Research Of Scholarly Paper Recommendation Via Hybrid Algorithm Based On Co-author Network

The amount of scientific literatures grows very fast, it increases the difficulty on lookingfor appropriate papers. Search engine lightens the work of searching for scholarly paperobviously, but it lacks consideration of researcher’s personalization. It’s difficult to find usefulinformation from the results. The recommender system is to solve those problems efficiently.In this thesis, we compare several common recommendation algorithms and theiradvantages and disadvantages. Then we introduce co-author network analysis methods andtechniques, study the phenomenon of cooperation in Sociology and analyze the overallfeatures of co-author network and the importance of scholars. Besides, we validate thesmall-world and scale-free properties of co-author network and explain the sociality andcommunity of scholars in co-author network. Du to the defects and limits of applicationscenarios, traditional single algorithms are not ideal in scholarly paper recommendation.Considering of the present shortcomings of existing researches, we studies the hybridalgorithms of scholarly paper recommendation from the following aspects:1. To describe the user’s preference correctly, we build the dynamic user interest modelvia his published papers and describe the importance of scholarly paper via quality evaluationmethod. Then we propose a hybrid algorithm on the basis of methods those mentioned before.2. To reduce the blindness of recommendation, we lead co-author network into thehybrid algorithm of scholarly paper recommendation proposed before and defined thecalculation of cooperation strength between different scholars in co-author network. To limitthe spread of cooperation strength, we divide the expanded co-author network intocommunities which contains similar scholars.3. The users have a high propensity to read the papers on the top of recommendation list.To measure the ability of results ranking of the hybrid algorithm, we use mean averageprecision and mean reciprocal rank to evaluate the performance of Top-N itemrecommendation problem.The results of experiments show that hybrid algorithms are better than traditionalalgorithms on scholarly paper recommendation; especially the hybrid algorithm which appliesco-author network and community division performs better than the others.