Bangla Aspect-Based Sentiment Analysis Based on Corresponding Term Extraction

Aspect-based sentiment analysis is a text analysis technique that extracts and separates each aspect term and identifies the sentiment polarity associated with each aspect term. Bangla is the seventh most spoken language in the world. Sentiment analysis in the Bangla language is considered a crucial and well-timed research topic. Aspect-based sentiment analysis of the Bangla language is treated as a complicated task because of the scarcity of resources like annotated datasets, corpora, etc. In this research, we have proposed a new technique named PSPWA (Priority Sentence Part Weight Assignment) to perform aspect category or term extraction on publicly available datasets named Cricket and Restaurant. We have used conventional supervised learning algorithms and Convolutional Neural Network (CNN) to demonstrate results. Dataset preparation, feature engineering, description of PSPWA, CNN architecture, experimental results along with a state-of-art comparison has been shown in this paper. The public dataset was imbalanced. CNN has performed better among other learning algorithms. CNN has achieved an f1-score of 0.59 and 0.67 for the cricket and the restaurant dataset respectively.

I. Introduction

The number of social media users is increasing enormously. At least 3.96 billion people across the world are using social media today, equating to 51% of the total global population. On average, they are spending 2 hours and 22 minutes using social media each day [1]. Alike, online shopping is popularizing day by day. An estimation of 1.79 billion people buys digital goods worldwide in 2019. The forecast says the number of online buyers will increase by over 2.14 billion in 2021 [2]. People on the internet are from different regions, casts, cultures, and languages. As a result, a large number of online contents are generated every day in various aspects. According to Forbes, 2.5 quintillion bytes of data are created each day at the current pace [3].