This is an end‑to‑end deep learning project that is about classification of Google Maps reviews. I wrote an academic paper that related to the project. My paper on the project was published on the YBS Encyclopedia website. I also collected and published a dataset for the project. The artificial intelligence model in this project that I developed categorizes the reviews made about the business into certain classes and finally presents the average score of the business in each category to the user as data. With this application, businesses will be able to easily see their shortcomings and correct them.
Data includes reviews of different restaurants on Google Maps. There are 1100 comments in total and pictures of each comment in the data set. The data is labeled according to 4 classes (Taste, Menu, Indoor atmosphere, Outdoor atmosphere) for the artificial intelligence model to predict. The dataset has been prepared in a way that can be used in both text processing and image processing fields. The dataset contains the following columns: business_name, author_name, text, photo, rating, rating_category IMPORTANT: The rating_category column is related to the photo of the review. If you want to use this dataset for NLP, you need to label it yourself. I will label it for you when I am available.
My academic article written in Turkish was shared in YBS Encyclopedia. If you review the article, you can see that I tested many popular ready-made deep learning models and reported the results. I also examined the performance of the neural networks at different depths that I created and included them in my article. For details, you can access my article from the link below.
I developed using Python and Tensorflow/Keras. I prepared the data for model training using various methods such as data augmentation, image preprocessing, and data analysis. I have tried many deep learning architectures and reviewed them all. As a result of the analysis, I saw that the best architecture was the model with 15 layers and 74.11% accuracy on the test data.