In information retrieval systems, the match between query and document is severely unbalanced, due to the huge difference in the sizes of query and document. The ability to strengthen the query with related knowledge, such as context, should improve the search result. Contextual information can help reveal the semantics beneath the query terms, or specify the facets of the topic.

In this paper, the authors present a new context-based search approach. It assumes a search is started from a document that the user is reading, and the context is based on words surrounding the phrase that the user has selected as a query. This assumption is quite reasonable, since a large number of queries occur in this situation. This approach is not applicable, however, for other situations in which the user is comparing multiple documents, or has made a mental leap while reading.

The IntelliZap system used in this paper captures and analyzes context automatically, and sends the augmented query to some general or domain-specific search engines. The results are re-ranked and synthesized into final results. Experiments show that this outperforms other major search engines when the number of query terms is small, but it seems that the advantage shrinks fast when the term number increases. Readers would benefit from understanding this query-performance curve, so they can reconsider using the system in situations where the extra processing cost is not worth the small performance improvement.