Overview / Usage

Consider Zomato has newly formed a dedicated customer engagement team. As a first step, they want to build a system to reach out to those Zomato users who were dissatisfied with their restaurant experiences. Now, dissatisfaction in this context can be inferred from the review(s) a certain user is giving for a particular restaurant. hence, the conceptualization of this project. Based on their reviews, we can associate a binary sentiment to them. For example, the reviews having less than or equal to 3 stars' ratings convey negative sentiment. So, the main problem reduces to classifying sentiments from given Zomato reviews.

There is no Zomato reviews' dataset available on the internet, hence the project started with data collection (which is done by Nilabhra Roy Chowdhury (refer this notebook: https://github.com/Nilabhra/kolkata_nlp_workshop_2019/blob/master/scraping.ipynb)).

The final sentiment classification model is deployed as a REST API on Heroku. Here's the demo: https://www.loom.com/share/c561a10760aa4e658932f51282557627

Methodology / Approach

• Used binary bag of words model to preprocess to build the corpus
• Used shallow neural nets to train using the corpus using tf-keras
• Built a custom sklearn prediction pipeline by combining the tf-keras model and the data preprocessor

Technologies Used

• Python (language)
• scikit-learn, tensorflow, flask (main libraries)
• Heroku (Deployment platform)

Repository

https://github.com/sayakpaul/Reviews-Classifier-Heroku-Deployment