Selected Projects
The project in this paper is the subproject in a huge research project, involving the Universities of British Columbia, Ottawa, and Toronto as well as SAP/Business Objects, aims at conducting a proof-of-concept study of a requirements-driven data warehouse construction. It is the phase for bidirectional transition between the Cube ER visualization and internal representation Cube EDM.
This paper proposes a way to do parallel query processing based on distributed inverted index. The conventional approach of search engines is to construct a complete inverted index and query on the inverted index in a high-computation ability machine. However, this approach suffers from two problems, one is the space limit and the other is the slow response time, particularly in a large scale. In this paper, we propose to divide the inverted index and distribute them to several nodes in a cluster, and then query on these nodes in a parallel way.
This paper presents a Hybrid Document Oriented Model (Hybrid DOM), which is more suitable for SOAP intermediaries in distributed SOA computing environments. Based on an event-driven model, the Hybrid DOM can parse XML document into DOM tree on the fly, and serialize in an efficient and lazy way. Compare with the original paper, this paper gives much more comprehensive evaluations and analysis in terms of performance under various parsing and serializing cases. In addition, several contemporary mainstream XML parsers are examined and compared to Hybrid DOM.
With the development of Web2.0 in the past few years, social annotation has drawn more and more attention. By studying the link relationships among them, we attempt to optimize the search results with this net information. Inspired by the idea of collaborative filtering, for the first time we propose a model based on the classification of users and voting with weights. In this model, we firstly identify similar users with similar interests then these users will vote for the web pages with their weights calculated by the user similarity. Therefore, web pages with more ballots from users whose similarity to the query user is higher will be recommended. At the same time, we also calculate the similarity between pages with the links to tags in order to select representative pages.
