Abstract:
With the popularization of social media, a vast amount of textual content with additional geo-located and time-stamped information is directly generated by human every day. Both tweet meaning and extended message information can be analyzed in a purpose of exploration of public mood variations within a certain time periods. This paper aims at describing the development of the program for public mood monitoring based on sentiment analysis of Twitter content in Russian. Machine learning (naive Bayes classifier) and natural language processing techniques were used for the program implementation. As a result, the client-server program was implemented, where the server-side application collects tweets via Twitter API and analyses tweets using naive Bayes classifier, and the client-side web application visualizes the public mood using Google Charts libraries. The mood visualization consists of the Russian mood geo chart, the mood changes plot through the day, and the mood changes plot through the week. Cloud computing services were used in this program in two cases. Firstly, the program was deployed on Google App Engine, which allows completely abstracts away infrastructure, so the server administration is not required. Secondly, the data is stored in Google Cloud Datastore, that is, the highly-scalable NoSQL document database, which is fully integrated with Google App Engine.
Keywords:sentiment analysis, public mood, mood patterns, twitter, social media.
Fulltext:
PDF file (944 kB)