Product recommendation method based on personality characteristics and FCM clustering optimization

Abstract

The invention discloses a product recommendation method based on personality features and FCM clustering optimization. The method comprises the steps that 1, a user and article set and a user big five personality feature score set are obtained; 2, clustering is carried out based on big five personality data of the user, and an optimal clustering center and an optimal membership matrix are obtained; 3, complementing the missing value of the user-article scoring matrix based on user personality feature clustering, so as to calculate the user similarity; and 4, calculating the preference similarity between the user and the target user according to the user-article scoring matrix, calculating the personality similarity between the user and the target user according to the personality characteristics of the user, and carrying out weighted operation to obtain the user similarity so as to calculate the predicted score of the user. According to the method, the cold start problem, the data sparsity problem and the interest drifting problem of user interests along with time change in a recommendation system can be effectively solved, so that the individuation, accuracy and reliability of product recommendation can be improved, and higher conversion rate and user stickiness are brought to a sales platform.

Description

Technical Field

[0001] The present invention belongs to the field of recommendation systems, and specifically relates to a product recommendation method optimized based on personality traits and FCM clustering. Background Art

[0002] With the rapid development of the market and the transformation of the economic structure, various products emerge in an endless stream, and the information choices faced by users become increasingly complex. Users cannot comprehensively understand all available products, which may lead to greater decision-making risks for users when choosing to purchase products due to the lack of necessary information. By using algorithms to predict users' interest preferences, personalized information and services can be provided to users. Through personalized recommendations, the system can recommend the most suitable products for users according to their preferences and historical behaviors, thereby improving user satisfaction and usage experience, and at the same time bringing higher revenues and user stickiness to the sales platform. However, recommendation systems also face many challenges, such as the cold start problem, data sparsity problem, dynamic changes in user interests, and the accuracy and diversity of recommendation results. To address these challenges, researchers have continuously explored new technologies and methods, such as combining user behavior, social relationships, context information, and psychological characteristics, to improve the performance of recommendation systems and user experience.

[0003] However, research shows that there are still many problems in recommendation systems. For example, the cold start problem mainly refers to the situation where there is insufficient historical data of users or items. In addition, the rating matrix between users and items is often sparse, that is, the number of ratings given by users to items is small. This makes it difficult for recommendation systems based on algorithms such as collaborative filtering to capture users' true interests and preferences, thus affecting the recommendation effect. Users' interests and preferences also change dynamically. Many traditional recommendation algorithms assume that users' preferences are fixed and do not change over time. However, users' behaviors may change over time and in different contexts, which may cause the recommendation method to become ineffective after a period of time. Summary of the Invention

[0004] The present invention is proposed to solve the above-mentioned deficiencies of the prior art, and provides a product recommendation method optimized based on personality traits and FCM clustering, aiming to effectively solve the cold start problem, data sparsity problem, and interest drift problem of users' interests changing over time in the recommendation system, so as to improve the personalization, accuracy, and reliability of product recommendations, and bring higher conversion rates and user stickiness to the sales platform.

[0005] To achieve the above-mentioned invention purpose, the present invention adopts the following technical solutions:

[0006] The characteristics of a product recommendation method optimized based on personality traits and FCM clustering of the present invention are as follows: the method is carried out according to the following steps:

[0007] Step 1. Obtain the user set U = , item set , where represents the k-th user, represents the -th item, n represents the total number of users, and m represents the total number of items;

[0008] Let the k-th user rate the -th item as R; R represents the rating matrix with dimensions m×n. If = 0, it means the k-th user has no rating for the -th item ;

[0009] Let the k-th user have the normalized personality trait score set , where represents the normalized u-th personality trait score of the k-th user, , and represents the total number of personality traits of each user;

[0010] Step 2: Cluster the user personality traits to obtain the optimal cluster centers and the optimal membership matrix ;

[0011] Step 3: Based on and , construct the completed rating matrix for calculating the total similarity between users;

[0012] Step 4: Use Equation (11) to obtain the predicted score of the k-th user for the -th item :

[0013] (11)

[0014] In Equation (11), represents the predicted score of the k-th user for the -th item , , respectively represent the average scores of the k-th user and the q-th user , represents the q-th user rating for the -th item The preprocessed score, where N represents the k-th user Among all users with the highest membership degree in the optimal membership matrix and with membership degrees greater than θ in the cluster with the highest membership degree, where θ is a boundary value is the k-th user and the q-th user The total similarity between them

[0015] Another feature of the product recommendation method based on personality traits and FCM clustering optimization according to the present invention is that step two is carried out as follows

[0016] Step 2.1: Define the number of current cluster centers as j and initialize j = 1

[0017] Let the j-th cluster center = ;

[0018] Step 2.2: Calculate the distance between the normalized personality trait score set of the k-th user and the cluster center of the j-th iteration, and use Equation (1) to obtain the selection probability of the k-th user , thereby obtaining the selection probabilities of n - 1 users

[0019] (1)

[0020] Step 2.3: Randomly select a user from the selection probabilities of n - 1 users as the cluster center for the (j + 1)-th iteration ;

[0021] Step 2.4: After assigning j + 1 to j, return to step 2.2 and execute sequentially until j = p, thereby obtaining the cluster center set V = ;

[0022] Step 2.5: Use Equation (2) to obtain the membership degree value of the personality trait score set of the k-th user belonging to the j-th cluster center , thereby obtaining the membership matrix B

[0023] (2)

[0024] In Equation (2), s is the fuzzy index is the h-th cluster center

[0025] Step 2.6: Define the current iteration number as t and initialize t = 1;

[0026] Step 2.7: Let the membership degree value belonging to the j-th cluster center in the t-th iteration be = ; the j-th cluster center in the t-th iteration be = ;

[0027] Obtain the objective function in the t-th iteration using Equation (3) :

[0028] (3)

[0029] Step 2.8: Use Equation (4) and Equation (5) to obtain the membership degree value belonging to the j-th cluster center in the (t + 1)-th iteration and the j-th cluster center in the (t + 1)-th iteration, so as to obtain the membership degree matrix in the (t + 1)-th iteration and the cluster center set : :

[0030] (4)

[0031] (5)

[0032] Step 2.9: Use Equation (3) to obtain the objective function in the (t + 1)-th iteration ;

[0033] Step 2.10: Judge whether holds. If it holds, then take as the optimal cluster center , take as the optimal membership degree matrix , where represents the optimal membership degree value belonging to the j-th cluster center ; is the threshold value.

[0034] Furthermore, Step 3 is carried out as follows:

[0035] Step 3.1 Assume = 0, and the optimal cluster center of the cluster where is located is , the corresponding optimal membership degree is ; then calculate the k-th user after completion For the th item rating × ; The preprocessed rating matrix is composed of all the ratings after completion and the original ratings ;

[0036] Step 3.2: Calculate the personality similarity between the k-th user and the q-th user using Equation (6): :

[0037] In Equation (6), represents the normalized u-th personality trait score of the q-th user , , respectively represent the average scores of the personality traits of the k-th user and the q-th user ;

[0038] Step 3.3: Calculate the preference similarity between the k-th user and the q-th user using Equation (7): :

[0039] (7)

[0040] In Equation (7), represents the preprocessed rating of the k-th user for the th item , represents the preprocessed rating of the q-th user for the th item , and respectively represent the average preprocessed ratings of all items for the k-th user and the q-th user ; represents the time difference penalty factor between the midpoint t of the rating times of the k-th user and the q-th user and the current time , and there is:

[0041] (8)

[0042] In Equation (8), is the current time, and t is the k-th user and the q-th user the midpoint of the rating time; representing the k-th user and the q-th user the time penalty parameter, and there is:

[0043] (9)

[0044] In formula (9), is a random parameter;

[0045] Step 3.4, obtain the total similarity between the k-th user and the q-th user by using formula (10): :

[0046] (10)

[0047] In formula (10), α is a weight coefficient, α ∈ [0, 1].

[0048] An electronic device of the present invention includes a memory and a processor, characterized in that the memory is used to store a program for supporting the processor to execute the product recommendation method, and the processor is configured to execute the program stored in the memory.

[0049] A computer-readable storage medium of the present invention, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the product recommendation method when being run by a processor.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] 1. The present invention uses the FCM clustering algorithm to calculate user similarity and adopts a more effective density-based initialization method to select the initial clustering centers, avoiding the poor clustering effect that may be caused by random initialization, significantly improving the efficiency and accuracy of the FCM clustering algorithm, so that more accurate personalized recommendation content can be obtained more quickly in the item recommendation scenario, reducing the user's decision-making time, improving the user satisfaction and shopping experience, and at the same time bringing higher conversion rates and user stickiness to the sales platform.

[0052] 2. The present invention clusters users based on the user personality trait parameters of the Big Five personality model, effectively solving the cold start problem of the recommendation system. By using the BFI-2 scale questionnaire to obtain the personality trait information of all users, users without any purchase or click data can also apply this recommendation model, thus accelerating the user adaptation and participation process and bringing a better user experience to new users of the sales platform.

[0053] 3. The present invention completes the missing values of the user-item rating matrix through a clustering algorithm based on the Big Five personality, and completes the sparse rating matrix through the nearest neighbor rating mean to form a dense user-item rating matrix, solving the sparsity problem of missing rating data in the recommendation system, further improving the accuracy of user similarity calculation, so that the recommendation system can more comprehensively reflect the true interests and preferences of users, improving the diversity and personalization of recommendation results, and effectively avoiding the recommendation deviation problem caused by data sparsity.

[0054] 4. The present invention performs hybrid recommendation by combining the similarity of users and user ratings, improving the overall performance of the recommendation system, being able to consider both the personality traits and interest preferences of users, more comprehensively and accurately depicting the similarity relationship between users, improving the personalization and diversity of recommended content, effectively balancing the accuracy and novelty of recommendation results, and providing users with a more satisfactory product recommendation experience.

[0055] 5. The present invention introduces a time penalty model based on outlier mitigation, introducing an exponential penalty time factor in the process of similarity calculation, weighting the historical behavior and rating data of users, solving the problem of interest drift in the recommendation system where users' interests and preferences change over time, so as to better reflect the current interests of users, further improving the response ability of the recommendation system to users' real-time needs, enhancing the timeliness and relevance of recommendation results, and at the same time taking into account the abnormal fluctuations of users' behaviors on the basis of maintaining personalized recommendations, improving the stability and reliability of the recommendation system. Brief Description of the Drawings

[0056] Figure 1 is a flowchart of the method of the present invention. Detailed Embodiments

[0057] In this embodiment, a product recommendation method based on the Big Five personality and an optimized FCM clustering algorithm is to construct a user simulation matrix based on the Big Five personality characteristics, obtain the behavioral characteristics of users' historical ratings, purchases, and clicks on products on the sales platform, construct a user-item interaction matrix, and then complete the missing values of the user-item interaction matrix. The user information matrix and the user-item interaction matrix obtained based on the above process are evaluated for similarity based on the FCM clustering algorithm. The FCM clustering algorithm uses a distance estimation method to select the initial clustering center, improving the stability and accuracy of clustering. After obtaining the top N similar users, product recommendations are made. In addition, users' interests change over time, so a time factor is introduced in the clustering process to construct a time decay model, weighting the historical behavior and rating data of users, thereby improving the accuracy and personalization of product recommendations. Specifically, the method includes the following steps:

[0058] Step 1: First, extract and analyze user behavior characteristics. The extraction and analysis of user behavior characteristics are key steps in the recommendation system, aiming to mine valuable information from the user's historical behavior data for predicting the user's future preferences. Obtain the user set U = and the item set , where represents the k-th user, represents the -th item, n represents the total number of users, and m represents the total number of items;

[0059] Let the score of the k-th user for the -th item be ∈R; R represents the rating matrix with dimensions m×n. If = 0, it means that the k-th user has no rating for the -th item ;

[0060] The Big Five personality model is a widely accepted personality psychology model used to describe an individual's personality traits in five main dimensions. These five dimensions are openness, conscientiousness, extraversion, agreeableness, and neuroticism. Use the BFI-2 scale to collect users' personality trait data. The BFI-2 scale contains 5 main dimensions (openness, conscientiousness, extraversion, agreeableness, and neuroticism) and 15 sub-dimensions. For each user, calculate the scores of each dimension and sub-dimension according to their answers on the BFI-2 scale. Let the normalized personality trait score set of the k-th user be , where represents the normalized u-th personality trait score of the k-th user, and represents the total number of each user's personality traits.

[0061] Step 2: Cluster based on user personality characteristics:

[0062] The FCM fuzzy clustering algorithm uses fuzzy concepts to classify users by calculating the membership degree of users to clustering categories. However, the traditional FCM algorithm largely depends on the selection of the initial clustering center and is prone to falling into local optimum, thus affecting the clustering effect of the algorithm. In the present invention, the process of randomly initializing the membership matrix in the FCM algorithm is optimized using a density-based initialization method to make the initial clustering centers as scattered as possible, significantly improving the efficiency and accuracy of the FCM clustering algorithm. Therefore, in the item recommendation scenario, more accurate personalized recommendation content can be obtained more quickly, reducing the user's decision-making time, improving the user's satisfaction and shopping experience, and at the same time bringing higher conversion rates and user stickiness to the sales platform.​

[0063] Step 2.1: Define the number of current cluster centers as j and initialize j = 1;

[0064] Let the j-th cluster center = ;

[0065] Step 2.2: Calculate the distance between the normalized personality trait score set of the k-th user and the cluster center of the j-th iteration , and use Equation (1) to obtain the selection probability of the k-th user , thereby obtaining the selection probabilities of n - 1 users;

[0066]

[0067] (1)

[0067] Step 2.3: Randomly select a user from the selection probabilities of n - 1 users as the cluster center of the (j + 1)-th iteration ;

[0068] Step 2.4: After assigning j + 1 to j, return to Step 2.2 and execute sequentially until j = p, thereby obtaining the cluster center set V = ;

[0069] Step 2.5: Use Equation (2) to obtain the membership value of the feature score set of the k-th user belonging to the j-th cluster center , thereby obtaining the membership matrix B:

[0070] (2)

[0071] In Equation (2), s is the fuzzy index, is the h-th cluster center;;

[0072] Step 2.6: Define the current iteration number as t and initialize t = 1;

[0073] Step 2.7: Let the membership value of the k-th user belonging to the j-th cluster center = ; The j-th cluster center of the t-th iteration=

[0074] Use Equation (3) to obtain the objective function of the t-th iteration:

[0075] (3)

[0076] Step 2.8: Obtain the membership degree value belonging to the j-th cluster center at the (t + 1)-th iteration and the j-th cluster center at the (t + 1)-th iteration by using Equation (4) and Equation (5), so as to obtain the membership degree matrix and the cluster center set at the (t + 1)-th iteration: belonging to the j-th cluster center of the membership degree value and the j-th cluster center at the (t + 1)-th iteration , thereby obtaining the membership degree matrix at the (t + 1)-th iteration and the cluster center set :

[0077] (4)

[0078] (5)

[0079] Step 2.9: Obtain the objective function at the (t + 1)-th iteration by using Equation (3); ;

[0080] Step 2.10: Judge whether holds. If it holds, then take as the optimal cluster center, take as the optimal membership degree matrix, where represents the optimal membership degree value belonging to the j-th cluster center; is the threshold; whether it holds. If it holds, then take as the optimal cluster center , take as the optimal membership degree matrix , where represents belonging to the j-th cluster center of the optimal membership degree value; is the threshold;

[0081] Step 3: Construct the completed scoring matrix , which is used to calculate the personality similarity between users:

[0082] There are missing values in the user-item scoring matrix R defined in Step 1. For each missing value in the user-item scoring matrix, find other users in the cluster group to which the user belongs after the Big Five personality clustering, calculate the average score of all users who rate the item in the cluster, and fill the missing value in the user-item scoring matrix:

[0083]

[0084] For each missing value in the user-item scoring matrix, fill the missing value in the user-item scoring matrix based on the mean value of the nearest neighbor ratings of the user obtained in Step 1. Output the dense user-item scoring matrix , it solves the sparsity problem of missing rating data in the recommendation system, further improves the accuracy of user similarity calculation, enables the recommendation system to more comprehensively reflect users' true interests and preferences, enhances the diversity and personalization of recommendation results, and effectively avoids the recommendation bias problem caused by data sparsity.

[0085]

[0086] Step 3.1 Assume = 0, and the optimal cluster center of the cluster where is and the corresponding optimal membership degree is ; then calculate the rating of the completed k-th user for the m-th item × ; The preprocessed rating matrix is composed of all the completed ratings and the original ratings.

[0087] Step 3.2: The present invention adopts the Pearson similarity calculation method of the fusion model to perform weighted operations on the personality similarity and preference similarity of users. First, it can be carried out by combining two dimensions of user personality similarity and preference similarity, more comprehensively and accurately depicting the similarity relationship between users, improving the personalization degree and diversity of recommended content, and effectively balancing the accuracy and novelty of recommendation results; Second, the calculation based on user personality similarity does not require users to make purchases and ratings on the sales platform, enabling new users without any purchase or click data to also apply this recommendation model, thereby accelerating the adaptation and participation process of users and bringing a better user experience to new users of the sales platform.

[0088] Calculate the personality similarity between the k-th user and the q-th user using Equation (6):

[0089] (6)

[0090] In Equation (6), represents the normalized u-th personality feature score of the q-th user , and , respectively represent the average scores of the personality features of the k-th user and the q-th user ;

[0091] Step 3.3: Calculate the k-th user and the q-th user Preference similarity between :

[0092] (7)

[0093] In formula (7), represents the preprocessed score of the k-th user for the th item , and represents the preprocessed score of the q-th user for the th item . During the similarity calculation process, an exponential penalty time factor is introduced to weight the historical behavior and rating data of users, solving the problem of interest drift in the recommendation system where users' interests and preferences change over time. Additionally, abnormal rating values can more clearly reflect users' preferences. In response to this situation, an outlier mitigation is added to abnormal ratings, using to increase the abnormal rating and reduce the time penalty ratio and respectively represent the average preprocessed scores of all items by the k-th user and the q-th user ; represents the time difference penalty factor between the midpoint t of the rating times of the k-th user and the q-th user and the current time , and there is:

[0094] (8)

[0095] In formula (8), is the current time, t is the midpoint of the rating times of the k-th user and the q-th user ; represents the time penalty parameter of the k-th user and the q-th user , and there is:

[0096] (9)

[0097] In formula (9), is a random parameter;

[0098] Step 3.4. Obtain the total similarity between the k-th user and the q-th user using formula (10):

[0099] (10)

[0100] In formula (10), α is a weight coefficient, and α ∈ [0, 1];

[0101] Step 4: Obtain the predicted user scores calculated based on this model, perform TopN recommended item recommendations based on the score ranking, and use formula (11) to obtain the k-th user For the th item predicted score :

[0102] (11)

[0103] In formula (11), represents the predicted score of user for item , , respectively represent average scores, represents the preprocessed score of user for item , N represents the set of all users whose membership degrees in the optimal membership matrix are greater than θ for the cluster in which user has the highest membership degree, and θ is a boundary value.

[0104] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0105] In this embodiment, a computer-readable storage medium stores a computer program, and when the computer program is run by a processor, it executes the steps of the above method.

[0106] In summary, the method of the present invention combines the Big Five personality traits with an optimized FCM clustering algorithm, effectively solves the cold start problem, data sparsity problem, and interest drift problem, thereby providing an accurate user personalized recommendation strategy and an efficient product recommendation mechanism for e-commerce platforms, significantly improving the accuracy and recall rate of product recommendations, and enhancing the user experience and platform stickiness.

Claims

1. A product recommendation method based on personality traits and FCM clustering optimization, characterized in that, It is carried out according to the following steps: Step 1. Obtain the user set U = and the item set , where represents the k-th user represents the -th item, n represents the total number of users, and m represents the total number of items; Let the k-th user rate the i-th item as ∈ R; R represents the rating matrix of dimension m × n. If = 0, it means that the k-th user has not rated the i-th item; Let the k-th user The set of normalized personality trait scores , where represents the k-th user The normalized score of the u-th personality trait, represents the total number of personality traits of each user; Step 2. Cluster the user personality traits to obtain the optimal clustering center and the optimal membership degree matrix ; Step 3. Based on and , construct the completed rating matrix , which is used to calculate the total similarity between users; Step 4: Obtain the k-th user using Equation (11) For the i-th item's predicted score : (11) In formula (11), represents the k-th user for the th item predicted score, , respectively represent the average scores of the k-th user and the q-th user , represents the preprocessed score of the q-th user for the th item , N represents all users in the cluster with the highest membership degree of the k-th user in the optimal membership degree matrix whose membership degrees are greater than θ, where θ is a boundary value, is the k-th user and the q-th user total similarity between.

2. The product recommendation method based on personality characteristics and FCM clustering optimization according to claim 1, wherein, The second step is carried out according to the following steps: Step 2.1: Define the number of current clustering centers as j, and initialize j = 1; Let the j-th cluster center = ; Step 2.2: Calculate the k-th user The normalized personality trait score set and the cluster center of the j-th iteration distance , and use Equation (1) to obtain the selection probability of the k-th user , so as to obtain the selection probabilities of n - 1 users; (1) Step 2.3: Randomly select a user from the selection probabilities of n - 1 users as the clustering center for the (j + 1)-th iteration ; Step 2.4: After assigning j + 1 to j, return to Step 2.2 and execute sequentially until j = p, so as to obtain a set of cluster centers V composed of p cluster centers ; Step 2.

5. Obtain the personality trait score set of the k-th user and the membership degree value belonging to the j-th cluster center so as to obtain the membership degree matrix B: ​ (2) In formula (2), s is the fuzzy exponent; is the h-th clustering center; Step 2.6: Define the current iteration number as t, and initialize t = 1; Step 2.7, let at the t-th iteration belong to the j-th cluster center membership value = ; the j-th cluster center at the t-th iteration = ; The objective function at the t-th iteration is obtained using Equation (3). : (3) Step 2.8, obtaining the membership degree value belonging to the j-th cluster center at the (t + 1)-th iteration and the j-th cluster center at the (t + 1)-th iteration , so as to obtain the membership degree matrix and the cluster center set at the (t + 1)-th iteration: (4) (5) Step 2.9: Obtain the objective function at the (t + 1)-th iteration using Equation (3). ; Step 2.10, determine whether it holds. If it holds, then take as the optimal clustering center , and take as the optimal membership degree matrix . Among them, represents the optimal membership degree value belonging to the j-th clustering center ; is the threshold value.

3. The product recommendation method based on personality characteristics and FCM clustering optimization according to claim 2, wherein, The third step is carried out according to the following steps: Step 3.1 Assume = 0, and the optimal cluster center of the cluster where is and the corresponding optimal membership degree is ; then calculate the score of the k-th user after completion for the -th item × ; The preprocessed score matrix is composed of all the scores after completion and the original scores ; Step 3.2: Calculate the personality similarity between the k-th user and the q-th user using Equation (6). : In formula (6), represents the q-th user The normalized personality feature score of the u-th person,[[]] , respectively represent the k-th user and the q-th user The average score of the personality traits; Step 3.3: Calculate the preference similarity between the k-th user and the q-th user using Equation (7). : (7) In formula (7), represents the k-th user for the item after preprocessing, represents the q-th user for the item after preprocessing, and respectively represent the average scores after preprocessing for all items by the k-th user and the q-th user ; represents the time difference penalty factor between the midpoint t of the scoring times of the k-th user and the q-th user and the current time , and there is: (8) In formula (8), is the current time, and t is the midpoint of the rating times of the k-th user and the q-th user ; represents the time penalty parameter of the k-th user and the q-th user , and there is: (9) In formula (9), is a random parameter; Step 3.

4. Obtain the total similarity between the k-th user and the q-th user using Equation (10): ​ (10) In formula (10), α is a weight coefficient, and α ∈ [0, 1].

4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program for supporting the processor to execute any one of the product recommendation methods in claims 1-3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it executes the steps of any one of the product recommendation methods in claims 1-3.

Images