|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Distributed Indexing and Searching
P2P Search and Recommendation
Participants : Esther Pacitti, Maximilien Servajean.
In crossdiscipline domains, users belonging to different communities produce various scientific material that they own, share, or endorse. In that context, we are interested in querying and recommending scientific material in the form of documents. Such documents cover various topics such as models for plant phenotyping, statistics on specific kinds of plants, or biological experiments.
In [40] , we investigate profile diversity, a novel idea in searching scientific documents. Combining keyword relevance with popularity in a scoring function has been the subject of different forms of social relevance. On the other hand, content diversity has been thoroughly studied in search and advertising, database queries, and recommendations.
We introduce profile diversity for scientific document search as a complement to traditional content diversity. Profile diversity combines the discipline and communities to which a user belongs. We propose an adaptation of Fagin’s threshold-based algorithms to return the most relevant and most popular documents that satisfy content and profile diversities. To validate our scoring function, DivRSci, we ran experiments that use two benchmarks: a realistic benchmark with scientists and TREC’09. We show that DivRSci presents the best compromise between all requirements we have identified. DivRSci also shows to be the best generating list of inter-disciplinary and inter-community documents. Finally, it yields very good gains (by a factor of 6), suited for profile diversification
Spatial Queries in Wireless Data Broadcasting
Participant : Patrick Valduriez.
The main requirements for spatial query processing via mobile terminals include rapid and accurate searching and low energy consumption. Most location-based services (LBSs) are provided using an on-demand method, which is suitable for light-loaded systems where contention for wireless channels and server processing is not severe. However, as the number of users of LBSs increases, performance deteriorates rapidly since the servers’ capability to process queries is limited. Furthermore, the response time of a query may significantly increase with the concentration of users’ queries in a server at the same time. That is because the server has to check the locations of users and potential objects for the final result and then individually send answers to clients via a point-to-point channel. At this time, an inefficient structure of spatial index and searching algorithm may incur an extremely large access latency.
To address this problem, we propose in [27] the Hierarchical Grid Index (HGI), which provides a light-weight sequential location-based index structure for efficient LBSs. We minimize the index size through the use of hierarchical location-based identifications. And we support efficient query processing in broadcasting environments through sequential data transfer and search based on the object locations. We also propose Top-Down Search and Reduction- Counter Search algorithms for efficient searching and query processing. HGI has a simple structure through elimination of replication pointers and is therefore suitable for broadcasting environments with one-dimensional characteristics, thus enabling rapid and accurate spatial search by reducing redundant data. Our performance evaluation shows that our proposed index and algorithms are accurate and fast and support efficient spatial query processing.