Provision of Efficient Sentiment Analysis for Unstructured Data

Abstract

Applications on sentiment analysis via diverse context in analyzing the individual opinion on various issues such as political events, e-governance, and product reviews. Decision-making through the sentiment analysis improves the understanding of the public opinion. Opinion mining can be achieved by retrieving data through social network, microblogs, blogs, and search engines. Twitter tweets are an invaluable source of knowing the individual opinion from lots and lots of unique personality. However, the unstructured data to the huge volume and unwanted punctuation used in context and the emoticons used in the context is the main task to analyze the efficiency data and with greater accuracy. Most of the existing computational methods/algorithms identify sentiment on unstructured data via algorithms on machine learning like (BOW) bags of word approach. Here is the work of both the supervised and unsupervised approaches on various training datasets used. Automatic generation of the sentiment for the tweets extracted is provided by the unsupervised approach. And various algorithms in machine learning are used to determine the sentiment analysis they are as Maximum entropy (ME), Multinomial Naïve Bayes (MNB) and support vector machines (SVM) and are used to identify sentiment from the tweets. Here in this work I have achieved an accuracy of 87% unsupervised 97% in supervised approach. The ngram, unigram, bigram, and parts-of-speech (POS) were combined together to identify the hidden emotion and sentiment in the context that mentioned in the tweet that are all in an unstructured format. The lexicon-based approach 75.20% is achieved, based on the sentiment prediction the opinion is given.