Research articles suggestion using topic modelling

Abstract

Searching research articles effectively is significantly important to researchers. Currently, researchers use search engines like Google Scholar and search by keywords. The typical search result includes a lot of articles which match the keywords exactly; however, they are on many different topics. It is very time and effort consuming to go through the articles manually and select desirable ones. We propose a method to search articles by topics. Our method trains on a large set of articles and analyzes every article in it to generate its distribution over topics. The method recommends articles to the researcher by analyzing the input given by her/him and comparing with the topic distribution of all articles in the training set. The articles with most similar distributions are recommended to the researcher. In this way, articles are recommended by matching by topics rather than keywords. Our experimental results are analyzed using Mean average precision (MAP) and Normalized Discounted Cumulative Gain (NDCG). The obtained results demonstrate that our method successfully extracts the topics beneath the words of an article and recommend closely related ones.