31.Oct.2009 The new Google patent granted: Rank adjustments via click data
Imagine that a number of people use Google to perform a search for “orange,” and then “banana,” and then “pineapple” and then choose the web page “http://www.example.com/fruit.htm” in the search results they see.
Now imagine that Google looks at the information it collects about what people do when they search, and finds in its query logs and click logs that there are a large number, a statistically significant number, of people who search for “orange,” and then “banana,” and then “pineapple,” or possibly the same search terms in a slightly different order, and then tend to click on “http://www.example.com/fruit.htm.”
Google may also notice that there are people looking for some very related terms during query sessions, such as consecutive searches for “banana,” “apple” and “pineapple.”
Since this second set of queries for “banana,” “apple” and “pineapple,” is so similar to the query sessions that contained the search terms “orange” and “banana” and “pineapple,” where people were choosing the page “http://www.example.com/fruit.htm,” Google may choose to adjust the ranking for “http://www.example.com/fruit.htm,” for people using those very related terms in their search sessions.
Google was granted a patent on this process this past week:
Rank-adjusted content items
Invented by Mayur Datar, Kedar Dhamdhere, and Ashutosh Garg
Assigned to Google
US Patent 7,610,282
Granted October 27, 2009
Filed March 30, 2007
Abstract
Click logs and query logs are processed to identify statistical search patterns. A search session is compared to the statistical search patterns. Content items responsive to a query of the search session are identified, and a ranking of the content items is adjusted based on the comparison.
Content items, e.g., video and/or audio files, web pages for particular subjects, news articles, etc., can be identified by a search engine in response to a query. The query can include one or more search terms, and the search engine can identify and rank the content items based on the search terms in the query. Typically the content items are displayed according to the rank.
The content items, however, are often identified only in response to a particular query, i.e., the search engine may identify and rank content items to independently for each query. For example, for three different queries, the search, engine may return a particular identification and rank of content items for each particular query, regardless of the other queries. In such implementations, a particular content item that may be highly relevant to a user’s current interests may not be identified and/or highly ranked and presented to the user until the user has conducted multiple searches. Additionally, other users may experience similar challenges when searching for content.
SUMMARY
Disclosed herein are systems and methods of identifying content items. In one implementation, click logs and query logs are processed to identify statistical search patterns based on the click logs and query logs. A search session is compared to the statistical search patterns. Content items responsive to a query of the search session are identified, and a ranking of the content items is adjusted based on the comparison.
In another implementation, query paths and content terminuses associated with query paths are identified. Additionally, a context of a search session is identified and a determination of whether the context is related, to one or more of the query paths is made. Content items responsive to a query of the search session are identified based on the determination.
In another implementation, a system includes a mining engine and an adjusting engine. The mining engine mines click logs and query logs to identify query paths and content terminuses associated with the query paths. The adjusting engine adjusts a ranking of content items responsive to a search session query based on the identified query paths and content terminuses.
In one implementation, identification of a context of a search session facilitates the adjusting of a ranking of one or more content items in response to a search session query. The adjustment can, for example, be based on the likelihood that a current user is searching for the rank-adjusted content items because a statistically significant number of prior users that exhibited a similar behavior to the current user selected the rank-adjusted content items.
