Arabic search results disambiguation : a set of benchmarks

Abstract

Web search engines aim at retrieving relevant results in response to a user information need. The query expressing the user information need can be ambiguous by potentially referring to different meanings or senses. Search results clustering (SRC) attempts to disambiguate query results by grouping them into groups of sense-relevant clusters. Little research was done on Arabic SRC, and one important reason may be the lack of quality benchmarks for SRC testing and evaluation. The main contribution of this paper is to introduce a set of benchmarks for Arabic SRC, called AMBIGAr-abic to aid in performing SRC experiments. The benchmarks include manually labeled datasets and a dataset based on blind relevance feedback (BRF). The designed benchmarks were used in a series of SRC experiments we performed and the results were encouraging. The benchmarks are being made available for use by research-ers working on Arabic SRC.