Research On Time Weighted Recommendation Algorithm Based On User Features And Item Tags

Recommendation system is the main method to solve the problem of "information overload" on the Internet,of which collaborative filtering recommendation algorithm is very successful technology.Collaborative filtering algorithm is very useful,but it also faces many challenges: for example,how to accurately mine users’ interests in sparse data,how to solve the problem of users’ interests drifting with time,and how to solve the problem of cold start when new users enter the system.These problems affect the recommendation algorithm to generate high-quality recommendation results and reduce users’ satisfaction with the recommendation system.Therefore,aiming at these problems,this thesis studies and improves the collaborative filtering recommendation algorithm.The specific contents are as follows:Aiming at the problem that traditional collaborative filtering algorithms do not fully consider user interest drifting,this thesis proposes ITW-CF.The algorithm considering the influence of time factors on the value of user scoring data,improves the time weighting function by using the time decay function and recent item similarity,acts on the user-item scoring matrix,and affects the process of user similarity calculation and scoring prediction.The experimental results show that ITW-CF is better than the traditional algorithm in recommendation accuracy and alleviates the influence of interest drifting on recommendation effect to some extent.Aiming at the cold start problem of users in traditional collaborative filtering algorithms,this thesis proposes CF-UFC.The algorithm takes user activity and scoring diversity as criteria,selects a plurality of expert users from the user set as initial center points of the K-means algorithm,clusters users according to user feature vectors and obtains user feature similarity,searches for similar neighbors from the clustering group where users are located,and comprehensively considers the relationship between user features and user scoring in combination with user feature similarity and user scoring similarity.The experimental results show that CF-UFC can alleviate the problem of users’ cold start to some extent.Aiming at data sparsity problem of traditional collaborative filtering algorithms,this thesis proposes CF-IT.The algorithm extracts item tags from items above the average score of users,calculates and constructs a user-tag interest matrix using TF-IDF algorithm,analyzes the interest degree of users in tags,and improves the similarity calculation strategy based on user rating and tag interest.The experimental results show that CF-IT can alleviate the problem of poor performance of recommendation algorithm caused by data sparsity to some extent.Finally,the above three methods are combined to effectively combine the advantages of the three algorithms,this thesis proposes TWH-UFIT.The experimental results show thatTWH-UFIT is better than the single algorithm in accuracy,and can predict the user score more accurately and improve the recommendation quality.