Finding Key Terms Representing Events from Thai Twitter

Abstract

In the fast and big data era, we all desire to understand trend or big picture of a story instantly. This work wants to find an automatic approach to extract the good-enough key terms of each event appear in Thai Twitter society. The core idea is to help reducing time for human to do the key term extraction, yet the quality of such selected key terms are acceptable by human and is better than our previous implementation. Our studied approaches focus to work on Thai language and covered preprocessing, feature selections and weighting schemes on three Thai real tweet events with different characteristics. Our experiment comprise four main approaches and a number of hypothesis. Our findings confirm the usefulness of hashtag terms with five or more character length, the benefit of bigram with stop words and the importance of event characteristics. In fact, we conclude to use different approaches for different types of event. The performance and rational evaluations are done by statistical analysis, evaluators voting, and F-Score measurement and are confirmed to be better than previous work twice as much.