Abstract
Recent years witness a significant increase in research related to knowledge extraction from web social networks or media. The enormous volume of posted comments, and related media can be a rich source of information. In Middle East and the Arab world in particular, social media websites continue to be the top visited websites especially with the current social and political changes in this part of the world.
Sentiment analysis and opinion mining focus on identifying and evaluating positive and negative opinions and comments. This study aims to identify the sentiment polarity for collected comments or posts from Twitter using a hybrid approach and a modest dataset of Arabic (Text and audio) comments. Two machine learning classification techniques are used to perform the required classification to identify the polarity of the collected opinions. We extended the evaluation of prediction algorithms and enhance them using Bagging and Boosting algorithms. We extracted a unified dataset of texts, audios and images and applied processing methods to extract final sentiment opinions. We noticed that some special expressions specially in recoding (such as laughing, yelling, etc. within the recording) have a negative effect on the accuracy of the automatic sentiment prediction system.