Multilingual sentiment mapping using twitter, open source tools, and dictionary based machine translation approach

Abstract

Online social networks are a popular communication tool for internet users. Millions of users share opinions on different aspects of everyday life. Therefore, microblogging websites are rich sources of data for opinion mining and sentiment analysis. Our current research based on the analysis of migration using various social networks required to implement a tool for automated multilingual analysis of sentiment from as many languages as possible. Usually, all available tools handle to work only with English written texts which are the most common on the social media. Few open source tools which can process French, German and Spanish texts exist too, but it is not optimal to reimplement and join different approaches together. Another requirement is the ability to process dynamic data streams and static historical datasets with high efficiency. Lesser accuracy and completeness of evaluated messages is acceptable as a counterweight for these general requirements. The paper presents sample data collection from Twitter for the opinion mining purposes. We perform multilingual sentiment analysis of the collected data and briefly explain experimental results. The analysis is made with the use of custom built solution utilising the AFINN-165 which is manually evaluated dictionary of English words. This dictionary was translated into other languages using Google Translate API that was tested during the process. It is then possible to determine positive, negative and neutral sentiment. Results of the research bring new insights, offer a possibility for wider use and allow optimisation of the wordlists/tool resulting in the better results of future research. Geospatial analysis of first experimental results undercovers interesting relation between time, location and a sentiment which enables readers to think of various use cases.