Automatically Building a Corpus for Sentiment Analysis on Indonesian Tweets

PACLIC 28 !185 Automatically Building a Corpus for Sentiment Analysis on Indonesian Tweets Alfan Farizki Wicaksono, Clara Vania, Bayu Distiawan T., Mirna Adriani Information Retrieval Lab. Faculty of Computer Science, University of Indonesia Depok, Republic of Indonesia {alfan, c.vania, b.distiawan, mirna}@cs.ui.ac.id Abstract The popularity of the user generated content, such as Twitter, has made it a rich source for the sentiment analysis and opinion min- ing tasks. This paper presents our study in automatically building a training corpus for the sentiment analysis on Indonesian tweets. We start with a set of seed sentiment corpus and subsequently expand them using a classi- fier model whose parameters are estimated us- ing the Expectation and Maximization (EM) framework. We apply our automatically built corpus to perform two tasks, namely opin- ion tweet extraction and tweet polarity clas- sification using various machine learning ap- proaches. Experiment result shows that a clas- sifier model trained on our data, which is automatically constructed using our proposed method, outperforms the baseline system in terms of opinion tweet extraction and tweet polarity classification. 1 Introduction There are millions of textual messages or posts gen- erated by internet users everyday on various user generated content platfroms, such as microblogs (e.g. Twitter1), review websites, and internet fo- rums. They post about their stories, experiences, current events that are happening, as well as opin- ions about products. As a result, the user generated content has become a rich source for mining useful information about various topics. Twitter, one the popular microblogging platforms, is currently getting a lot of attention from internet 1http://twitter.com users because it allows users to easily and instantly post their thoughts of various topics. Twitter cur- rently has over 200 million active users and produce 400 million posts each day 2. The posts, known as tweets, often contain useful knowledge so that many researchers focus on Twitter for conducting NLP- related research. McMinn et al. (2014) harnessed millions of tweets to develop an application for de- tecting, tracking, and visualizing events in real-time. Previously, Sakaki et al. (2013) also used twitter as a sensor for earthquake reporting system. They claimed that the system can detect an earthquake with high probability merely by monitoring tweets and the notification can be delivered faster than Japan Meteorology Agency announcements. More- over, Tumasjan et al. (2010) demostrated that Twit- ter can also be used as a resource for political fore- casting. Due to the nature of Twitter, tweets usually ex- press peoples personal thoughts or feelings. There- fore, tweets serve as good resources for sentiment analysis and opinion mining tasks. Many compa- nies can benefit from tweets to know how many pos- itive responses and/or negative responses towards their products as well as the reasons why consumers like/dislike their products. They can also leverage tweets to gain a lot of insight about their competi- tors. Consumers can also use information from tweets regarding the quality of a certain product. They commonly learn from peoples past experiences who have already used the product …

Content truncated to 500 words