Neural Network based Emotion Recognition Model in Twitter Text Data

Abstract

Recently, the online social networks are rising as the new field for the individuals to share their perspectives and viewpoints on various issues and subjects with their companions, family, family members, and so on. Twitter being the well-known micro-blogging tools permits the people to share their musings, mental status on explicit social, public, worldwide issues through textual content, photographs, voice and video messages and posts. In reality, in spite of the accessibility of the different types of correspondence, text is one of the most well-known methods of communication in an informal community. In the tweets, the emotional articulation has an essential function in different impact of consistent communication. The emotion extraction from twitter sentiments is more troublesome particularly from multi-dialects messages. Therefore, in this research work a novel sentiment based emotion recognition model is introduced with five steps as follows: 1) Pre-processing, 2) Keyword extraction and Sentiment Categorization, 3) Semantic similarity score measurement, 4) Feature extraction and 5) Classification. Initially, the collected raw twitter data is subjected to pre-processing, where the tokenization, stemming and stop word removal is undergone. After pre-processing, the keywords are extracted and the similarity score is measured for each of the emotion based sentiments. Then, from the computed semantic similarity score, the differential holoentropy based features and weighted probability based holoentropy features are extracted. These extracted features are subjected to emotion detection (classification) by Neural Network (NN), which outputs the detected emotion. Finally, the proposed work is compared over the existing works in terms of positive, negative and other measures.