Sentibase: Sentiment Analysis in Twitter on a Budget

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 590–594, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics Sentibase: Sentiment Analysis in Twitter on a Budget Satarupa Guha ∗, Aditya Joshi ∗, Vasudeva Varma Search and Information Extraction Lab International Institute of Information Technology, Hyderabad Gachibowli, Hyderabad, Telengana, India {satarupa.guha,aditya.joshi}@research.iiit.ac.in vv@iiit.ac.in Abstract Like SemEval 2013 and 2014, the task Sen- timent Analysis in Twitter found a place in this year’s SemEval too and attracted an un- precedented number of participations. This task comprises of four sub-tasks. We partici- pated in subtask 2 — Message polarity classi- fication. Although we lie a few notches down from the top system, we present a very simple yet effective approach to handle this problem that can be implemented in a single day! 1 Introduction Social media not only acts as a proxy for the real world society, it also offers a treasure trove of data for different types of analyses like Trend Analysis, Event Detection and Sentiment Analysis, to name a few. SemEval 2015 Task 10 subtask B (Rosen- thal et al., 2015) specifically deals with the task of Sentiment Analysis in Twitter. Sentiment Analy- sis in social media in general and Twitter in par- ticular has a wide range of applications — Compa- nies/services can gauge the public sentiment towards the new product or service they launched, political parties can estimate their chances of winning the upcoming elections by monitoring what people are saying on Twitter about them, and so on. In spite of the availability of huge amount of data and the huge promises they entail, working with social media data is far more challenging than regular text data. Be- ing user-generated, the data is noisy; there are mis- spellings, unreliable capitalization, widespread use ∗The first two authors made equal contribution to this work of creative acronyms, lack of grammar, and a style of writing that is very typical of its own which makes the problem of Sentiment Analysis on Twitter more challenging. Also, the cues for positive or negative sentiment in social media text are starkly different, thereby generating a whole new domain for explo- ration. 2 Related Work SemEval 2013 (Nakov et al., 2013) and 2014 tasks (Rosenthal et al., 2014) on Sentiment Analysis in Twitter not only contributed to this field by mak- ing huge amounts of annotated datasets available for research, but also encouraged researchers to come up with better solutions for this challenging prob- lem. There has been numerous initiatives outside SemEval too. (Pak and Paroubek, 2010) is one of the early attempts at using Twitter as a corpus for Sentiment Analysis, which shows how to automat- ically collect a corpus for the same and performs linguistic analysis of the collected corpus. (Bakli- wal et al., 2012) presents a simple sentiment scoring function which uses prior information to classify and weight various sentiment bearing words/phrases in tweets. (Wilson et al., 2005) demontrates an efficient technique for automatically identifying the contex- tual polarity …

Content truncated to 500 words