IITPSemEval: Sentiment Discovery from 140 Characters

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 601–607, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics IITPSemEval: Sentiment Discovery from 140 Characters Ayush Kumar, Vamsi Krishna Akella and Asif Ekbal Department of Computer Science and Engineering Indian Institute of Technology, Patna Patna, 800013 Email: ayush.cs12@iitp.ac.in, vamsi.cs11@iitp.ac.in, asif@iitp.ac.in Abstract This paper presents an overview of the sys- tem developed and submitted as a part of our participation to the SemEval-2015 Task 10 that deals with Sentiment Analysis in Twitter. We build a Support Vector Machine (SVM) based supervised learning model for Subtask A (term level task) and Subtask B (message level task). We also participate in Subtask E viz., determining degree of polarity, and build a very simple system by employing the available lexical resources. Experiments with the 2015 official datasets show F1 scores of 81.31% and 58.80% for Task A and Task B, respectively. For Subtask E, our model achieves a score of 0.413 on Kendal’s Tau metric. 1 Introduction The use of social media platforms has become cen- tral to many teenager’s and adult’s lives. With the emerging forms of communication, much of the freely available texts in the opinionated texts are lin- guistically unstructured. People have adopted cre- ative spellings and abbreviations, and are exces- sively using more intelligent forms of messages that involves typos, hash-tags and emoticons to convey their messages. The huge abundance of inexpen- sive data, rich in applications, can prove handy for public and corporate institutions. This has urged the scientific community to extract the substantive in- formation from these texts. The proliferation of mi- croblogging sites like Twitter which boasts of user’s comments on everything trending in real time opens up an unprecedented opportunity to explore and de- velop techniques to mine the information. Task 10 in Semantic Evaluation 2015 provides a research platform promoting the knowledge dis- covery in Twitter. Task 10 consists of five dif- ferent subtasks: Contextual Polarity Disambigua- tion (A), Message Polarity Classification (B), Topic- Based Message Polarity Classification (C), Detect- ing Trends Towards a Topic (D) and Determining degree of polarity of Twitter terms with the senti- ment (E). Complete details of the task can be found at (Rosenthal et al., 2015). We participated in Sub- tasks A, B and E, the first two of which require the sentiments to be classified into positive, negative and neutral classes for a given segment of the tweet (for A) or the entire message (for B), while the Task E needs to compute the strength of association of the given terms to the sentiment on a scale of 0 to 1 with 1 denoting the maximum strength. The technical study of public sentiment has been a subject of trending research and a significant amount of extensive work is being carried out in the domain. Sentiment Analysis has been handled at the various levels of granularity. Early research works (Pang and Lee, 2004) focussed on the document level classification with further studies at message and …

Content truncated to 500 words