Date of Award

Winter 2020

Project Type


Program or Major

Computer Science

Degree Name

Master of Science

First Advisor

Laura Dietz

Second Advisor

Elizabeth Varki

Third Advisor

Marek Petrik


In this work, we propose a method to retrieve a human-readable explanation of how a retrieved entity is connected to the information need, analogous to search snippets for document retrieval. Such an explanation is called a support passage.

Our approach is based on the idea: a good support passage contains many entities relevantly related to the target entity (the entity for which a support passage is needed). We define a relevantly related entity as one which (1) occurs frequently in the vicinity of the target entity, and (2) is relevant to the query. We use the relevance of a passage (induced by the relevantly related entities) to find a good support passage for the target entity. Moreover, we want the target entity to be central to the discussion in the support passage. Hence, we explore the utility of entity salience for support passage retrieval and study the conditions under which it can help. We show that our proposed method can improve performance as compared to the current state-of-the-art for support passage retrieval on two datasets from TREC Complex Answer Retrieval.