An Attention Based Approach for Sentiment Analysis of Food Review Dataset

Abstract

Sentiment Analysis is a technique related to text analysis and natural language processing used to detect various types of insights or information from a portion of text. Over the past few years, researchers have done many works regarding this. In Bangladesh, many online services like-e-com become very popular day by day. One of them is online food delivery services. We can order various foods of our choice from online and sometimes people gives reviews based on that food. Those reviews are usually discarded as unstructured data which of them have no work in further. In this piece of research focus primarily on those unstructured data to analyze them in a correct manner to find insight into customers' behavior and their reactions on those online platforms. To do this experiment first we collect data from websites. Later deep learning-based techniques applied here. For baseline structure, we have used both CNN and LSTM models. Then for improving the model accuracy an attention mechanism applied followed by CNN which gives us 98.45% accuracy. We've also evaluated our model performances with some evaluation metrics also. From them, CNN based attention model gives a higher f1-score of 0.93.