文摘
In this paper, we study query expansion strategies that improve the relevance of retrieved documents in a news and social media monitoring system, which performs real-time searches based on complex queries. We propose a two-step retrieval strategy using textual features such as bi-gram word dependencies, proximity, and expansion terms. We compare two different methods for query expansion: (1) based on word co-occurrence information; (2) using semantically-related expansion terms. We evaluate our methods and compare them with the baseline version of the system by crowdsourcing user-centric tasks. The results show that word co-occurrence outperforms semantic query expansion, and improves over the baseline in terms of relevance and utility.