Research On Recommender Algorithms Based On Collaborative Filtering

The popularity of mobile device and the rapid development of the internet have brought great convenience to people's life,users on the internet can easily read news,watch movies,study and shop on various e-commerce websites or apps on mobile devices.However,the explosive internet resources make it more time-consuming and tiresome for users to select their interested content.Therefore,recommender algorithms have received more and more attention and become a research hotspot in academia and industry.Collaborative Filtering is one of the most popular recommender techniques.Such technique aims at analyzing the historical information of users and items,and extracting the corresponding implicit characteristics,then predicts users' preference on items and provides personalized recommendations.Nevertheless,the real world data are always sparse and the distribution of different data sets are also quite different,which poses great challenges of collaborative filtering methods about low prediction accuracy,high time complexity and poor adaptability.And researches about group recommendation and diversityaware recommendation have the space to improve.In view of above,we develop some researches on the difficulties of collaborative filtering technology.Particularly,the research contents and innovations are listed as follows:(1)Aiming at improving the prediction accuracy of recommendations,we propose a Bagging-based matrix factorization model which assigns dynamic weights to every base learner according to the number of users' and items' ratings,then acquires the prediction ratings by weight summation.The experimental results demonstrate our model has the same efficiency as the traditional matrix factorization model,and it is superior to the traditional matrix factorization on all measures.(2)Aiming at improving high-order rating distance model's adaptability and efficiency,we propose an improved high-order rating distance model with omitting rules based on slack variable,in which the static parameter used to balance the first-order rating distance and the second-order rating distance is replaced by a data scale sensitive function.We choose the Newton method to solve the convex recommendation optimization problem defined in this paper instead of stochastic gradient descent.Our model not only achieves the adaptability by eliminating several static parameters for module balancing,reduces the computation complexity,but also accelerates the optimization function convergence speed.Experimental results show the proposed model has good performance in terms of prediction accuracy and efficiency.(3)Aiming at improving the performance of group recommendations,we propose a local optimization framework,using sub-group profiles to compute the item relevance.Such method can capture and remove the bias existed in the traditional group recommendation algorithms in a certain degree and it can also be used to derive single-user recommendation.Experimental analysis for group and personal recommendation show fairly good results,our method consistently outperform several state-of-thearts in efficiency.(4)Aiming at improving the adaptability and efficiency of diversity-aware recommendations,we propose a coverage-based model according to the concepts of user-coverage and users' interest domain defined in this paper.This model is parameter-free and suitable for either implicit data or explicit data.Furthermore,in order to improve the efficiency of model,we design an improved greedy algorithm to achieve user interest domain coverage maximization and also provide solid theoretical proof about performance guarantee on efficiency and recommendation quality.Experimental results demonstrate the superiority of proposed model over the state-of-the-art techniques in terms of item relevance and diversity.In summary,based on the collaborative filtering algorithm and combining the knowledge of machine learning,social network and objective optimization,we propose a series of improved algorithms for individual recommendation,group recommendation and diversity recommendation,and provide theoretical proofs and experimental verifications for the performance of the algorithms.