Sentiment Analysis of Twitter Data Using Sentiment Influencers

Abstract

Much of the research work has been carried out on sentiment analysis and is still under consideration by many researchers. There are many documents that present author’s subjective views on particular topics. Sentiment analysis is the extraction of attitudes and opinions from human authored documents. Sentiment analysis has attained much attention in the recent years due to its significance in various fields as it captures and analyzes such attitudes and opinions in an automated and structured fashion and offers a powerful technology to a number of problem domains, including business intelligence, marketing, national security, crime prevention and healthcare/wellbeing services. Social media in particular becomes an interesting and practical source for sharing sentiments, emotions and opinions therefore they are also useful for sentiment mining. The text written by any author is spontaneous, unstructured and disordered. There is a need to classify this text and to acquire some significant outcomes such as whether the author is admiring or criticizing. Sentiment analysis can also be used to filter emails and other messages, or indicate abusive messages in newsgroup or helps users navigate via the Internet not only using topic keywords but also opinions. This research aims to develop technologies for extraction and analysis of sentiment from twitter data known as tweets, using an unsupervised machine learning technique with the focus on POS tagging of the tweet in which the parts-of-speech are ranked according to their sentiment describing influence. A novel term is devised for POS which is named as “sentiment influencers” and are ranked according to their influence on detecting sentiment. The proposed framework consists of five main phases named as data acquisition, pre-processing, Sentiment influencers identification, Rule based search engine and sentiment classifier. The gathered data is filtered in the pre-processing step and passed to the POS tagger that tags the tweet into noun, verb, adverb and adjective. SentiWordNet dictionary is used to calculate the score of each word in the tweet. The scores of these parts-of-speech are summed together and divided by total number of words in the tweet to get the final score. The final score predicts whether the tweet should be classified as positive, negative or neutral. This research has been implemented in java using Eclipse with Oracle database as backend data store. A number of APIs have also been used to supplement its functionality. Results are calculated and compared with other author’s work. Evaluation measurements such as Precision, Recall and F-measure are calculated. The research can be placed in the context of different communities, including research (e.g. biomedical/bioinformatics forums), medical/health wellbeing, business intelligence, ecommerce, market analysis, social media etc.