Separating Brands from Types: an Investigation of Different Features for the Food Domain

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 2291–2302, Dublin, Ireland, August 23-29 2014. Separating Brands from Types: an Investigation of Different Features for the Food Domain Michael Wiegand and Dietrich Klakow Spoken Language Systems Saarland University D-66123 Saarbr¨ucken, Germany {Michael.Wiegand|Dietrich.Klakow}@lsv.uni-saarland.de Abstract We examine the task of separating types from brands in the food domain. Framing the problem as a ranking task, we convert simple textual features extracted from a domain-specific corpus into a ranker without the need of labeled training data. Such method should rank brands (e.g. sprite) higher than types (e.g. lemonade). Apart from that, we also exploit knowledge induced by semi- supervised graph-based clustering for two different purposes. On the one hand, we produce an auxiliary categorization of food items according to the Food Guide Pyramid, and assume that a food item is a type when it belongs to a category unlikely to contain brands. On the other hand, we directly model the task of brand detection using seeds provided by the output of the textual ranking features. We also harness Wikipedia articles as an additional knowledge source. 1 Introduction Brands play a significant role in social life. They are the subject matter of many discussions in social me- dia. Their automatic detection for information extraction tasks is a pressing problem since, despite their unique property to refer to commercial products of specific companies, in everyday language they often occur in similar contexts as common nouns. A typical domain where such behaviour can be observed is the food domain, where food brands (e.g. nutella or sprite) are often used synonymously with the food type1 of which the brand is a prototypical instance (e.g. chocolate spread or lemonade). Such usage is illustrated in (1) and (2). (1) In the evening, I eat a slice of bread with either nutella or marmalade. (2) I prepare my pancakes with baking soda, water and a lacing of sprite instead of sugar. This particular phenomenon of metonymy (Lakoff and Johnson, 1980), commonly referred to as generi- cized trademarks, of course, has consequences on automatic lexicon induction methods. If one automat- ically extracts food types, one also obtains food brands. In this paper, we examine features to detect brands automatically. Solving the issue with the help of a manually-compiled list of brands neglects parts of the nature of brands. Brands come and go. Some products may be discontinued after a certain amount of time (e.g. due to limited popularity) while, on the other hand, new products constantly enter the market. For instance, popular food brands, such as sierra mist or kazoozles, did not exist a decade ago. Therefore, a list of brands that is manually created today may not reflect the predominant food brands that will be available in a decade. The features we introduce to detect brands consider both the intrinsic properties of brands and their contextual environment. Even though in many contexts, brands are used as ordinary type expressions (1), there might be specific contexts …

Content truncated to 500 words