Contextual Phrase-Level Polarity Analysis Using Lexical Affect Scoring and Syntactic N-Grams

Proceedings of the 12th Conference of the European Chapter of the ACL, pages 24–32, Athens, Greece, 30 March – 3 April 2009. c⃝2009 Association for Computational Linguistics Contextual Phrase-Level Polarity Analysis using Lexical Affect Scoring and Syntactic N-grams Apoorv Agarwal Department of Computer Science Columbia University New York, USA aa2644@columbia.edu Fadi Biadsy Department of Computer Science Columbia University New York, USA fadi@cs.columbia.edu Kathleen R. Mckeown Department of Computer Science Columbia University New York, USA kathy@cs.columbia.edu Abstract We present a classifier to predict con- textual polarity of subjective phrases in a sentence. Our approach features lexi- cal scoring derived from the Dictionary of Affect in Language (DAL) and extended through WordNet, allowing us to automat- ically score the vast majority of words in our input avoiding the need for manual la- beling. We augment lexical scoring with n-gram analysis to capture the effect of context. We combine DAL scores with syntactic constituents and then extract n- grams of constituents from all sentences. We also use the polarity of all syntactic constituents within the sentence as fea- tures. Our results show significant im- provement over a majority class baseline as well as a more difficult baseline consist- ing of lexical n-grams. 1 Introduction Sentiment analysis is a much-researched area that deals with identification of positive, negative and neutral opinions in text. The task has evolved from document level analysis to sentence and phrasal level analysis. Whereas the former is suitable for classifying news (e.g., editorials vs. reports) into positive and negative, the latter is essential for question-answering and recommendation systems. A recommendation system, for example, must be able to recommend restaurants (or movies, books, etc.) based on a variety of features such as food, service or ambience. Any single review sentence may contain both positive and negative opinions, evaluating different features of a restaurant. Con- sider the following sentence (1) where the writer expresses opposing sentiments towards food and service of a restaurant. In tasks such as this, there- fore, it is important that sentiment analysis be done at the phrase level. (1) The Taj has great food but I found their ser- vice to be lacking. Subjective phrases in a sentence are carriers of sentiments in which an experiencer expresses an attitude, often towards a target. These subjective phrases may express neutral or polar attitudes de- pending on the context of the sentence in which they appear. Context is mainly determined by con- tent and structure of the sentence. For example, in the following sentence (2), the underlined subjec- tive phrase seems to be negative, but in the larger context of the sentence, it is positive.1 (2) The robber entered the store but his efforts were crushed when the police arrived on time. Our task is to predict contextual polarity of sub- jective phrases in a sentence. A traditional ap- proach to this problem is to use a prior polarity lexicon of words to first set priors on target phrases and then make use of the syntactic and semantic information in …

Content truncated to 500 words