Learning to Weight Translations using Ordinal Linear Regression and Query-generated Training Data for Ad-hoc Retrieval with Long Queries

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1725–1733, Osaka, Japan, December 11-17 2016. Learning to Weight Translations using Ordinal Linear Regression and Query-generated Training Data for Ad-hoc Retrieval with Long Queries Javid Dadashkarimi Masoud Jalili Sabet School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran {dadashkarimi,jalili.masoud,shakery}@ut.ac.ir Azadeh Shakery Abstract Ordinal regression which is known with learning to rank has long been used in information re- trieval (IR). Learning to rank algorithms, have been tailored in document ranking, information filtering, and building large aligned corpora successfully. In this paper, we propose to use this algorithm for query modeling in cross-language environments. To this end, first we build a query-generated training data using pseudo-relevant documents to the query and all translation candidates. The pseudo-relevant documents are obtained by top-ranked documents in response to a translation of the original query. The class of each candidate in the training data is deter- mined based on presence/absence of the candidate in the pseudo-relevant documents. We learn an ordinal regression model to score the candidates based on their relevance to the context of the query, and after that, we construct a query-dependent translation model using a softmax function. Finally, we re-weight the query based on the obtained model. Experimental results on French, German, Spanish, and Italian CLEF collections demonstrate that the proposed method achieves better results compared to state-of-the-art cross-language information retrieval methods, particu- larly in long queries with large training data. 1 Introduction The multilingual environment of the Web has long required the researchers in information retrieval (IR) to introduce powerful algorithms for bridging the gaps between the languages (Nie, 2010; Ganguly et al., 2012; Dadashkarimi et al., 2016). Generally, these algorithms can be categorized as follows: (1) translating the query of the user to the language of the documents (Ganguly et al., 2012), (2) translat- ing all of the documents into the language of the user (Oard, 1998), (3) translating the query and the documents into a third language (Kishida and Kando, 2005), (4) bringing the query and the documents into a shared low-dimensional space (Vulic and Moens, 2015; Dadashkarimi et al., 2016), and (5) using semantic/concept networks (Franco-Salvador et al., 2014). Usually the query translation approach has been opted as the most efficient and effective approach in the literature (Vulic and Moens, 2015; Nie, 2010). Ma et al. (2012), have shown that cross-language information retrieval (CLIR) takes more advan- tage of weighting all translations than selecting the most probable ones. But, building this translation model demands a statistical analysis of translation candidates over an aligned corpus or a single target collection (Talvensaari et al., 2007; Liu et al., 2005; Ganguly et al., 2012). Aligned corpora have been exploited in CLIR successfully (Rahimi et al., 2016; Talvensaari et al., 2007). But, these resources are either scarce in some languages or specific to a few number of domains. Therefore, recently query-dependent collections have been shown to be more effective and are available to …

Content truncated to 500 words