A Machine Learning Based Approach to Analyze Food Reviews from Bengali Text

Nowadays, people share their opinion about restaurants and food items on different online platforms. Before visiting any restaurant people often read the reviews about the food items. Again, the food industry measures the satisfaction level of consumers through these reviews and tries to improve the food quality according to the demand. However, it’s a difficult task to manually go through the reviews. To make the task easier analyzing the reviews automatically would be a great achievement. Although there exists a good number of research works to analyze food reviews written in the English language. Unfortunately, very few focuses on food review analysis in Bangla language. Because of this unstructured research works in this field, we explore six different machine learning algorithms to classify food reviews from Bangla text into three categories i.e. good, neutral and bad. We collect almost 4000 food reviews from different online sites. Among them, 80% data is used for training and 20% is used for the testing purpose. To extract the feature two different feature extraction techniques Term Frequency – Inverse Document Frequency (TF-IDF) and CountVectorizer (CV) are used using unigram, bigram and tri-gram models. Our experimental results reveal that SGD classifier can reach up to 93% accuracy on bi-gram features that are extracted using TF-IDF technique.