An Integrated Probabilistic and Logic Approach to Encyclopedia Relation Extraction with Multiple Features

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 1065–1072 Manchester, August 2008 An Integrated Probabilistic and Logic Approach to Encyclopedia Relation Extraction with Multiple Features ∗ Xiaofeng YU Wai LAM Information Systems Laboratory Department of Systems Engineering & Engineering Management The Chinese University of Hong Kong Shatin, N.T., Hong Kong {xfyu,wlam}@se.cuhk.edu.hk Abstract We propose a new integrated approach based on Markov logic networks (MLNs), an effective com- bination of probabilistic graphical models and first- order logic for statistical relational learning, to ex- tracting relations between entities in encyclopedic articles from Wikipedia. The MLNs model en- tity relations in a unified undirected graph col- lectively using multiple features, including contex- tual, morphological, syntactic, semantic as well as Wikipedia characteristic features which can cap- ture the essential characteristics of relation extrac- tion task. This model makes simultaneous statisti- cal judgments about the relations for a set of related entities. More importantly, implicit relations can also be identified easily. Our experimental results showed that, this integrated probabilistic and logic model significantly outperforms the current state- of-the-art probabilistic model, Conditional Random Fields (CRFs), for relation extraction from encyclo- pedic articles. 1 Introduction Relation extraction is a growing area of research that discovers various predefined semantic relations (e.g., visited, associate, and executive) between en- tity pairs in text. As a subtask in Information Ex- traction (IE), this problem has generated much in- terest and has been formulated as part of Message Understanding Conferences (MUC) and Automatic Content Extraction (ACE) Evaluation. Reliably extracting relations between entities in natural-language documents is still a difficult, un- solved problem. A large number of engineered systems were developed for identifying relations of interest. Recent approaches to this problem in- ∗The work described in this paper is substantially sup- ported by grants from the Research Grant Council of the Hong Kong Special Administrative Region, China (Project Nos: CUHK4193/04E and CUHK4128/07) and the Direct Grant of the Faculty of Engineering, CUHK (Project Codes: 2050363 and 2050391). This work is also affiliated with the Microsoft- CUHK Joint Laboratory for Human-centric Computing and Interface Technologies. ∗c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. clude statistical parsing (Miller et al., 2000), lo- gistic regression (Kambhatla, 2004), feature-based methods (Zhou et al., 2005; Toru et al., 2007), and kernel methods (Zelenko et al., 2003; Culotta and Sorensen, 2004; Bunescu and Mooney, 2005, 2006). In text, this usually amounts to examining pairs of entities in a document and determining whether a relation exists between them. In general, the above approaches to relation extraction suffer from the following three difficulties: (1) enumerating all pairs of entities, even when restricted to pairs within a sentence, results in a low density of pos- itive relation examples, (2) these approaches as- sume that relations only exist within document, and classify them independently without considering dependencies between entities. However, this as- sumption does not hold in practice, and ignoring dependencies between entities may lead to reduced performance, and (3) implicit …

Content truncated to 500 words