Using top-ranked documents in response to a query has been shown to be aneffective approach to improve the quality of query translation indictionary-based cross-language information retrieval. In this paper, wepropose a new method for dictionary-based query translation based on dimensionprojection of embedded vectors from the pseudo-relevant documents in the sourcelanguage to their equivalents in the target language. To this end, first welearn low-dimensional vectors of the words in the pseudo-relevant collectionsseparately and then aim to find a query-dependent transformation matrix betweenthe vectors of translation pairs appeared in the collections. At the next step,representation of each query term is projected to the target language and then,after using a softmax function, a query-dependent translation model is built.Finally, the model is used for query translation. Our experiments on four CLEFcollections in French, Spanish, German, and Italian demonstrate that theproposed method outperforms a word embedding baseline based on bilingualshuffling and a further number of competitive baselines. The proposed methodreaches up to 87% performance of machine translation (MT) in short queries andconsiderable improvements in verbose queries.
展开▼
