Goharian, N., E1-Ghazawi, T., Grossman, D., and Chowdhury, A. On the enhancements of a sparse matrix information retrieval approach. Proceedings of the International Conference on Parallel and distributed Processing Techniques and Applications, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....x 4 = 800 Mbytes of main memory. Although we exploit from the distributed memory, reading this matrix can increase disk access costs. Moreover, it also causes network congestion, since we work on cluster. In order to reduce the matrix size, we look at a compression method called Scalar ITPACK [5]. The idea is to store non zero elements of the matrix with their rows and column indices. To measure the computational efficiency of our parallel algorithm, we examined the speedup (S) The speedup is the ratio of the execution time for learning and classifying a document collection on a single ....

Goharian, N., E1-Ghazawi, T., Grossman, D., and Chowdhury, A. On the enhancements of a sparse matrix information retrieval approach. Proceedings of the International Conference on Parallel and distributed Processing Techniques and Applications, 1999.

....is the value of the last element of row vector incremented by the number of non zero elements in the last row of sparse matrix. The size of the row vector is M 1, M being the number of the rows in the sparse matrix. 1.2. Overview of the Information Retrieval Matrix Approach In our prior work [Goharian, et.al. 1999, 2000], we demonstrated the application of Sparse Matrix Vector Multiplication in an Information Retrieval (IR) System. The motivation of our work is to utilize other techniques and codes to implement a scalable IR system. Thus, minimizing the need for the redevelopment of software. Our approach relies ....

....4 shows the result of our experiments to store the collection data with the term offsets for proximity search, both as inverted index and sparse matrix. 3. Conclusion and Directions for Future Work Previously, we built an analytical model of the Sparse Matrix approach to Information Retrieval [Goharian et al. 1999, 2000]. In this paper, we built an experimental prototype to validate this analysis. The storage reduction of compressed sparse row matrix is 35 40 over the storage of conventional inverted index. Furthermore, we also experimentally evaluated our proposed Proximity Search structure to improve the ....

N. Goharian, T. El-Ghazawi, D. Grossman, A. Chowdhury, On the Enhancements of a Sparse Matrix Information Retrieval Approach, PDPTA'2000.

....engines. In this paper, we evaluate the result of our implementation of a parallel information retrieval engine as the application of sparse matrix vector multiplication. The application of sparse matrixvector multiplication in an IR System was described in the prior work of Goharian, et al. [2] and [3] They showed that using sparse matrix IR, with compressed sparse row format (CSR) a storage reduction of 35 40 could be achieved over the storage requirement of the conventional inverted index while the accuracy of the search would remain the same. The readers are referred to [4] for ....

....document scores to the query client. Query Processors: Each query processor node receives the query from the server. The index, i.e. vector elements resides in the memory of each node. Each node processes the query by using the CSR sparse matrix vector multiplication algorithm described in [2] and [3] The result of the multiplication computation is the relevance ranking score demonstrating the relevance of each document to the query. Each document has a score associated with the document. The query processor nodes sort their document scores and send the scores to the server. 3.2.1 ....

N. Goharian, T. El-Ghazawi, D. Grossman, A. Chowdhury, On the Enhancements of a Sparse Matrix Information Retrieval Approach, PDPTA'2000.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC