Classification of Food Reviews from Bengali Text using LSTM

Abstract

People of this modern era are very much dependable on online reviews when it is the matter of purchasing any product. It is vital to bring out information from the huge amount of accessible text reviews. People of almost every age often visit restaurants. In today’s world food review is the fundamental requirement for visiting restaurants. But selecting a restaurant based on reviews is not quite an easy task. Deciding whether a food is worth having or not can be difficult. Several factors including the price, quality, taste, quantity can influence the actual worth of a food. From the perspective of a consumer, it is a dilemma to select a food appropriately. Food quality prediction can be a challenging task due to the high number of reviews that should be considered for the accurate prediction. Most people nowadays select restaurants based on their preferred food’s review. But the reviews present on the social platforms are mostly broad. People don’t find it useful to read the whole review. Therefore, a model which is capable of accepting reviews as input and is able to predict the food quality as output can become a great solution to this problem. So in this study, we have introduced a method which will be able to classify long Bengali food reviews into precise classes such as Good, Bad and Best using LSTM. The whole dataset which was used in our experiment has been collected from Facebook food review groups. Among them 80% was used for model training and 20% data used for the validation. Our model was able to classify 1000 Bengali review with 98% training and 80% validation accuracy.