Research On Collaborative Filtering Algorithm Based On Product Time Effect

The development of Internet technology and the continuous updating and iteration of various information terminal devices have promoted the generation and dissemination of a large amount of information.It is so hard for users to find useful information from large amounts of data.Therefore,many e-commerce websites use personalized recommendation technology for users: by using corresponding recommendation algorithms,users 'preferences are calculated from users' historical behavior data,and products they may like are recommended.Although recommendation algorithms can effectively reduce the time users spend on information retrieval and improve the user's retrieval efficiency,many traditional recommendation algorithms still have shortcomings.Except for problems such as cold start and sparse samples,many recommendation algorithms do not take time into account.Therefore,this paper will deeply study the effect of time effects on product popularity,comprehensively analyze the cyclical characteristics of product heat and the characteristics of user interest changes with time,and propose improvements to the original algorithm.The main work of this paper can be divided into the following points:(1)First,this paper deeply researches various traditional recommendation algorithms,including content-based,knowledge-based and association rule-based recommendation algorithms.Through the analysis of various algorithms to compare their respective shortcomings,and then propose the concept of dynamic recommendation,leading to the effect of time effects on the precision of the recommendation algorithm.(2)In this paper,the traditional cosine similarity algorithm is improved for the characteristic that the heat of some products will change periodically with time.The similarity algorithm of the product is optimized by calculating the time effect of the product.At the same time,considering that the heat of some products is not greatly affected by the time change,this paper will fuse the improved part with the original algorithm according to a certain proportion.The experiment proves that using the cosine similarity algorithm based on time effect improves the precision of the system recommendation results.(3)Based on the improved similarity algorithm,this paper also optimizes the traditional collaborative filtering recommendation algorithm.The collaborative filtering algorithm needs to calculate user ratings for different products to obtain user preferences,but this method does not take into account the characteristics of user preferences that change over time.Therefore,this paper proposes a time decay model for user interest.The longer the scoring time is from the current time,the lower its corresponding weight will be.Experiments show that the improved algorithm improves the accuracy of recommendation results to a certain extent.(4)Finally,this paper uses the recommendation engine of mahout to build a recommendation system based on the JFinal framework and demonstrates the recommendation function of the system.