A New Method for Cross-Language Information Retrieval by Summing Weights of Graphs

Abstract

Disambiguation is the aim of most translation techniques used in cross-language information retrieval. In this paper, we present a new method for query translation which only needs a bilingual dictionary and a monolingual corpus. Unlike the traditional statistical approach, our method uses co-occurrences between pairs of terms as statistical measure. By adding up all the weights of a k-complete subgraph, we can compare different combinations of target terms. The output of our method is in the form of probability distribution. Then the result is converted to the query in the target language. The method is easy to implement, and experiment shows it performs well.