ABSA
Aspect Based Sentiment Analysis
Retrieves aspects from the user reviews with sentiment score.
** Aspects: {Value, Food, Drinks, Staff, Service, Ambience, Location and Price}
Datasources Extraction:
Training Datasources: (Fetch data from) :
** Google places API - • https://maps.googleapis.com/maps/api/place/nearbysearch/json • https://maps.googleapis.com/maps/api/place/details/json ** Zomato API - • https://developers.zomato.com/api/v2.1/locations, • https://developers.zomato.com/api/v2.1/location_details, • https://developers.zomato.com/api/v2.1/reviews) ** Twitter • tweepy.OAuthHandler • tweepy.API • tweepy.Cursor ** CitySearch scraping • http://www.citysearch.com/listings/houston-tx-metro/restaurants.html
Test Data of: (Manually fetch and test)
** TripAdvisor • from Kaggle and directly using tools, for testing
Data Preparation:
- Data cleaning (emojis, date formats, text formats)
- XML formatting of Review and each aspect in it.
Steps for Training:
- Pretrained BERT
- Parse and pass XML for training.
- Vectorization using SpaCy.
- 4 fold Cross validation
Model Preparation:
- Aspects extraction
- MultinomialNaiveBayes
- Sentiment Score
Testing Accuracy:
- Confusion matrix
- Multiclass classification