Search

CN-122019890-A - Collaborative filtering recommendation method and device based on user classification

CN122019890ACN 122019890 ACN122019890 ACN 122019890ACN-122019890-A

Abstract

S1, user grading processing, namely obtaining user data of a target platform, grading users based on preset grading dimensions, and obtaining at least two user grades and grade weights corresponding to each user grade; S2, data preprocessing and feature extraction, namely collecting user behavior data and content data, cleaning the data, constructing a user-content interaction matrix, and extracting user features and content features. According to the scheme, the user grade weight is deeply integrated into similarity calculation and recommendation priority sorting, so that the preferences of high-value users and paid users can be remarkably amplified, high-quality recommendation results which are more accurate and more in line with the identity and the requirements of the users are obtained, and the satisfaction degree, the retention rate and the commercial conversion efficiency of core users are effectively improved.

Inventors

  • ZHENG JIANPING

Assignees

  • 苏州蓝豚互动信息科技有限公司

Dates

Publication Date
20260512
Application Date
20260205

Claims (10)

  1. 1. A collaborative filtering recommendation method based on user grading is characterized by comprising the following steps: S1, user grading processing, namely acquiring user data of a target platform, grading users based on preset grading dimensions, and obtaining at least two user grades and grade weights corresponding to each user grade; S2, data preprocessing and feature extraction, namely collecting user behavior data and content data, after data cleaning, constructing a user-content interaction matrix, and extracting user features and content features; S3, collaborative filtering recommendation calculation based on grading: s31, adjusting similarity calculation among users based on user grade weight to obtain a similarity overload value; S32, screening a similar user set of the target user according to the similarity overload value; S33, extracting a candidate content set from the similar user set, and calculating recommendation priority scores of candidate contents by combining the target user grade weight and the heat attenuation coefficient; s4, cold start and diversity optimization, namely if the target user is a new user, generating an initial recommendation list by adopting a mode of combining content-based recommendation with a hierarchical cold start strategy; performing diversity regulation and control on the candidate content set to control the category distribution and the long tail content ratio of the content in the recommendation list; and S5, generating a recommendation result, namely sorting the candidate contents according to the recommendation priority score, and selecting the content with the preset quantity, which is ranked at the front, as a final recommendation result to push to a target user.
  2. 2. The collaborative filtering recommendation method based on user ratings according to claim 1, wherein the preset ratings dimensions include at least two of a business value dimension, a behavioral activity dimension, and a credit compliance dimension; The user data includes user attribute data, behavior data, and business data.
  3. 3. The collaborative filtering recommendation method based on user ratings according to claim 1, wherein the user ratings include core users, normal users and new users, different user ratings correspond to different ratings weights, and the ratings weights are dynamically adjustable.
  4. 4. The collaborative filtering recommendation method based on user classification according to claim 1, wherein in the step S31, a calculation formula of a similarity overload value is: similarity overload value = base similarity x (target user class weight x 0.6 + user class weight to be matched x 0.4); wherein the base similarity is calculated by pearson correlation coefficient or cosine similarity.
  5. 5. The collaborative filtering recommendation method based on user classification according to claim 1, wherein in the step S33, a calculation formula of a heat attenuation coefficient is: Heat attenuation coefficient=1/(1+α×content cumulative exposure amount); Wherein, alpha is a regulating coefficient, and the value range is [0.001, 0.01].
  6. 6. The collaborative filtering recommendation method based on user classification according to claim 1, wherein the diversity adjustment in step S4 comprises at least one of: Limiting the duty ratio of the content of the same category in the recommendation list not to exceed a preset upper limit; limiting the pay content duty ratio for the core user not to exceed a preset upper limit; the lower limit is preset for ensuring that the content duty ratio of the long tail is not lower for common users.
  7. 7. An apparatus for implementing the collaborative filtering recommendation method based on user ratings of any one of claims 1-6, comprising: the user grading module is used for acquiring user data, grading the users based on preset grading dimensions and outputting user grades and corresponding grade weights; The data processing module is used for collecting and cleaning user behavior data and content data, constructing a user-content interaction matrix and extracting user characteristics and content characteristics; The collaborative filtering calculation module is used for improving a collaborative filtering algorithm based on the user grade weight, calculating a user similarity overload value and calculating a recommendation priority score of candidate contents according to the similar user set and the target user grade weight; the optimizing module is used for executing cold start processing of the new user and carrying out diversity optimizing and controlling on the recommendation list; and the recommendation output module is used for generating and outputting a final recommendation result according to the recommendation priority score ranking.
  8. 8. The collaborative filtering recommendation device based on user classification according to claim 7, wherein the user classification module comprises a data acquisition unit, a classification calculation unit and a weight configuration unit; the data processing module comprises a data cleaning unit, a matrix construction unit and a feature extraction unit.
  9. 9. The collaborative filtering recommendation device based on user ratings according to claim 7, wherein: The collaborative filtering computing module comprises a similarity computing unit, a candidate set extracting unit and a score computing unit; The optimizing module comprises a cold starting unit and a diversity regulating unit.
  10. 10. The collaborative filtering recommendation device based on user ratings of claim 7, further comprising a processor and a memory, the memory storing at least one instruction that is loaded and executed by the processor to implement the method of any one of claims 1 to 6.

Description

Collaborative filtering recommendation method and device based on user classification Technical Field The invention relates to the technical field of Internet information, in particular to a collaborative filtering recommendation method and device based on user classification. Background With the rapid development of internet platforms (such as recruitment, electronic commerce and service platforms), a recommendation system has become a core bridge for connecting users with goods, services or contents, and the performance of the recommendation system directly influences the user experience and the platform business efficiency. Collaborative filtering is used as a classical scheme of a recommendation algorithm, and is used for recommending by analyzing the behavior similarity or the object correlation of a user group, so that the collaborative filtering has the advantages of no dependence on objects, the service attribute, the discoverable potential interest association of the user and the like, and is widely applied to various scenes. However, the existing collaborative filtering recommendation technology still has a number of significant drawbacks in practical applications: First, conventional collaborative filtering algorithms typically treat all users as homogenous objects, ignoring the inherent variability of the user population. In a practical platform, there are obvious grading features for users, such as general users, paid users, high value users, etc. In the prior art, the user grading information is not deeply combined with the recommendation algorithm, so that the personalized requirements of high-value users cannot be met preferentially, the interests of paid users are difficult to be effectively reflected through a recommendation mechanism, and the retention of core users and the commercial value conversion of a platform are influenced. Secondly, traditional collaborative filtering recommendation is easily affected by the 'Martai effect', namely popular content is repeatedly recommended, exposure is higher and higher, and a large number of high-quality long-tail content (such as the posts of the masses, special commodities and emerging services) lack exposure opportunities, so that platform content ecological imbalance is caused, and user interest exploration is limited. And thirdly, under a cold start scene with sparse user behavior data (such as a new user or a user without history behavior), the collaborative filtering algorithm is difficult to accurately calculate the similarity of the users, so that the recommendation accuracy is greatly reduced, and the initial experience and conversion of the new user are affected. In addition, although there is application of user classification in the prior art, the application is limited to authority management or simple tagging, and classification results (such as class weights) are not deeply fused to core calculation links (such as similarity calculation, candidate set screening, score sorting and the like) of a recommendation algorithm, so that the value of a user classification system cannot be fully exerted in a recommendation system. Therefore, a technical scheme capable of fusing user classification and collaborative filtering recommendation depth, not only guaranteeing high-value user demands preferentially, but also relieving the Martai effect and improving the cold start recommendation effect is needed. Disclosure of Invention The invention aims to solve the defects in the prior art and provides a collaborative filtering recommendation method and device based on user classification. In order to solve the background technical problems, the invention adopts the following technical scheme: In a first aspect, the present invention provides a collaborative filtering recommendation method based on user classification, including the following steps: S1, user grading processing, namely acquiring user data of a target platform, grading users based on preset grading dimensions, and obtaining at least two user grades and grade weights corresponding to each user grade; S2, data preprocessing and feature extraction, namely collecting user behavior data and content data, after data cleaning, constructing a user-content interaction matrix, and extracting user features and content features; S3, collaborative filtering recommendation calculation based on grading: s31, adjusting similarity calculation among users based on user grade weight to obtain a similarity overload value; S32, screening a similar user set of the target user according to the similarity overload value; S33, extracting a candidate content set from the similar user set, and calculating recommendation priority scores of candidate contents by combining the target user grade weight and the heat attenuation coefficient; s4, cold start and diversity optimization, namely if the target user is a new user, generating an initial recommendation list by adopting a mode of combining content-based reco