Bangla News Recommendation Using doc2vec

Abstract

We present a content-based Bangla news recommendation system using paragraph vectors also known as doc2vec. doc2vec is a neural network driven approach that encapsulates the document representation in a low dimensional vector. doc2vec can capture semantic relationship effectively between documents from a large collection of texts. We perform both qualitative and quantitative experiments on a large Bangla news corpus and show that doc2vec performs better than two popular topic modeling techniques LDA and LSA. In the top-10 recommendation scenario, the suggestions from doc2vec are more contextually correct than both LDA and LSA. doc2vec also outperforms LDA and LSA on human-generated triplet dataset with 91% accuracy where LDA and LSA give 85%,84% accuracy respectively.