Semantic modeling of the natural language of Wikipedia annotations

Knowledge bases (KB) store relational facts and constitute a significant resource for a variety of natural language processing (NLP) tasks. Improving their coverage and refining the relations is a basic and pressing research effort. In this thesis we propose a novel approach towards this canonical task by using the unstructured Wikipedia corpus: we extract low-dimensional embeddings for title pages of the Wikipedia corpus and show that they can be used to significantly outperform state-of-the-art approaches on a variety of metrics in three concrete tasks: measuring semantic relatedness, solving semantic analogies, and KB completion and refinement. A central feature of our work is a new log-linear discriminative model for the annotations inside a Wikipedia document that we name IBOE (isotropic bag-of-entities): we hypothesize that the parameters of the model satisfy a geometric symmetry property (isotropy). We show that the isotropy property leads to self-normalization allowing for the design of an efficient parameter estimation algorithm that we christen wiki2vec. The self-normalization property of IBOE is validated empirically on the Wikipedia corpus and is also of independent mathematical interest.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2016

Author dc:creator

• Mu, Jiaqi

Contributors dc:contributor

• Viswanath, Pramod
• Bhat, Suma P.

• entity embedding
• knowledge base completion

Statement dc:rights

• Copyright 2016 Jiaqi Mu

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/92851

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/92851