This SBIR Phase I project from Unconventional Wisdom aims to build a recommendation system for the World Wide Web (WWW). Finding relevant information on the WWW is becoming increasingly difficult. Most current search engines produce too much irrelevant information because they search syntactically. This proposal shows how to produce more accurate and personally-relevant search recommendations through three key innovations: a method to learn and combine user information (e.g. demographics and ratings) with WWW source information (e.g. web page features); a method to leverage the power of human-categorized information sources on the WWW; and a method to learn mega-features, that is, large sets of homogeneous features through running a machine-learning algorithm over the sets. In Unconventional Wisdom's framework, people implicitly assign semantics to web pages by their pattern of usage and machines act as the transporters of those semantics to other users so that retrieval can be based on semantics implicit in the user's interests. This makes it possible to retrieve complex multimedia information based on the user's interests. Unconventional Wisdom proffers a technology addressing a significant opportunity in e-commerce. Since a web site can contain information such as text, audio, video or catalog items, the system envisioned by Unconventional Wisdom is flexible in its application to both the WWW and military or corporate Intranets. The area of research and development that this project treats has important implications for knowledge management in general, such as in developing packaged inference mechanisms to address information selection of many types and in systems to generate and support relationships among persons with related information requirements and values.
