Deep learning based sentiment analysis and text summarization in social networks

Sentiment analysis aims to reveal semantic knowledge on written texts that share the emotions and thoughts of users like personal blog and social network. Because the data shared by users on social networks is made up of short texts, it can be seen as a text classification problem. Although language libraries have been developed in other languages to solve the problem of emotion analysis, studies for Turkish language are still limited. In this study, two categories of emotion analysis were studied with the data obtained from many social networks. Sentiment analysis is considered as a classification problem, and the success rate is increased by using semantic context word embedding methods. LSTM, which is a deep learning method, was used instead of classical text classification methods. In order not to lose the semantic understanding between words, success rates were observed by using more than one word embedding method. LSA was used for text summation. In this study, where both emotion analysis and text summation are carried out, the main goal is to analyze the emotions and thoughts about a subject and present brief information to the user. The main model was created with data collected from many social networks. Analysis and summary of the text was made with data from a hashtag on Twitter. The highest 93% success rate was achieved with the methods used in the analysis of emotions. A summary of the text has been successfully presented.