Augmenting Business Entities with Salient Terms from Twitter

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 121–129, Dublin, Ireland, August 23-29 2014. Augmenting Business Entities with Salient Terms from Twitter Riham Mansour Microsoft Research ATL Cairo, Egypt rihamma@microsoft.com Nesma Refaei Cairo University Cairo, Egypt nesma.a.refaei@eng.cu.edu.eg Vanessa Murdock Microsoft Seattle, WA vanessa.murdock@yahoo.com Abstract A significant portion of search engine queries mention business entities such as restaurants, cin- emas, banks, and other places of interest. These queries are commonly known as “local search” queries, because they represent an information need about a place, often a place local to the user. A portion of these queries is not well served by the search engine because there is a mismatch be- tween the query terms, and the terms representing the local business entity in the index. Business entities are frequently represented by their name, the category of entity (whether it is a restaurant, an airport, a grocery store, etc.) and other meta-data such as opening hours and price ranges. In this paper, we propose a method for representing business entities with a term distribution gener- ated from web data and from social media that more closely aligns with user search query terms. We evaluate our system with the local search task of ranking businesses given a query, in both the U.S. and in Brazil. We show that augmenting entities with salient terms from social media and the Web improves precision at rank one for the U.S. by 18%, and for Brazil by 9% over a competitive baseline. For precision at rank three, the improvement for the U.S. is 19%, and for Brazil 15%. 1 Introduction Search engine queries, particularly queries issued from mobile devices, often mention business entities such as restaurants, cinemas, banks, and other places of interest. These “local search” queries represent an information need about a place. Often there is a mismatch between the query terms, and the terms representing the local business entity in the index, making it difficult for the search engine to find results that satisfy the user. Local data consists largely of listings of businesses, annotated with metadata. This metadata includes the name of the location, category information (is the business a clothing retailer, or a Thai restaurant, for example), address and phone number, opening hours, and indicators such as price range, popularity, star ratings, etc. Figure 1 shows an example of the type of information available to local search systems. Some local search queries are known item searches, where the user knows the name of a business and they seek other information about the place, such as the opening hours. Other local search queries are category searches where the user does not know the name of a specific business but is using the Internet in much the same way they might have used the Yellow Pages in pre-Internet days. An example of a category search is “Thai restaurants in Denver”. There are also descriptive local queries such as “pizza delivery” or “romantic brunch in Seattle” where the user does …

Content truncated to 500 words