Penentuan Kerelevanan Dokumen Menggunakan Rangkaian Rambatan Balik

Information retrieval (IR) is one of the Computer Science branches that deals with accessing relevant information from a database. Several search engines have been developed to assist users in retrieving the relevant information from the Internet. However, due to information overload, some search e...

Full description

Saved in:
Bibliographic Details
Main Author: Fadhilah, Mat Yamin
Format: Thesis
Language:eng
eng
Published: 2002
Subjects:
Online Access:https://etd.uum.edu.my/491/1/FADHILAH_BT._MAT_YAMIN.pdf
https://etd.uum.edu.my/491/2/FADHILAH_BT._MAT_YAMIN.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Information retrieval (IR) is one of the Computer Science branches that deals with accessing relevant information from a database. Several search engines have been developed to assist users in retrieving the relevant information from the Internet. However, due to information overload, some search engines are still incapable of returning only the most relevant documents to the users. Hence, this research aims to explore the use of Artificial Intelligence (AI) technique, particularly neural network (NN) in measuring the relevancy of each document compared to the users requests. Backpropagation learning algorithm has been used as a basis for learning in this study. Several phases are involved, namely as the identification of the document's atributes, implementation of NN, identification of NN parameters and development of simple search engine prototype. 53 documents have been uploaded into the database for evaluation purpose. These documents have been downloaded from the Seventh International World Wide Web Conferences. The documents are then used to test with two different queries; 'metadata' and 'multimedia'. A test for 'metadata' query achieved 100 percent recall and 50 percent precision. Whereas, the test for 'muItimedia ' query achieved 75 percent recall and 60 percent precision. The result shows that the usage of NN approaches has produced a high recall. The result is also tested using fallout and generality measurement. Fallout for both queries are 6 and 5.666 percent respectively. Whereas, the generality for both queries are 4.08 and 7.54 respectively.