Research On Personalized Recommendation Based On Collaborative Filtering

The human society has already entered the age of information explosion, and lots of information floods around people everyday. As the information overload problem increases seriouly, the recommender system emerges. The system provides personalized recommendations for people by collecting the interested and personalized information of users. Since there is no significant difference in the acquired information, recommendation algorithm as the core recommendation technology, decides the performance of recommendation system. Collaborative filtering technology is the most widely used recommendation technology. However, the technology still has many problems need to be solved such as the problem of data sparse. Besides, collaborative filtering algorithm only analyses the static behavior without considering the changes of system, which leads to low dynamic accuracy of recommendation.In view of above problems, this thesis researches the collaborative filtering technology including: improving collaborative filtering algorithm using neural networks and trust relationship; analysing the dynamic data set and proposing a dynamic model of collaborative filtering algorithm.The main contributions of this thesis are as follows:1. Summarize the background, significance and progress of collaborative filtering recommendation; introduce the information filtering technology and data mining technology; analysis the process, advantages and disadvantages of several commonly used collaborative filtering algorithm; point out challenges and hotspots of recommendation technology providing theoretical basis for further study.2. Train the generalized regression neural network model and predict the missing rates to decrease the data sparseness.3. Study the relationship between trust and traditional similarity measurements; design personalized recommendation model based on trust; propose methods to measure the trust; improve traditional rating prediction formula based on trust relationship; improve accuracy of collaborative filtering recommendation based on trust.4. Analyse the dynamic characteristics of recommendation system; research on the timeliness on the similarity measurement and the accuracy of recommendation; induce timeliness to the similarity and rating prediction process of collaborative filtering algorithm; improve the baseline model and matrix decomposition model based on the dynamic changes of the system.