hot topic in current technological research. Traditional information search methods, such as database indexing, mainly rely on manually set search rules for information filtering. However, this method cannot intuitively reflect users' interests and preferences, nor can it provide personalized services. Therefore, personalized recommendation has emerged as a new information service model. Personalized recommendation refers to the technology of collecting, filtering, and classifying user information based on their interests and preferences, extracting other information that users are interested in, and recommending it to them. Personalized recommendation systems provide personalized information and support services to their customers through social networking sites, e-commerce, and other platforms.

[0003] Personalized recommendation systems typically employ recommendation algorithms based on three main techniques: multi-objective optimization, collaborative filtering, and multi-objective community detection.

[0004] Multi-objective optimization is a technique for solving optimization problems. When optimizing two or more conflicting objectives, it can successfully find a set of solutions. These solutions are optimal for all objectives as much as possible and conform to Pareto optimality theory. For example, in the travel problem, time and cost are two conflicting objectives. When time cost is lower, cost tends to be higher; when time cost is higher, cost tends to be lower. Therefore, there exists an optimal set of solutions for these two objectives, distributed among their optimal solutions. Figure 1 As shown.

[0005] Collaborative filtering is designed for discrete, independent data. It calculates the similarity between user data to identify similar neighborhoods of users with similar interests and hobbies. Based on the interests of other users within these neighborhoods, it recommends potentially interesting related items to the original user.

[0006] Multi-objective community detection is an unsupervised clustering technique for network structure data. Its core is to model the community detection problem as a multi-objective optimization problem, utilizing the information contained in the network topology to extract the community structure from complex networks and construct network partitions. For social network data, it can be abstracted into a complex network topology graph, and clustering can be performed based on the connections between users.

[0007] However, existing recommendation technologies based on collaborative filtering often recommend items or projects that are highly correlated with the user's interests, leading to overly simplistic recommendations and a lack of diversity. Secondly, existing multi-objective community detection technologies suffer from high time overhead, and mutation strategies are prone to generating invalid solutions. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a method for constructing a personalized recommendation system based on multiple individual variants, which can ensure that personalized data recommendations are more diverse and efficient.

[0009] The present invention discloses a method for constructing a personalized recommendation system based on multiple individual variants, comprising the following steps:

[0010] Step 1: Collect user information data, construct a user relationship network topology graph based on the user information data, and construct an adjacency matrix based on the user relationship network topology graph;

[0011] Step 2: Build a multi-target community detection algorithm model based on multiple individual mutations, and input the adjacency matrix to divide the user relationship network into several relationship network communities;

[0012] Step 3: Build a similarity model and analyze the similarity of users within the community based on the relationship network community to obtain the target user's nearest neighbor users;

[0013] Step 4: Recommend the personalized data of the neighboring users to the target user.

[0014] Step two specifically includes,

[0015] The first step is to determine the population size (popSize) and the number of iterations (iterSize).

[0016] The second step is to initialize the encoding of the user relationship network topology to obtain encoded individuals whose number is consistent with the population size popSize.

[0017] The third step is to calculate the fitness value of all the coded individuals.

[0018] Step 4: Sort all the fitness values ​​non-dominantly and calculate the crowding distance to obtain the individual set;

[0019] Step 5: Select the two individuals with the smallest set value as superior individuals, and perform bidirectional crossover recombination on the two superior individuals to obtain two random individuals;

[0020] Step 6: Select the individual with the smallest median among the two excellent individuals as the optimal individual, and perform multi-individual mutation on the optimal individual and two random individuals to obtain mutated individuals;

[0021] Step 7: Determine whether the current iteration number t is less than the iteration number iterSize; if t < iterSize, use the mutated individual as the encoded individual and repeat Steps 3 to 6; until t ≥ iterSize, use the currently obtained mutated individual as the target individual;

[0022] Step 8: Evaluate each solution in the target individual through the performance index function to obtain the target solution.

[0023] Calculate the fitness value of the encoded individual through the following objective function

[0024] ;

[0025] In the formula, is the number of network nodes in the adjacency matrix; The relational network is divided into communities; , are both any natural numbers from 1 to ; is the nd community; is the th community; represents the sum of the degrees of the nodes inside the th community; represents the sum of the degrees of the connections between the nodes inside the th community and the nodes of the th community.

[0026] The performance index function is

[0027] ;

[0028] In the formula, is the number of edges in the entire network; represents that there is an edge between the nd node and the th node in the adjacency matrix; represents that there is no edge; and respectively represent the degrees of the nd node and the th node; represents whether the th node and the th node are partitioned in the same community. If they are in the same community, then It equals 1, otherwise It equals 0.

[0029] The second step specifically includes,

[0030] Generate an individual with the same number of individuals as the population size popSize based on the node information stored in the adjacency matrix; decode each individual and encode it using a label-based vector representation to obtain the encoded individual.

[0031] The multiple variants are specifically:

[0032] Let the optimal individual be... The two random individuals are respectively , ;

[0033] Select a factor of one dimension from the optimal individuals. As a calibration factor;

[0034] The optimal individual and two random individuals are subjected to a mutation function. ;

[0035] In the formula, N is any natural number from 1 to 3; N is a natural number greater than or equal to 1.

[0036] The function In denoted as daughter gene ; ;like ,but ,otherwise ;in, It is any natural number from 1 to N;

[0037] Suppose a gene set The gene set For all daughter genes sum;

[0038] According to the gene set Hezi gene The value of , the optimal individual and two random individuals are mutated.

[0039] The function In The operation rules are as follows:

[0040] When gene set There is only one daughter gene in it. When the value is greater than or equal to 2, the node data of the optimal individual or random individual will be used. Change it to the community label of its neighboring nodes;

[0041] When gene set There are daughter genes When the value is greater than 2, the node data from the optimal individual or random individuals will be used. Assigned value ;

[0042] When gene set There are daughter genes = At that time, the node data of the optimal individual or random individuals will be used. Change it to the community label of its neighboring nodes.

[0043] The similarity model is constructed using the Pearson coefficient similarity method.

[0044] The relationship network community is decoded, and the nearest users of the target user are found based on the Pearson correlation coefficient.

[0045] Beneficial effects

[0046] The advantage of this invention is that it transforms the user relationship network topology. Figure 1 The adjacency matrix is ​​input into a multi-target community detection algorithm model based on multiple volume mutations to obtain relational network communities. Similarity processing is then performed on these communities to obtain the desired nearest-neighbor users, and personalized data from these nearest-neighbor users is recommended to the target user. Compared to traditional recommendation algorithms, this invention combines multiple volume mutations with a multi-target community detection algorithm, solving the problem of single-user recommendations and improving the diversity and efficiency of recommendations. Attached Figure Description

[0047] Figure 1 A schematic diagram of the Pareto front;

[0048] Figure 2 This is a flowchart illustrating the overall construction method of the present invention;

[0049] Figure 3 This is a flowchart of the multi-objective community detection algorithm model based on multiple individual mutations of the present invention;

[0050] Figure 4 This is a schematic diagram of the encoding method used in the construction method of the present invention. Detailed Implementation

[0051] The present invention will be further described below with reference to embodiments, but this does not constitute any limitation on the present invention. Any limited modifications made by any person within the scope of the claims of the present invention are still within the scope of the claims of the present invention.

[0052] See Figures 2-4 The present invention provides a method for constructing a personalized recommendation system based on multiple individual variants, comprising the following steps.

[0053] Step 1: Obtain user information data through the website database server cluster, including user basic attributes, user behavior, user reviews, user consumption and other related information, and construct user profiles.

[0054] Based on user information data, analyze the relationships between users, construct a user relationship network topology graph, and then construct an adjacency matrix based on the user relationship network topology graph. When two users in the user relationship network topology graph share a common edge, it indicates that there is a data connection between the two users, and the corresponding position in the adjacency matrix is ​​1; when there is no edge, it indicates that there is no data connection between the two users, and the corresponding position in the adjacency matrix is ​​0.

[0055] The adjacency matrix in this embodiment is a symmetric matrix, so space efficiency can be improved by compressing the adjacency matrix.

[0056] Step 2: Construct a multi-objective community detection algorithm model based on multiple individual mutations to model the community detection problem as a multi-objective optimization problem. Input the adjacency matrix into the multi-objective community detection algorithm model to divide the user relationship network into several relationship network communities.

[0057] The specific implementation method for dividing relational network communities using a multi-objective community detection algorithm model based on multiple individual mutations is as follows.

[0058] The first step is to determine the number of network nodes N, the population size popSize, and the number of iterations iterSize based on the adjacency matrix.

[0059] The second step involves generating individuals of the same number as the population size `popSize` based on the node information stored in the adjacency matrix. Each individual is then decoded and encoded using a label-based vector representation to obtain the encoded individual, represented as... .in, Represents a node.

[0060] The third step is to introduce two objective functions to calculate the fitness values ​​of all encoded individuals.

[0061] The objective function is,

[0062] .

[0063] In the formula, The number of network nodes in the adjacency matrix; The relationship network is divided into One community; , All are 1 to Any natural number in the range; For the first One community; For the first One community; Indicates the first The sum of node degrees within the community; Indicates the first Community internal nodes and the first The sum of the connectivity of community nodes. KKM reflects the internal connectivity of a community, while RC reflects the external connectivity between communities. min means minimizing the two objective functions.

[0064] The fourth step is to sort all fitness values ​​non-dominantly and calculate crowding distance to obtain the individual set.

[0065] Step 5: Select the two individuals with the smallest individual set value as superior individuals, and perform bidirectional crossover recombination on the two superior individuals to obtain two random individuals.

[0066] Step 6: Select the individual with the smallest value among the two excellent individuals as the optimal individual, and perform multiple mutations on the optimal individual and the two random individuals to obtain mutated individuals.

[0067] In this embodiment, the implementation of multiple body mutations is specifically as follows:

[0068] Let the optimal individual be... The two random individuals are respectively , Select a factor of one dimension from the optimal individuals. As a calibration factor.

[0069] Next, the three selected individuals are subjected to a mutation function. This leads to the discovery of mutated individuals.

[0070] In the above variogram, among which denoted as daughter gene . .like ,but ,otherwise And set a gene set. Gene set For all daughter genes sum.

[0071] Based on the above settings, the mutation function In The operation rules are as follows:

[0072] When gene set There is only one daughter gene in it. When it is greater than or equal to 2, the node data in the optimal individual or random individual is modified to the community label of its adjacent node.

[0073] When the gene set has a sub-gene greater than 2, the node data in the optimal individual or random individual is assigned to .

[0074] When the gene set has a sub-gene = , the node data in the optimal individual or random individual is modified to the community label of its adjacent node.

[0075] Step 7: Determine whether the current iteration number t is less than the iteration number iterSize. If t < iterSize, use the mutant individual as the encoded individual and repeat Steps 3 to 6; until t ≥ iterSize, use the currently obtained mutant individual as the target individual.

[0076] Step 8: Evaluate each solution in the target individual through the performance metric function to obtain the target solution.

[0077] The performance metric function is specifically

[0078] .

[0079] In the formula, is the number of edges in the entire network; represents that there is an edge between the th node and the th node in the adjacency matrix; represents that there is no edge; and respectively represent the degrees of the th node and the th node; represents whether the th node and the th node are partitioned in the same community. If they are in the same community, then is equal to 1, otherwise is equal to 0.

[0080] Since a larger modularity indicates a better community structure, therefore, the solution with the largest Q value is finally selected as the target solution.

[0081] Step 3: Build a similarity model based on the Pearson correlation coefficient method. Decode the relational network communities. The decoded relational network communities will be decomposed into individual communities, such as {P1, P2, P3…, Ps}. Find the nearest neighbors of the target user based on the Pearson correlation coefficient.

[0082] Step 4: Recommend personalized data of nearby users, such as their interests and hobbies, and items they are interested in, to the target user and display them in the web application.

[0083] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention, and these will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A method for constructing a personalized recommendation system based on multiple individual variations, characterized by, It includes the following steps: Step 1: Collect user information data, construct a user relationship network topology graph based on the relationships between users, and construct an adjacency matrix based on the user relationship network topology graph; Specifically, obtain user information data through the website database server cluster, which includes user basic attribute information, user behavior information, user evaluation information, and user consumption information, and form a user portrait; Step 2: Build a multi-objective community detection algorithm model based on multi-agent mutation, input the adjacency matrix, and divide the user relationship network into several relationship network communities; Step 3: Build a similarity model, and analyze the similarity of users within the community based on the relationship network community to obtain the neighbor users of the target user; Step 4: Recommend the personalized data of the neighbor users to the target user; Step 2 specifically includes: The first step: Determine the population size popSize and the number of iterations iterSize; The second step: Initialize the encoding of the user relationship network topology graph to obtain encoded individuals with the same number as the population size popSize; The third step: Calculate the fitness values of all the encoded individuals; The fourth step: Perform non-dominated sorting and crowding distance calculation on all the fitness values to obtain an individual set; The fifth step: Take the two individuals with the smallest values in the individual set as excellent individuals, and perform two-way crossover recombination on the two excellent individuals to obtain two random individuals; The sixth step: Take the one with the smallest value among the two excellent individuals as the optimal individual, and perform multi-agent mutation on the optimal individual and the two random individuals to obtain mutant individuals; The seventh step: Judge whether the current iteration number t is less than the iteration number iterSize; if t < iterSize, use the mutant individuals as the encoded individuals, and repeat the third step to the sixth step; until t ≥ iterSize, take the currently obtained mutant individuals as the target individuals; The eighth step: Evaluate each solution in the target individuals through a performance metric function to obtain the target solution.

2. The method of constructing a personalized recommendation system based on multiple body variations according to claim 1, wherein, Calculate the fitness value of the encoded individuals through the following objective function ； In the formula, The number of network nodes in the adjacency matrix; The relationship network is divided into One community; , All are 1 to Any natural number in the range; For the first One community; For the first One community; Indicates the first The sum of node degrees within the community; Indicates the first Community internal nodes and the first The sum of the connectivity of community nodes; In the adjacency matrix, the th Node and the The node has edges; This indicates that no edge exists.

3. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 2, characterized in that, The performance metric function is ； In the formula, Let be the number of edges in the entire network; and They represent the first Node and the The degree of a node; Indicates the first The node and the first Are the nodes assigned to the same community? If they are in the same community, then... It equals 1, otherwise It equals 0.

4. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 1, characterized in that, Step 2 specifically includes: Generate individuals with the same number as the population size popSize according to the node information stored in the adjacency matrix; decode each individual to encode the individual based on the label vector representation method to obtain encoded individuals.

5. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 1, characterized in that, The multi-agent mutation is specifically Let the optimal individual be... The two random individuals are respectively , ; Select a factor of one dimension from the optimal individuals. As a calibration factor; The optimal individual and two random individuals are subjected to a mutation function. ; In the formula, N is any natural number from 1 to 3; N is a natural number greater than or equal to 1.

6. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 5, characterized in that, The function In denoted as daughter gene ; ;like ,but ,otherwise ;in, It is any natural number from 1 to N; Suppose a gene set The gene set For all daughter genes sum; According to the gene set Hezi gene The value of , the optimal individual and two random individuals are mutated.

7. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 6, characterized in that, The function In The operation rules are as follows: When gene set There is only one daughter gene in it. When the value is greater than or equal to 2, the node data of the optimal individual or random individual will be used. Change it to the community label of its neighboring nodes; When gene set There are daughter genes When the value is greater than 2, the node data from the optimal individual or random individuals will be used. Assigned value ; When gene set There are daughter genes = At that time, the node data of the optimal individual or random individuals will be used. Change it to the community label of its neighboring nodes.

8. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 1, characterized in that, The similarity model is built by the Pearson correlation coefficient similarity method.

9. The method for constructing a personalized recommendation system based on multiple individual variants according to claim 8, characterized in that, Decode the relationship network community, and find the neighbor users of the target user according to the Pearson correlation coefficient.

Images