الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 183 |  |  |
| | | from | 183 | | |
Abstract
The variety of knowledge existing in various language resources disseminated the demanding of using Multi-Lingual Information Retrieval (MLIR) rather than traditional information retrieval. This thesis introduces fine-grained MLIR system where it focuses on three main issues to enhance MLIR performance. Firstly, this thesis proposes a method for measuring the quantity of online content of a set of languages at domain level. This measurement is used for building a MLIR system that identifies which languages are strongly represented on the internet about a specific query topic. Secondly, this thesis proposes a query translation disambiguation approach based on measuring the semantic relatedness between the terms of the target query. It uses the semantic relatedness to select the suitable translation candidate of query terms and it doesn’t rely on parallel corpus. BabelNet lexical resource is used for extracting translating candidates of words and multi-word expressions. Thirdly, this thesis proposes LTR approach based on simple Term Weighting Scheme (TWS) which is Term Frequency-Average Term Occurrence (TF-ATO) in CLIR datasets. Various machine learning ranking methods were examined for building ranking models for CLIR system including: SVMLight, SVMRank.