Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

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

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Computer Science

Year Degree Awarded

Spring 2014

First Advisor

James Allan

Subject Categories

Computer Sciences | Databases and Information Systems

Abstract

The goal of this work is to leverage cross-document entity relationships for improved understanding of queries and documents. We define an entity to be a thing or concept that exists in the world, such as a politician, a battle, a film, or a color. Entity-based enrichment (EBE) is a new expansion model for both queries and documents using features from similar entitymentions in the document collection and external knowledge resources. It uses task-specific features from entities beyond words that include: name aliases, fine-grained entity types, categories, and relationships to other entities. EBE addresses the problem of sparse or noisy local evidence due to multiple topics, implicit context, or informal writing. With the ultimate goal of improving information retrieval effectiveness, we start from unstructured text and through information extraction build up rich entity-based representations linked to external knowledge resources. We study the application ofentity-based enrichment to each step in the pipeline: 1) Named entity recognition, 2) Entity linking, and 3) Ad hoc document retrieval. The empirical results for EBE in each of these tasks shows significant improvements. Our first application of entity-based enrichment is the task of detecting entities in named entity recognition. We enrich the representation of observed words likely to represent entities. We perform weighted feature copying of recognition features from similar tokens in the corpus and external collections. The evaluation shows statistically significant improvements on in-domain newswire accuracy and demonstrates that the models are more robust on out-of-domain data. In the second part of this work, we apply EBE to the task of entity linking. The proposed entity linking method for disambiguating the detected mentions to entries in an external knowledge base is based on information retrieval. Theneighborhood relevance model, an enrichment model, identifies salient associations between an entity mention and otherentity mentions in the document. The results show that the enrichment model outperforms other context models and results in a system that is competitive with leading methods. Using the constructed entity representation, the final task is ad hoc document retrieval. We enrich the query representation with entity features. Retrieval is performed over documents annotated with entities linked to Wikipedia and Freebase from our system as well as the publicly available Google FACC1 annotations. To effectively leverage linked entity features, we extend dependency-based retrieval models to include structured attributes. We also define a new query-specific entity context model which builds a model for disambiguated entities from retrieved documents. Our results demonstrate significant improvements over competitive query expansion baselines for several standard test collections.

Share

COinS