TALP at GeoCLEF 2007: Results of a Geographical Knowledge Filtering Approach with Terrier

This paper describes and analyzes the results of our experiments in Geographical Information Retrieval (GIR) in the context of our participation in the CLEF 2007 GeoCLEF Monolingual English task. Our system uses Linguistic and Geographical Analysis to process topics and document collections. Geographical Document Retrieval is performed with Terrier and Geographical Knowledge Bases. Our experiments show that Geographical Knowledge Bases can be used to improve the retrieval results of the Terrier state-of-the-art IR system by filtering out non geographically relevant documents. 1 System Description Our GIR system is a modified version of the system presented in GeoCLEF 2006 [1] with some changes in the Retrieval modes and the Geographical Knowledge Base. The system has four phases performed sequentially: i) a Linguistic and Geographical Analysis of the topics, ii) a thematic Document Retrieval with Terrier (a state-of-the-art search engine that implements relevance feedback and several retrieval models such as: TFIDF, BM25, and Divergence From Randomness), iii) a Geographical Retrieval task with Geographical Knowledge Bases (GKBs), and iv) a Document Filtering phase. In addition, we have developed a toolbox based on Shape Files 1 for countries, following [2]. A Shape File is a popular geospatial vector data format for geographic information systems software. Shape Files spatially describe geometries: points, polylines, and polygons. In this paper we focus on the analysis of the experiments at GeoCLEF 2007. For a more detailed description of the system architecture, collection processing, and tuning, consult [3].