PowerPoint Presentation Problem Automatic determination of the most interesting segments in the content of a given form of media (i.e. video, audio, or webpages). Significance Used for generating automatic preview information based on the most important segments. Allows for faster and more efficient extraction of important data within large media collections (i.e. segments of a single file or files in a collection). Helps users to avoid long downloads, especially when searching for specific content over a collection of files. Removes guesswork for content designers looking to focus the attentions of end users. Current Approaches Finite rule based techniques where each condition has a static meaning. This has primarily been done with website analysis software. References and Acknowledgements Tanveer Syeda-Mahmood Ph.D. IBM Almaden Research Center University of Nevada, Reno George Bebis Ph.D. Dwight Egbert Ph.D. The National Science Foundation UNR Office of Research UNR Department of Computer Science Results Using a Hidden Markov Model as a machine learning foundation, we were able to develop two products that demonstrate the application of smart user interest analysis technology. VideoCharger Plus - a multimedia content interest level analyzer (based on IBM's VideoCharger software) VideoCharger Plus main viewing window VideoCharger Plus User Interest Graph Displays user interest level changes with time Instantaneous user state display Detects events from player interactions (i.e. play, pause, stop, fast forwarding, slow scanning, etc.) For each event, the learning model is used to determine which state (or interest level) best describes the user's behavior. Interesting segments can be analyzed across users. Web Analyzer - a webpage content interaction analyzer (uses IBM's WBI technology) From the detection (through proxy embedded JavaScript) of URL information, time spent on each page, scrolling, form submissions, advertisement following, etc. Web Analyzer is able to detect both user interest levels as well as general user action meaning. The system can distinguish between a user that is searching, browsing, or aimlessly wondering the Internet. Using this knowledge, the system can further infer whether a user is becoming frustrated or likes what he/she is viewing. Can be used to target user or scenario specific information. This is beneficial to both users and site operators.