Adaptive Information Filtering Based on PTM Model (APTM)

Adaptive information filtering (AIF) is a challenging issue for web search, as the Web contains non-structured data used by many different users. One of the main questions in AIF is how to keep the system up-to-date over time by increasing training on line with changes in the userpsilas needs and updating the parameters values accordingly. This paper investigates the use of Pattern Taxonomy Models (PTM) in adaptive filtering by adding an updating feature. We developed a mathematical model that updates training based on sliding windows over the positive and negative examples. Merging the scores of documents in the new windows with the old score of the system takes in to account the size of the training window and the type of document in each window. In order to test this approach, the mathematical model was implemented and tested with RCV1 data collection. The experimental results indicated that the new system improves performance of PTM.