Encoding Adjective Scales for Fine-grained Resources

Encoding Adjective Scales for Fine-grained Resources Cédric Lopez*, Frédérique Segond*, Christiane Fellbaum** * Viseo Technologies, 4 avenue Doyen Louis Weil, Grenoble (France) ** Princeton University, Computer Science Department, 35 Olden Street, Princeton (USA) E-mail: cedric.lopez@viseo.com, frederique.segond@viseo.com, fellbaum@princeton.edu Abstract We propose an automatic approach towards determining the relative location of adjectives on a common scale based on their strength. We focus on adjectives expressing different degrees of goodness occurring in French product (perfumes) reviews. Using morphosyntactic patterns, we extract from the reviews short phrases consisting of a noun that encodes a particular aspect of the perfume and an adjective modifying that noun. We then associate each such n-gram with the corresponding product aspect and its related star rating. Next, based on the star scores, we generate adjective scales reflecting the relative strength of specific adjectives associated with a shared attribute of the product. An automatic ordering of the adjectives “correct” (correct), “sympa” (nice), “bon” (good) and “excellent” (excellent) according to their score in our resource is consistent with an intuitive scale based on human judgments. Our long-term objective is to generate different adjective scales in an empirical manner, which could allow the enrichment of lexical resources. Keywords: adjectives scales, resource, opinion mining 1. Introduction Adjectives modify entities (expressed by nouns or noun phrases) by providing additional information about the entities’ properties. Some properties (or attributes) are gradable. For example, degrees of “coldness” can be expressed by modifying the adjective “cold” with an adverb, as in “rather cold”, “very cold”, etc. But languages usually also have adjectives that encode such degrees; English distinguishes, “arctic,” “icy” and “chilly”. Typically, such adjectives fall on two scales, each centered about a polar adjective (e.g., “hot and “cold”); scales with polar adjectives can also be united into a single scale. We can reasonably assume that speakers know where on the scale of “coldness” and “hotness” the relevant English adjectives fall; at least, they are likely to agree on the relative position of two adjectives. But can one empirically determines this? Our work focuses on a subset of gradable adjectives, evaluative adjectives such as “good,” “great,” “terrible” and “awful.” Our long-term objective is to generate different adjective scales in an empirical manner, which could allow the enrichment of lexical resources. Such a resource could be valuable for opinion mining, especially for the prediction of rating scores of reviews (Ifrim & Weikum, 2010) (Liu & Seneff, 2009). 2. Related Work Evaluative adjectives express a “sentiment” of the speaker/writer with respect to the entity to which the adjective is applied. Thus, “a great car” expresses a positive sentiment, while “a lousy car” expresses a negative sentiment. Sentiment Analysis is often performed with the help of a lexical resource where entries are annotated with sentiment values. For example, SentiWordNet (Baccianella et al., 2010) is the result of automatically annotating all WordNet synsets (Miller, 1995) (Fellbaum, 1998); Hu and Liu (2004) is a list of 6,800 words annotated for sentiment. A simple model may determine the sentiment of a document based …

Content truncated to 500 words