Time Series Analysis of Sentiment Polarity Trends: A Case Study

The availability of a large number of opinion about the particular product or service may cause user confusion about final opinion. More and more methods for sentiment analysis are developed, that can judge the polarity of each opinion. The problem arises during the process of determining the final polarity and to predict future opinion. In the paper we analyze a set of opinions written by the group of users (e.g. familiar users in a social network) using time series analysis methods. The main aim is to analyze the trend of polarity in a group of people and to predict sentiment towards a particular topic in this group. We analyzed a dataset of real opinions using time series decomposition and prediction methods for determining the trend of sentiment score. The performed experimental evaluations have shown the efficiency of the proposed method. The results can be useful for future development of a complex model for opinion forecasting.