A Combination of Text Mining Techniques for Relevant Literature Search and Extractive Summarization

Over the past few years, the amount of research papers published has dramatically increased. Consequently, researchers spend a lot of time reviewing relevant literature in order to better understand their domain of interest and keep up with new developments. After doing literature reviews in the area of text mining, we found many works proposing the means of sentence representation in machine learning for finding sentence similarity. These include average bag of words, weight average word vectors, bag of n-grams, and matrix-vector operations. However, these techniques are limited in word ordering and semantic analysis. This paper proposes a framework that combines two text mining techniques, paragraph vectors and TextRank, for the selection of relevant research paper and extractive summarization, respectively. Our training corpus includes over 20 million research papers. The aim of this work is to build a supplementary research tool that assists researchers in saving time conducting literature reviews. As the result, we can rank all relevant research papers potentially within the corpus, and utilize the outputs in our literature reviews. Moreover, the tool can extract all potential keywords in a single task as well.