Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.

Non-WSU users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Access Type

WSU Access

Date of Award

January 2021

Degree Type


Degree Name



Computer Science

First Advisor

Alexander Kotov


The recent years have witnessed the increase in popularity of knowledge graphs in various applications, such as information extraction and retrieval systems, and intelligent assistants. Traditionally, retrieval from knowledge graphs is performed by submitting queries in SPARQL, a rigid query language based on triple patterns and logical operations. In this work, we propose several approaches to ad hoc and conversational entity retrieval that transcend the limitations of this approach by allowing the user to either submit queries using natural language in an ad hoc retrieval setting or have a conversation with an intelligent retrieval system by asking a series of questions that seek to find particular entities of interest from a knowledge graph. We propose several novel methods to effectively rank entities in such scenarios. For an ad hoc entity retrieval setting, we propose Parameterized Fielded Term Dependence Models, which infer the user's intent behind each individual query concept by dynamically estimating its projection onto the fields ofstructured entity representations based on a small number of statistical and linguistic features. Next, we propose Knowledge graph Entity and Word Embedding for Retrieval (KEWER), a random walk-based embedding model of both words and entities in the same embedding space, which allows us to calculate the similarity between query and entity embedding for relevance matching in a low-dimensional space. Finally, we propose a Neural Architecture for Sequential Simple Question Answering (NASS-QA), or conversational entity retrieval, which computes the matching score between a question and a given KG entity by taking into account the entity's neighboring structural components, such as entities, categories, and literals, as well as prior dialog history.

Off-campus Download