Professor of Computer Science
Center for Advanced Computer Studies
University of Louisiana at Lafayette
Lafayette, LA 70504-4330
Phone: (337)482-6603
Fax: (337)482-5791
Email: raghavan@cacs.louisiana.edu
Dr. Raghavan is a Distinguished Professor of Computer Science in The Center for Advanced Computer Studies at the University Louisiana at Lafayette (UL Lafayette), where he joined in 1986. He was an Associate Professor at UL Lafayette between 1986-88. Prior to that, he was with the Computer Science Department at the University of Regina, Regina, Canada for nine years. During 1983-84, Dr. Raghavan was at the Institut fur Informatik, Technische Universitat in W. Berlin and at the Computer Center, University Malaya in Kuala Lumpur as a Visiting Professor.
Dr. Raghavan has expertise in several areas of computer science such as Information Retrieval, DBMS, Image Database Management and Design and Analysis of Algorithms. His research activities address various approaches to designing effective retrieval strategies and access structures. In particular, he has developed strategies through which the performance of a retrieval system can be improved through prior interactions with its users. The techniques involve various kinds of connectionist learning paradigms and the application domain includes both text and image databases. He has investigated various aspects of evaluating information retrieval system performance. His research contributions also include work in traditional business databases, where he developed approaches for multiple query processing, index selection and database mining. Dr. Raghavan has published widely in international conferences, workshops, and technical journals. Raghavan guest edited a special issue of the September 1995 issue of the IEEE Computer magazine on "Content-Based Picture Retrieval Systems." A special issue of the J. of American Society for Information Science on "Knowledge Discovery and Data Mining," guest edited by Raghavan et al., was published in April 1998.
Currently, his research activities are supported by the DoE, the NSF, the U.S. Geological Survey (USGS) and the Louisiana Board of Regents. As a part of the research funded by the DoE, he collaborates with the National Wetlands Research Center of the USGS and the NASA/UL-Lafayette Regional Application Center (RAC) in the development of an Energy and Environmental Information Resources Center.
Raghavan co-chaired the NSF Workshop on Future Directions in Information Retrieval in 1991, which led to the sponsorship of the digital library initiative by the NSF, DARPA and NASA. He has served in numerous conference committees and is currently an ACM National Lecturer. Raghavan directs the DoD component of Louisiana's Experimental Program to Stimulate Competitiveness in Research (DEPSCoR) program. He is a member of the ACM, the IEEE, the Phi Kappa Phi and the Upsilon Pi Epsilon Computer Science honor societies.
Raghavan has served as a consultant to University of Malaya in the area of DBMS application design in the context of a DBMS based on the Network data model. He has collaborated with a research group at the NTT Laboratory in Japan on advanced text analysis and retrieval methods. He served as an expert consultant by STAR Software Corporation in order to evaluate the specifications and general design of the Laboratory Information Managemant System (LIMS) that is being developed for Armed Forces DNA Identification Laboratory (AFDIL). He was an advisor on a multi-national project on Biological Image Database Management coordinated by the Spanish National Center for Biotechnology.
Content-based retrieval requires the choice of distance functions for determining inter-image distances. Distance functions considered to be desirable for computing inter-image distances are often too expensive, computationally, to be used for on-line retrieval from large image databases. In this talk, we propose a generic and efficient content-based image retrieval architecture, where the original images are embedded onto an abstract feature space such that the desired (or real) inter-image distances are preserved in the new vector space. It is shown that it is more efficient to compute distances between these feature vectors and use them as estimates of the real distances. We have conducted experiments using color, texture and shape as image properties to define real distances. For this presentation, results based on color are used. The results show substantial reduction is possible in the size of the feature space and that high retrieval accuracy is achieved for the set of queries tested.
Most engines used for searching information resources via the Internet employ the Boolean Retrieval Model. Two main drawbacks of this model are that users have difficulty to precisely formulate their concept (or, topic) of interest using Boolean logic and the resulting output is not ranked. We propose to address both these problems by employing a Concept-based Retrieval Model, where a concept is defined by a set of production rules and the rule-base is represented as a rule tree. Features of a prototype called Concept-Set Structuring System (CS^3), which includes a graphical interface for defining and refining rule trees and for converting them into equivalent sets of conjunctions (called Minimal Term Sets), are described. By submitting MTSs generated for a concept to an existing search engine and by reordering the returned results according to the importance of MTSs they satisfy, our concept-based retrieval prototype enhances the capabilities of the underlying search engine. Results that demonstrate the use of the prototype, coupled with DOE InfoBridge, will be presented.
Business information received from advanced data analysis and data mining is a critical success factor for companies wishing to maximize competitive advantage. The use of traditional tools and techniques to discover knowledge is inefficient and does not lead to the delivery of the right information at the right time. In this paper, we introduce a dynamic approach that combines knowledge discovered in previous episodes with that of the current episode. The proposed approach is shown to be effective for solving problems related to the efficiency of handling database updates, accuracy of data mining results, gaining more knowledge and interpretation of the results, and performance. In our analysis, we have used an Apriori-like algorithm as the underlying procedure to generate all large itemsets. We show how the results of the Dynamic Data Mining algorithm are related to those obtained by Apriori-like algorithms. Our results are independent of the underlying algorithm used to generate itemsets.
Enhancements in data capturing technology have lead to exponential growth in amounts of data being stored in information systems. This growth in turn has motivated researchers to seek new techniques for extraction of knowledge implicit or hidden in the data. In this presentation, we motivate the need for an incremental data mining approach based on data structure called the item-set tree. The proposed approach is shown to be effective for solving problems related to efficiency of handling data updates, accuracy of data mining results, processing input transactions, and answering user queries. We present efficient algorithms to insert transactions into the item-set tree and to count frequencies of itemsets for queries about strength of association among items. We prove that the expected complexity of inserting a transaction is nearly O(1), and that of frequency counting is O(n), where n is the cardinality of the domain of items.