Holistic Multilingual Sentiment Analysis on Reviews in Social Media

Abstract

Decades ago, people used to represent their opinion by writing it manually or by speaking at public places. These reviews are further taken as an advice for the betterment. To process this data was a complete manual task. As the usage of internet grew, people started sharing their views regarding any entity through emails or social platforms. newlineThe intensification of data over the social media makes the task of deducing valuable information a bit complex. Automatic deduction of sentiments from web data is considered as a process of sentiment analysis. An algorithm devised for the same is known as sentiment analyser. Use of Sentiment analysers is at the peak for various enterprises to find the loopholes in their product or services. An optimal sentiment analyser is the one which works rationally as humans. The goal is thus to fill the research gaps associated with the effective sentiment processing. newlineSentiment analysis integrates many subtasks i.e. Named Entity Extraction, Anaphora resolution, Sentiscore, Feature extraction, etc. Effective pre-processing yields better results for all the natural language processing tasks. The reason for pre-processing of the data is that people use slangs, long tail words, multilingual content and visual language such as emoticons. People use unstructured format of writing along with all the above mentioned categories these days. The process to handle slangs, misspelled words, etc. is called as normalization. newlineThe primary aim of this study is to have effective pre-processing of the content i.e. normalization. Normalization here deals with two aspects: one is to deal with slangs and another is to deal with emoticons. In this study, a technique is used where each emoticon is mapped corresponding to its meaning for generating Sentiscore, instead of just adding or subtracting one for positive and negative smiley respectively. Slangs are also handled effectively by using cross word dictionary and corpus. It is aimed to get better results for pre-processing. newline