A Personalized Paper Recommendation System Based On User Interest Models

As a novel media of knowledge propagation, Sciencepaper Online provides a convient way for publishing and reviewing the academic words. However, like the traditional aggregation-based publishing platforms, the existing ones inherit the same deficiencies. For example, they wait for paper reviews passively. Moreover, they lack of effective personalized services to enhance the dynamic scheduling of paper resources. Based on these problems, we in-depth research all kinds of personalized recommendation algorithms, data mining algorithms and the characteristics of the paper publishing platform. Finally, we raise a personalized paper recommendation system, whick has the ability to automaticly collect the sentific experts'information to model the reviewer's interest, and actively explore the characteristics of paper newly published to construct the paper model. With these models, the system can find the most suitable reviewers for the paper, and then send each of them an email to recommand the papers.After comprehensively analyzing the characteristics of paper recommandation application, we innovatively create an interest model for reviewers. This model doesn't only effectively decrease the computation multeplexity in time and space, but alse improve the quality of recommandation result. Besides, this model can dynamicly update itself and trace the transfer of users'interests. By improving the existed algorithm, the system achieves a batter precesion rate and recall in result.In additon, to reduce the passive impact of stopwords when building models for reviewers'interst and papers being recommended, we develop another system based on the data mining algorithms. This system can create a Chinese stopwords list and English stopwords list to filter the meaningless stopwords, and then update them dynamicly.