Research On Personalized Recommendation Algorithm Based On Collaborative Filtering

Complex and diverse information is filled with the entire network because of the rapid development of the Internet and information technology,which makes information overload problems.Therefore,personalized recommendation systems which utilize users' information to recommend interesting items are applied in all aspects.Nowadays,personalized recommendation systems are still under developing in practical applications.There exist some problems such as data sparsity,low recommendation quality and real-time recommendation problem.Under the condition of data sparsity,the traditional similarity measurement in collaborative filtering is very simple and does not consider that the users' interest will change over time,which results in poor performance.According to these problems,the concrete research work and innovation in this thesis are as follows:1.An improved collaborative filtering recommendation algorithm based on user confidence and time context is proposed.This thesis integrates the confidence of the user's role and the change of the user's interest over time into the adjusted cosine similarity calculation method based on user.Firstly,the user confidence which is treated as a dynamic factor is added into the similarity calculation between users on the basis of the adjusted cosine similarity.Besides,the time context is integrated into the user's confidence calculation.Finally,the optimal results can be obtained by repeated experiments.The experimental results demonstrate that the improved algorithm is suitable for the parse data and effectively improves the prediction accuracy and the accuracy of recommendation.2.An improved collaborative filtering recommendation algorithm based on item confidence is proposed.The traditional item-based similarity calculation methods merely measure the similarity between different items by the rating data and ignore item confidence.Item confidence includes the similarity between item type,the number of common users between items and the influence of active users.The improved algorithm mainly considers the item confidence into the calculation on the basis of the adjusted cosine similarity based on item.The experimental results show that the performance of the improved algorithm is better than the traditional similarity calculation methods.What's more,it improves the accuracy of prediction.This thesis mainly improves the similarity calculation methods of the recommendation algorithm and integrates the condidence factor and dynamic interest over time into the adjusted cosine similarity calculation method.Experiments show that improved algorithms can improve the accuracy of prediction and improve the recommendation quality.