Recommendation System for Bangla News Article with Anaphora Resolution

Abstract

This paper represents an efficient approach for Bangla News Recommendation. In traditional Bangla news recommendation system, recommend news from the different newspapers of the same day. Actually, they contain the same news just from different sources. From the user's view, it is more desired if users get to know more diverse information on the same news. In this paper, we have represented a noble approach for recommending news on the same topic with more diverse information. At first, we have done news clustering; it is an automatic learning technique aimed to create clusters that are coherent internally, but substantially different from each other. In this approach, we have used anaphora resolution to increase the keywords frequency. We build an automatic word tagger for anaphora resolution, which can tag all nouns and pronouns with five different criteria (Number, Person, Status, Gender, and POS). Next we have counted document wise unique words to calculate tf-Idf algorithm with cosine similarity to make the recommendation. Finally, we have done three different modified technique of reverse hierarchical clustering on the same cluster to identify more distinct news which is related to the same subject.