On the Impact of Seed Words on Sentiment Polarity Lexicon Induction

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1557–1567, Osaka, Japan, December 11-17 2016. On the Impact of Seed Words on Sentiment Polarity Lexicon Induction Dame Jovanoski, Veno Pachovski University American College Skopje UACS, Macedonia {jovanoski,pachovski}@uacs.edu.mk Preslav Nakov Qatar Computing Research Institute HBKU, Qatar pnakov@qf.org.qa Abstract Sentiment polarity lexicons are key resources for sentiment analysis, and researchers have in- vested a lot of efforts in their manual creation. However, there has been a recent shift towards automatically extracted lexicons, which are orders of magnitude larger and perform much better. These lexicons are typically mined using bootstrapping, starting from very few seed words whose polarity is given, e.g., 50-60 words, and sometimes even just 5-6. Here we demonstrate that much higher-quality lexicons can be built by starting with hundreds of words and phrases as seeds, es- pecially when they are in-domain. Thus, we combine (i) mid-sized high-quality manually crafted lexicons as seeds and (ii) bootstrapping, in order to build large-scale lexicons. 1 Introduction The recent rise of social media has greatly democratized content creation. Facebook, Twitter, Skype, WhatsApp and LiveJournal are now commonly used to share thoughts and opinions about anything in the surrounding world. This proliferation of social media content has created new opportunities to study public opinion, with Twitter being especially popular for research due to its scale, representativeness, variety of topics discussed, as well as ease of public access to its messages. Naturally, this abundance of data has attracted business and research interest from various fields in- cluding marketing, political science, and social studies, among many others, which are interested in questions like these: Do people like the new Apple Watch? What do they hate about iPhone6? Do Amer- icans support ObamaCare? What do Europeans think of Pope’s visit to Palestine? How do we recognize the emergence of health problems such as depression? Do Germans like how Angela Merkel is handling the refugee crisis in Europe? What do republican voters in USA like/hate about Donald Trump? Answer- ing these questions requires studying the sentiment of opinions people express in social media, which has given rise to the fast growth of the field of sentiment analysis in social media. Initially, sentiment analysis was addressed as a text classification problem, but it was soon realized that sizable performance gains can be obtained from using carefully built sentiment polarity lexicons as a source of external knowledge. Thus, researchers have invested a lot of efforts in the manual creation of such lexicons, which were typically of small to moderate size, e.g., less than 10,000 words. Recently, there has been a shift towards using automatically extracted lexicons, which are orders of magnitude larger and perform much better. These lexicons are typically mined using bootstrapping, starting from very few seed words whose polarity is given, e.g., 50-60 words, and sometimes even just 5-6. Here, we demonstrate that sizable further performance gains can be observed by starting with mid- sized seeds (hundreds of words and phrases), thus getting the …

Content truncated to 500 words