A comparative evaluation of term weighting methods for information filtering

Users of information filtering systems can not be expected to provide large amounts of information to initialize a profile. Therefore, term weighting methods for information filtering have somewhat different requirements to those for information retrieval and text categorization. We present a comparative evaluation of term weighting methods, including a new method, relative document frequency, designed specifically for information filtering. The best weighting methods appear to be those that favor information provided by the user, over information from a general collection.