Matching for mobile users in the publish/subscribe paradigm

In a mobile publish/subscribe paradigm, user service discovery and recommendation requires matching user preferences with properties of published services. For example, a user may want to find if there is a moderately priced Chinese restaurant that does not require reservations close by. To generate accurate recommendations, the properties of each user subscription must be matched with those of existing services as accurately as possible. This is a difficult problem when users are mobile, wirelessly connected to a network, and dynamically roaming to different locations. The available data is very large, and the matching must be computed in real time. Existing heuristics are quite ineffective. We propose novel algorithms that use singular value decomposition as a dimension-reduction technique. We introduce "positive" nearest-neighbor matching to find services whose attribute values exceed those of a new user subscription. Making this idea effective requires careful attention to details such as normalization. Performance and quality of matches are reported for datasets representing applications in the mobile publish/subscribe paradigm. For n services and m preference attributes, reasonable matches can be found in time O (m log n), using O (nm) storage.