Relevant Term Suggestion Based on Pseudo Relevance Feedback from Web Contexts

Most search engines rely on query logs to give query suggestions. By mining potential relevant terms surrounding the query from Web resources, we aim at improving query formulation and retrieval effectiveness without query logs. In this paper, we propose a relevant term suggestion approach based on pseudo relevance feedback from Web contexts. Expansion term candidates are extracted and filtered by contextual relevance as calculated by mutual information and Web n-gram language model. Experimental results show a good performance in relevant term suggestion.