Research On Mature User Hybrid Collaborative Filtering Algorithm Fusion Matrix Decomposition

Nowadays,the Internet has become a part of people's daily life,and people can't do without the help of recommendation systems behind the convenience brought by the Internet.However,as the user's usage time on the platform increases,the user's historical data is also increasing.Compared with the need to focus on solving the cold start problem for new users in the early days of the Internet,how to make good use of the existing data to maintain the existing maturity Users are more important.Aiming at this problem,this paper firstly introduces the research status and development trend of traditional recommendation algorithms,and then introduces the commonly used recommendation algorithms and their improvements in detail.Mature users accumulate more data on the platform,and it is more difficult to find users with similar preferences.Therefore,this paper chooses an item-based collaborative filtering recommendation algorithm that can recommend according to user preferences.Time and other factors affect the problem,and a single traditional algorithm is not suitable for today's data of more than one million.This paper improves and studies the traditional collaborative filtering algorithm for the above problems.The specific work is as follows:A collaborative filtering recommendation algorithm with added penalty hot item and time decay(IFFF-ICF)is proposed.The model is inspired by Newton's cooling formula to design a new time decay function,which reduces the score to extreme scores while decaying according to time.It can reflect the real evaluation of users more accurately in the face of long-term data,and at the same time,in order to avoid the interference of popular items when calculating item similarity recommended accuracy.Considering that the accuracy of the item-based collaborative filtering recommendation algorithm is very dependent on the judgment of the user's favorite items,and a single recent rating data or only considering the items with higher ratings cannot accurately capture the user's interest,this paper uses the time window to classify the user's interest.It is divided into two parts: short-term interest and longterm interest.The user's real interest set is found by synthesizing the short-term interest obtained from the interest changes in the user's recent behavior and the longterm interest trend reflected in the historical score.Combining the user interest model with the improved collaborative filtering algorithm,a collaborative filtering algorithm(IFFF-ICF-Interest algorithm)based on the user interest model that adds penalty hot items and time decay is obtained.The improved collaborative filtering algorithm has the problem of insufficient efficiency in the face of today's exponentially increasing data volume.Therefore,this paper proposes an item-based collaborative filtering algorithm(FMT-ICF)model fused with matrix factorization as a recall module,which reduces the number of comparisons in recommendation ranking by making full use of the background information of users and items.The model uses a factorization machine for recall,which can fully discover the interactive information in the face of extremely sparse variables such as user information.The method extracts item latent features as feature variables of items.Integrating the above two algorithms can well combine the background information and rating information of users and items,reduce the amount of calculation of collaborative filtering items,so as to find the content that each user is really interested in more accurately,and greatly improve the collaborative filtering algorithm.efficiency and accuracy.Finally,this paper conducts experiments on Movie Lens 100 k dataset and proves the effectiveness of the algorithm proposed in this paper.