Combining resources with confidence measures for cross language information retrieval

Query translation in Cross Language Information Retrieval (CLIR) can be performed using multiple resources. Previous attempts to combine different translation resources use simple methods such as linear combination. Unfortunately, these approaches are insufficient to combine different types of resources such as bilingual dictionaries and statistical translation models. In this paper, we use confidence measures for this combination for the purpose of English-Arabic CLIR. Confidence measure is used to adjust the original scores of translations and to create a weight of the same nature for translations with different resources. We tested this technique on two test CLIR collections from TREC and obtained encouraging improvements compared to the results of linear combination. Categories and Subject Descriptors H.3.3 [Information storage and retrieval]: Information Search and Retrieval – retrieval models, query formulation. General Terms Algorithms, Performance, Experimentation, Theory. K...

• Type: Conference
• Year: 2007
• Where: CIKM
• Authors: Youssef Kadri, Jian-Yun Nie
• Keywords: CIKM 2007, Confidence Measures, Cross Language Information Retrieval, Linear Combination