Sentiment Analysis Based on Social Media Data

Abstract

Sentiment analysis has proven to be one of the most challenging tasks in natural language processing (NLP). Many AI systems have been developed which can detect the polarity of a sentence (degree of positivity, neutrality or negativity). But more information such as the emotion of the author can be detected. Our task here is to build an artificial agent – or an AI system – that is capable of detecting polarity in a document as well as the emotion of the author. Generally speaking, sentiment analysis detects the polarity of the opinion based on the object/subject in discussion. But emotion detection can identify the particular “mood” of the author. Social platforms – such as Facebook, Twitter, IMDB, or comment sections from online newspapers – to name a few – can provide a huge corpus of – usually unlabeled or extremely sparsely labelled – content. Using modern machine learning tools and the available computational power, an agent can analyze the content – given usually as a list of messages – and detect the general emotion “within the message”. It must be capable of identifying subjects with unstable or “chaotic” mood that might require attention. Our aim is to attempt detecting the underlying emotion as well as the polarity of a document. In this process we will make use of open source libraries and publicly available data sets. The program will be able to run locally, both in geographic and in cultural sense, and to analyze the results that were obtained. This work is the result of my own activity. I have neither given nor received unauthorized assistance on this work.