Content-Based Mining of Query Replacements
Hung Viet Nguyen · 2008
Keyword searching is the most common form of product search on the Web. Most search engines do their text query and retrieval using keywords. The average keyword query length is under three words (2.2 words). Online advertisers are eager to find all relevant popular search phrases to promote their products. However, features in natural language such as synonymy, relations of terms, and perceptions of terms pose difficulties for advertisers. Vendors cannot anticipate all possible ways in which shoppers search for their products. The ultimate aim of search engines is to increase the number of relevant results and to reduce irrelevant results as much as possible. Many times, users' search query may not be a perfect description of their information needs. Even when the information is somewhat well described, a search engine or information retrieval system may not be able to retrieve documents matching the query as stated. For example, if the shopper uses motorcycle jacket then unless the publisher knows that every leather jacket can be a motorcycle jacket it cannot produce all matches for the user's search. Thus, for certain phrases, there is a semantic gap between the search phrase used and the way the corresponding matching products are described. A serious consequence of this gap is that the users are not satisfied. Thus there is a critical need to boost product findability by bridging the semantic gap that exists between search phrases and product descriptions. Closing this gap has the strong potential to translate web traffic into higher conversion rates and more satisfied user experience. In this dissertation, a model for terms correlation is introduced and a robust automated algorithm for exploring definition rules is presented. This algorithm can use a small corpus of keyword matching product descriptions for an advertiser's search-phrase to effectively mine and rank alternative but highly relevant search-phrase definitions. These definitions, which are search-phrases themselves, can then be provided as alternative phrases to an advertiser. Experimental evaluations are also provided with large datasets to demonstrate the effectiveness of the approach.