Aplicação de técnicas de aprendizagem de máquina na geração de índices para sistemas de busca

Abstract

Search engines station between main Ways to Get INFORMATION on the Internet, receiving Million daily queries. This volume Query topping generates a considerable burden In Query processors of search machines, that should NOT JUST worry with a quality fi nal ANSWER received by users, but Also with a latency This consultation, JA What A delay excessive on MAY rhythm harm a RESPONSE Experience Users ITS. In the Last Year HAS A considerable effort been Research in Technical Application Machine Learning During Query Processing, aiming mainly um increase in the fi nal quality of YOUR answers. This thesis studied the Technical Application machine learning During the generation of indexes, rather than apply them to processing Queries, therefore opening a new front for the Technical Application of Learning Machine in search systems. Within the study, we propose two techniques for the Application Machine learning in the Document Indexing in search of machines with Showing THAT That there Space paragraph enhance an indexing with the use of these techniques. The advantage of this approach is that as THAT Processing And Done Before Consultations Made to be looking machine, Regardless of how costly this computationally BE Process, THIS IS NOT re fl directly etirá In Query Processing Time. We propose here Two techniques: LePrEF, One Evidence Fusion Technique Indexing Time Who Has As a search Purpose Machines Performance Improvement During Processing Consultations, THROUGH indexes Generation que CODI fi cam a Better importance of each term Each Document, AND A phrasal TERMS detection technique (phrases OS), with the objective of improve the quality of the responses obtained FOR THEM. The Technical LePrEF Performs a Relevance of Evidence Sources Fusion pace in Indexing using for both machine learning. The need for evidence of fusion not Drift Query Processing What do FACT search Systems in General Several use Evidence Sources To compute answers YOUR Such As the text of web pages, the text of IN pointers received each page, pointing Analysis Methods As the PageRank, Among many others. However, the recent increase Evidence Sources lev Also one New custodian processing. This is, coupled with the constant growth in the Content Quantity