Cluster recommendation of scientific literature for research area analytics

Abstract

Due to a fast-paced increase of the number of research articles published, it ishard for a researcher to find relevant articles. Research paper RecommendationSystems help in finding the relevant research article. Most Article RecommendationSystems are realized through clusteringIn this thesis, we mainly focus on clustering of documents which helps laterfor the recommendation of scholarly articles. When a user is searching for anypaper, he/she has a different kind of search purpose like "search for the paperwhich solves the same problem that a user wants to solve" or "the papers whichsolve the same problem with different approaches" etc. Here our aim is to groupall those papers that solve the same problem. Further in attempt to extract researchproblems from the article and attempt "problem-based" clustering. In thiswork, we explored various state-of-art clustering techniques for the said objective.We propose a hierarchical approach to cluster scientific papers which addressthe same problems. First we cluster research articles based on their abstracts.Within a topic, we further cluster them based on "problem" extracted from abstract.For first level clustering, we use topic modeling technique Latent Dirichletallocation.For second level clustering, we use Linguistic clues for extracting "problem" ofarticles. We try K-means algorithms with the different features. Also, merge twoclustering techniques K-means and K-nearest neighbors.We conducted experimental studies on DBLP. We show how the result of Kmeansalgorithm depends on features which are used to train algorithms. Comparethe results of different approaches.