Subjectivity Word Sense Disambiguation

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 190–199, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Subjectivity Word Sense Disambiguation Cem Akkaya and Janyce Wiebe University of Pittsburgh {cem,wiebe}@cs.pitt.edu Rada Mihalcea University of North Texas rada@cs.unt.edu Abstract This paper investigates a new task, subjec- tivity word sense disambiguation (SWSD), which is to automatically determine which word instances in a corpus are being used with subjective senses, and which are be- ing used with objective senses. We pro- vide empirical evidence that SWSD is more feasible than full word sense dis- ambiguation, and that it can be exploited to improve the performance of contextual subjectivity and sentiment analysis sys- tems. 1 Introduction The automatic extraction of opinions, emotions, and sentiments in text (subjectivity analysis) to support applications such as product review min- ing, summarization, question answering, and in- formation extraction is an active area of research in NLP. Many approaches to opinion, sentiment, and subjectivity analysis rely on lexicons of words that may be used to express subjectivity. Examples of such words are the following (in bold): (1) He is a disease to every team he has gone to. Converting to SMF is a headache. The concert left me cold. That guy is such a pain. Knowing the meaning (and thus subjectivity) of these words would help a system recognize the negative sentiments in these sentences. Most subjectivity lexicons are compiled as lists of keywords, rather than word meanings (senses). However, many keywords have both subjective and objective senses. False hits – subjectivity clues used with objective senses – are a signifi- cant source of error in subjectivity and sentiment analysis. For example, even though the follow- ing sentence contains all of the negative keywords above, it is nevertheless objective, as they are all false hits: (2) Early symptoms of the disease include severe headaches, red eyes, fevers and cold chills, body pain, and vomiting. To tackle this source of error, we define a new task, subjectivity word sense disambigua- tion (SWSD), which is to automatically determine which word instances in a corpus are being used with subjective senses, and which are being used with objective senses. We hypothesize that SWSD is more feasible than full word sense disambigua- tion, because it is more coarse grained – often, the exact sense need not be pinpointed. We also hy- pothesize that SWSD can be exploited to improve the performance of contextual subjectivity analy- sis systems via sense-aware classification. The paper consists of two parts. In the first part, we build and evaluate a targeted supervised SWSD system that aims to disambiguate members of a subjectivity lexicon. It labels clue instances as having a subjective sense or an objective sense in context. The system relies on common machine learning features for word sense disambiguation (WSD). The performance is substantially above both baseline and the performance of full WSD on the same data, suggesting that the task is feasi- ble, and that subjectivity provides a natural coarse- grained …

Content truncated to 500 words