alexrutherford / arabic_nlp

Tools to normalise and derive sentiment from Arabic text

Summary

Python scripts to read in raw Arabic text, remove diacritics, perform basic normalisation and stemming and then derive sentiment from term frequency. Positive and negative words are supplied and based on Egyptian and Levantine dialects as well as Fus'ha.

normalise_file.py reads input file, cleans and normalises text, produces normalised output file and file with all links and their frequency.

get_sentiment.py reads normalised text file, assigns a positive and a negative sentiment value to each line based on term frequency. Saves sentiments to a file.

Files

• exempt_words.txt Words that are exempt from cleaning since they trivially match stop words or lemmatisation e.g. 'و' in 'ﻭﺎﻠﻠﻫ'
• negation_words.txt Words that can be used to negate (for more sophisticated bigram analysis)
• neg_emojis.txt Emojis that imply negative emotion
• pos_emojis.txt Emojis that imply positive emotion
• neg_words.txt Arabic words with negative sentiment
• pos_words.txt Arabic words with positive sentiment
• stop_words.txt Arabic words that do not impart meaning e.g. 'و'

Citation

The list of stop words and labelled positive and negative words in based on the Masters Thesis of Amira Magdy Shoukry: Arabic Sentence Level Sentiment Analysis American University in Cairo Spring 2013 (link)

Dependencies

get_sentiment.py requires NumPy and Matplotlib, otherwise all code in pure Python

To Do

• Implement generator for memory friendly treatment of large files
• Update lists of emojis
• Parse arguments with argparser