Focused Query Expansion With Entity Cores For Patient-Centric Health Search

The Web provides a plethora of contents about diseases, symptoms and treatments. Most notably, users turn to health forums to seek advice from doctors and from peers with similar cases. However, the benefit of forums mostly lies in community QA and browsing. Expressive querying for patient-centric needs is poorly supported by search engines. This paper overcomes this issue by enriching user queries with judiciously chosen entities and classes from a large knowledge graph. Candidate entities are extracted from the full text of user posts. To counter topical drift that would arise from picking all entities, we devise ECO, a novel method that computes a focused entity core for query expansion. Experiments with contents from health forums and clinical trials demonstrate substantial gains that ECO achieves over state-of-the-art baselines.