Research On The Architecture And Key Problems Of Recommender System Based On Community Tag

Along with the development of the information technology and communicationtechnology, and the popularization of internet, how to help people to find the usefulknowledge in vast quantities of information is a problem that is exigent to be solved.The personalized recommendation system is a useful tool to conquer the informationoverload, and helps the users to find the necessary information. The recommendationsystem builds the interesting model through analyzing the characteristics of users(items), and the users’ operating behaviors on items. Based on the models, therecommendation system can predict the users’ preferences to the items, andrecommend the appropriate items. In recent years, the recommendation system hasbeen widely used in industry, such as e-commerce, and researched deeply in theacademic community. However, along with the information requirements of usershave developed into diversification, there are some challenges in recommendationsystem, such as the architecture of recommendation system, the efficiency ofalgorithms, the diversity and accuracy of recommended items. This paper mainlystudies on the architecture of multitasking recommendation system, the models andalgorithms of recommendation based on the collaborative filtering, and verify theiravailability through the standard data sets. The main contributions of our researchare as following:1. In terms of architecture of recommendation system: our paper designs theprototype of recommendation system, which has5main components, includinginteraction interface, data acquisition and preprocess, behavior analysis and featureextraction, multitasking recommendation module, and feedback analysis. Thepluggable recommendation architecture can choose and deploy the algorithm andmodel according to the users’ requirements. The recommender system can make useof the system architecture and efficient recommending algorithm to get the highaccuracy and diversified recommended items.2. In terms of improving the diversity of recommended items: we propose the cross domain recommendation based on the cross nearest neighbor model. Throughanalyzing the Folksonomy of different kinds of recommended items, we can minethe association rules in semantics among users’ interests in different domains. Theexperimental results show that our method can implement the cross domainrecommendation and get the association rules among users’ preferences in differentdomains with tags.3. In terms of improving the accuracy of recommended algorithm: we proposeto integrate the attributes into the graph based model and latent factor model in orderto improve the accuracy of Top-N recommend task and predictive rating task.4. In terms of improving the efficiency of recommended algorithm: we proposetwo models: one is the probabilistic clustering model based on clustering algorithm,the other is the incremental SVD model based on the singular value decomposealgorithm. The two models can be used to real-time recommendation.Based on the architecture and algorithms which are introduced in this paper, webuilt the book recommendation system: Readings. Our research satisfies algorithminnovation and engineering practices. Our multitasking recommending architectureand recommending algorithms can improve the effectiveness, accuracy and diversityof recommendation to a large extent.