Google Inc.

Jeremy Bem

Simon Tong

Noam M Shazeer

Georges R Harik

Joshua L Levenberg

A system ranks documents based, at least in part, on a ranking model. The ranking model may be generated to predict the likelihood that a document will be selected. The system may receive a search query and identify documents relating to the search query. The system may then rank the documents based, at least in part, on the ranking model and form search results for the search query from the ranked documents.

Ranking documents based on large data sets

A system for generating a model is provided. The system generates, or selects, candidate conditions and generates, or otherwise obtains, statistics regarding the candidate conditions. The system also forms rules based, at least in part, on the statistics and the candidate conditions and selectively adds the rules to the model.

Large scale machine learning systems and methods

The number of ads potentially relevant to search query information may be increased by relaxing the notion of search query keyword matching. This may be done, for example, by expanding a set of ad request keywords to include both query keywords (or derivatives of a root thereof) and related keywords. The related keywords may be words with a relatively high co-occurrence with a query keyword in a group of previous search queries (e.g., search queries in a session). The scores of ads with keyword targeting criteria that matched words related to words in a search query, but not the words from the search query, may be discounted. That is, the scores of ads served pursuant to a relaxed notion of matching may be discounted relative to the scores of ads served pursuant to a stricter notion of matching. This may be done by using a score modification parameter, such as an ad performance multiplier (for cases in which an ad score is a function of ad performance information). The score modification parameter may be updated to reflect observed performance data, such as performance data associated with {word-to-related word} mappings.

Increasing a number of relevant advertisements using a relaxed match

Model generation for ranking documents based on large data sets

The number of ads potentially relevant to search query information may be increased by relaxing the notion of search query keyword matching. This may be done, for example, by expanding a set of ad request keywords to include both query keywords and related keywords. The scores of ads served pursuant to a relaxed notion of matching (those with keyword targeting criteria that matched words related to words in the search query, but not the words from the search query) may be discounted relative to the scores of ads served pursuant to a stricter notion of matching. This may be done by using a score modification parameter, such as an ad performance multiplier (when an ad score is a function of ad performance information) The score modification parameter may be updated to reflect observed performance data, such as that associated with {word-to-related word} mappings.

If some aspect of serving or scoring an ad is subject to a performance (e.g., click-through rate, etc.) threshold, such a threshold may be adjusted using a confidence factor of the ad targeting used. For example, ads served pursuant to a more relaxed notion of match might have to meet a higher performance threshold (e.g., than the threshold applied to ads served pursuant to a stricter notion of match). Alternatively, or in addition, ads served pursuant to a stricter notion of match might be subject to a lower performance threshold (e.g., than the threshold applied to ads served pursuant to a more relaxed notion of match). Thus, in general, a performance threshold could increase as match confidence decreases, and/or a performance threshold could decrease as match confidence increases.

Growing community of inventors

Berkeley, CA, United States of America

Jeremy Bem