Monday, July 23, 2007

Understanding user intent when presenting search results

A paper "Random Walks on the Click Graph" by Nick Craswell, and Martin Szummer both from Microsoft Research will be presented SIGIR 2007.

“We are trying to find,” Craswell explains, “a high-quality set of synonymous queries and results. Because our work is based on a large set of user click data, we’re getting a handle on user intent. Whereas click data can only go from a precise query to a set of documents, we find larger clusters of related queries and information.”

Click data, on which the project relies, provides the means to analyze the results for a particular query. The engine devised by Craswell and Szummer considers the history of all the people who ever typed that query and the preference of those users according to which results were often clicked. The history of which results were selected for which queries is click data.

“Click-based methods,” Craswell notes, “are a very good way of understanding the dominant intent.”

In doing so, the researchers are able to take a look at queries that are synonymous, but “synonymous queries,” as it turns out, are not quite what you might expect.

“Interestingly,” Craswell says, “the definition of what’s a synonymous query is different from synonyms you would find in a thesaurus. It’s very specific to search and depends on understanding user intent.” via Microsoft Research

No comments: