An Automatic Personalized Context-Aware Event Notification System for Mobile Users: An Automatic Personalized Context-Aware Event Notification System for Mobile Users George Lee User Context-based Service Control Group Network Laboratories NTT DoCoMo R&D


Overview: Overview The Problem: Mobile users cannot easily get desired information Proposed solution: automatic, personalized, context-aware event notification approach Matching Engine to match users and events User Agent to learn user interests


Mobile users can’t easily get relevant information: Mobile users can’t easily get relevant information Relevant information is: Appropriate for their context Personalized based on individual interests Current and up-to-date Static menu is inadequate Too many choices Difficult to navigate Not personalized or context-aware Information retrieval has drawbacks Requires queries Not good for new or changing information


Automatic, personalized, context-aware event notification: MIT News CSAIL News The Tech Boston Dining News Italian Restaurants … Central Sq. Sports Red Sox Scores … Automatic, personalized, context-aware event notification Context: Going to lab CSAIL News: Talk at 3pm G825 Central Sq. Dining: New café opening Red Sox vs. Yankees: 4-3 (6th inning) Mobile Handset Automatic Personalized Context-aware


Matching events and learning user interests: Event Matching events and learning user interests User User User Event Event Matching Engine User Agent User input Event description User interests User Agent automatically learns user interests for the current context based on user input Matching Engine decides which users match an event based on event descriptions and user preferences


Describing events and user interests using an event model: Describing events and user interests using an event model matches Events and user interests are described and matched according to an Event Model Problem: existing event notification systems do not work well with complex event models


Choosing an appropriate event model: expressiveness vs. efficiency: Choosing an appropriate event model: expressiveness vs. efficiency Expressiveness Matching Speed Flat (e.g. Mailing lists) Content-based (e.g. XPath) Hierarchical (e.g. Newsgroups) Graph-structured (e.g. Yahoo!) Can we improve the matching efficiency of graph-structured event models?


Regular matching: Regular matching Event topic: “Red Sox” Matches all users with “Red Sox” as a subtopic in their interests: (Red Sox, Boston Sports, Baseball, Sports, and All) Sports Baseball Red Sox Yankees All … … Boston Sports … Event Topic: Red Sox Must search graph to find related topics


Optimized matching: Optimized matching Compute a table of all supertopics of each topic (transitive closure) Sports Baseball Red Sox Yankees All … … Boston Sports … Event Topic: Red Sox Gets all related topics in one table lookup


Evaluation of efficient matching : Evaluation of efficient matching Objective: Evaluate optimized matching with many users and a complex event model Event Model Topic: Open Directory Project (ODP) Location: Getty Thesaurus of Geographic Names (TGN) 44,506 topics, 6905 locations Simulated users 100 to 100,000 users Interests include 5 topics and 3 locations Simulated events Contain 3 random topics and 2 random locations


Slide11: Efficient: Optimized matching is 30 times faster than unoptimized matching Expressive: Works well with complex event models with 45,000 topics and 7000 locations Scalable: Can match 10,000 users in less than 10 seconds


A user agent for learning user interests: Event A user agent for learning user interests User User User User Agent automatically learns user interests for the current context based on user input Event Event Matching Engine User Agent User input Event description User interests Context Server Challenges: Implicitly learning user interests Recommending topics in new contexts


Learning and automatically updating user interests: User Agent Mobile Handset Learning and automatically updating user interests Event List Topic 1 Topic 2 Topic 3 Topic Rating Learner Selected Topics User Interests Matched events Matching Engine Context Server Topic Recommender Automatically recommends new topics based on ratings of past topics Implicitly learns user ratings for topics based on user selections


Recommending topics: Recommending topics Recommendations needed for new topics and contexts Possible approaches: Popularity: not personalized Rating History: recommendations based on previous topic ratings Collaborative Filtering (CF): recommendations based on interests of users with similar interests


Context-aware Collaborative Filtering: Context-aware Collaborative Filtering User X User A User D Is User X interested in “MIT News” for context “Go to lab”? User C User B To calculate a recommendation for topic T in context C: Find users who have rated topic T under context C Find users with similar interests Decide whether to recommend topic T based on ratings of similar users


Enhanced Context-aware Collaborative Filtering: Enhanced Context-aware Collaborative Filtering Is User X interested in “MIT News” for context “Go to lab”? User X User A User D Yes Yes Yes Yes Yes Model relationships between topics and contexts when calculating user similarity Give greater weight to similar topics and contexts (e.g. give greater weight to same topic and same context)


Recommender Evaluation: Recommender Evaluation Evaluate ability of Enhanced CF to provide relevant information in a new context User Interface: app on mobile handset 16 test subjects 8 for data collection 8 for evaluation 50 topics based on i-mode services 2 contexts Going to see a movie in Tokyo Going to Tokyo Disneyland 10 topics per recommender Interleave topics from two recommenders and observe which topics users selected vs. Random vs. Rating History vs. Regular CF Recommender A Recommender B


Slide18: Effective: Enhanced CF can recommend relevant topics in new contexts Compared to other approaches, enhanced CF topics selected 413% more than Random topics 49.7% more than Rating History topics 24.8% more than Regular CF topics More studies needed to increase confidence


Conclusion: Conclusion I proposed an event notification system for mobile users Automatic Personalized Context-aware Research contributions Optimized content-graph event matching algorithm Enhanced context-aware collaborative filtering topic recommender Future work Distributed architectures Learning and recommendation algorithms Context models User studies