Query Portals: Dynamically Generating Portals for Web Search Queries

Many informational web search queries seek information about named entities residing in structured databases. The information about entities that a user is seeking could be available from the structured database, or from other web pages. For example, consider an informational query such as [lowlight digital cameras]. A large structured database consisting of products may contain several products that are relevant for the user’s query. Surfacing a set of the most relevant products to the user and then enabling her to obtain more information about one or more of them would be very useful. Ideally, we would like a portallike functionality which provides an overview of all relevant products, and further allows drill down on them. The need for structured data search is illustrated by the proliferation of vertical search engines for products [3, 1], celebrities [2], etc. Current web search engines already federate queries to one or more structured databases containing information about named entities products, people, movies and locations. Each structured database is searched individually and the relevant structured data items are returned to the web search engine. The search engine gathers the structured search results and displays them along side the web search results. However, this approach does not enable a portal-like functionality due to two key limitations. Incomplete Results: Current federated search over each structured database is “silo-ed” in that it exclusively uses the information in the structured database to find matching entities. That is, the query keywords are matched only against the information in the structured database. The results from the structured database search are therefore independent of the results from web search. We refer to this type of structured data search as silo-ed search. While the silo-ed search works well for some queries, it would return incomplete or even empty results for a broad class of queries. Consider the query [lowlight digital cameras] against a product database containing the name, description, and technical specifications for each product. Canon EOS Digital Rebel Xti may be a relevant product but the query keyword {lowlight} may not occur in its name, description or technical specifications. Silo-ed search over the above product database would fail to return this relevant product. Reviews of the product may describe it using those keywords and can help deduce that the product is relevant to the query. However, the structured database may not contain the com-