In this paper we describe our participation in the INEX 2010 ad-hoc track. We participated in three retrieval tasks (restricted focused task, relevant-in-context, restricted relevant-in-context) and report our findings based on a single set of measure for all tasks. In this year's participation, we evaluate the performance of the standard language model that is more focused on a fixed number of relevant characters than on relevant paragraphs. Our findings are: 1) the simplest language model for document retrieval performs relatively well in the restricted focused task when using a fixed offset that is close to the average character distance from the beginning of a document to its main content; 2) a good result of document ranking does improve the performance of snippet retrieval; 3) stemming and stopword removal can further boost performance.
展开▼
