Large Multi-Domain Resources for Arabic Sentiment Analysis

3rd Best Paper winner at the International Conference on Computational Linguistics and Intelligent Text Processing CICLing2015

Download the paper from here

Overview :

The Repository includes the following :

• 33K Automatically annotated Reviews in Domains of Movies, Hotels, Restaurants and Products
• Domain specific lexicons, semi automatically generated from the datasets above (2K total)
• A total of 615 Experiments over each of the datasets experimenting :
  • Classifiers : Linear SVM, Logistic Regression, KNN, BNB, SGD training with SVM (Hinge loss and L1 penality)
  • Sandard Features : TFIDF, Term Count, Term Existence, Delta-TFIDF
  • Lexicon Based Features: domain specific and domain general
  • Combining features : Lexicon based feature vectors + Standard features
  • Classification Problems : with neutral class included or not
  • Balanced or unBalanced Datasets
• Results of Each of the Experiments

Dataset Statistics

Datasets :

####ATT.csv

• Dataset of Attraction Reviews scrapped from TripAdvisor.com
• 2154 reviews

####HTL.csv

• Dataset of Hotel Reviews scrapped from TripAdvisor.com
• 15572 reviews

####MOV.csv

• Dataset of Movie Reviews scrapped from elcinema.com
• 1524 reviews

####PROD.csv

• Dataset of product reviews scrapped from souq.com
• 4272 reviews

####RES1.csv

• dataset of restaurant reviews scrapped from qaym.com
• 8364 reviews

####RES2.csv

• dataset of restaurant reviews scrapped from tripadvisor.com
• 2642 reviews

####RES.csv

• RES1.csv and RES2.csv combined
• 10970 reviews

Lexicons

Domain specific lexicons, semi automatically generated from the datasets above (2K total)

lexicon	MOV	RES	PROD	HTL	BOOK	Total
size	87	734	369	218	874	1913