Tuesday, 5/22/2007
10:30 AM - 11:30 AM
Level: Technical-Intermediate/Case Study

CNET Networks has over a dozen websites, which in aggregate make up the 10th largest Web property in the US. A key resource to users of all these sites is the offering of "related searches," which suggest, for any given query, some alternative ways of retrieving similar information. Alas, all the sites are "long tail" sites and therefore suffer from a high degree of "cold start" blockage when it comes to generating related searches via collaborative filtering (CF). In other words, "users who searched for X also searched for Y" can't be done for most of the queries seen on CNET sites, due to a sparseness of examples. TextDigger, a start-up company developing a semantic search tool, has jointly conducted a case study with CNET to overcome the cold start problem on related searches in two ways: (1) where CF can produce no related searches, a semantically related search can be provided as an alternative; (2) the semantic tool can be used to "bridge the gap" between a query whose examples are too sparse and another, semantically similar query whose examples are plentiful - thus arriving at related searches which neither CF nor the semantic search tool would have produced working alone. We present preliminary results of three approaches to generating related searches (CF alone, semantics alone, CF and semantics combined) as regards coverage, impressions, and click-thrus. We also explain how the semantic search tool had to be adjusted within certain parameters in order to make the case study practical to implement.

Dr. Tim Musgrove is the Founder & CEO of TextDigger, Inc., a Silicon Valley startup developing a horizontal, consumer-facing semantic search engine, Digger.com. TextDigger was spun out of CNET Networks, where Tim was the Senior Research Fellow for Artificial Intelligence. At CNET his team developed patented linguistic technologies which today are used to auto-generate thousands of natural language texts posted on CNET's award winning websites. Tim's background areas of ontology, philosophy of language, and cognitive semantics are significantly visible in his development projects at TextDigger.