Research On Hybrid Collaborative Filtering Algorithm Based On Integrated User's Interest Preference

The personalized recommendation system is mainly aimed at helping users obtain their potential and vague requirements.When the user cannot accurately describe the keywords of his desired item,it is difficult for the user to obtain the required items in the massive data information of the Internet.The recommendation system can just make up for this shortcoming,based on the existing data modeling to actively provide users with recommendations.Collaborative filtering technology is not only a simple algorithm,but also has a high accuracy,so it is the most commonly used recommendation technology.However,in real life,there is a general phenomenon,namely the long-tail theory,especially in e-commerce.In the user-item scoring system,most scoring data are concentrated in a small number of users and items,which will seriously affect recommendation quality.This paper focuses on the distribution of scores of users and items in collaborative filtering algorithms and proposes two improved algorithms.First,when calculating user similarity,the user's interest preference is introduced,and when looking for nearest neighbors,a distribution search is used,ie,two similarity thresholds are set,and nearest neighbors are selected according to the threshold,thereby improving the problem of sparse user rating data.Then,in the scoring prediction,the scores of some items are pre-filled to make up for the recommendation error caused by the item score data.The specific content of the algorithm is as follows:(1)For the problem of uneven distribution of user rating data,a neighbor selection collaborative filtering algorithm that combines user interest preferences is proposed.Firstly,according to the label of the user rating item,the TF-IDF algorithm is used to calculate the user's interest degree for each label,so as to obtain the user's interest preference,and further to obtain the user's similarity in interest;then,by setting a similarity factor,the user is calculated.Comprehensive similarities in interest and ratings;Finally,set appropriate thresholds based on user similarity and similarity of scores,and select the user whose most similarity is greater than the set value and whose similarity is greater than the set value.Strictly neighboring user's rating data is used for rating prediction to achieve recommendation.(2)For the case that most items only have a few ratings,a hybrid collaborative filtering algorithm based on long tail items is proposed.In this method,users are divided into active users and inactive users according to the scores of users and items.The item is divided into regular items and long-tail items;then the neighbor selection collaborative filtering algorithm is used to integrate the user interest preferences for regular items.For the score forecasting,for the long-tailed item,the long-tailed item is predicted by the active user first,and the score is used to fill in the missing values in the score matrix,and then the score of the long-tail item is predicted;finally,the regular item and the long tail item are merged.The score is recommended.Finally,the experiment analysis and comparison of the above algorithms are carried out in this paper.