Methods, apparatuses, and articles for receiving a collection of documents and/or objects determined to be potentially relevant to a keyword, and processing the collection of documents and/or objects to extract one or more keyword characterizations for use as proxies for the keyword, are described herein. In various embodiments, the one or more keyword characterizations may be used to compute a measure of keyword similarity for the keyword, facilitate keyword behavior modeling of the keyword, and/or find one or more advertisements.