Connecting the Dots: Document-level Neural Relation Extraction with Edge-oriented Graphs

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4925–4936, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4925 Connecting the Dots: Document-level Neural Relation Extraction with Edge-oriented Graphs Fenia Christopoulou1, Makoto Miwa2,3, Sophia Ananiadou1 1National Centre for Text Mining, School of Computer Science, The University of Manchester, United Kingdom 2Toyota Technological Institute, Nagoya, 468-8511, Japan 3Artificial Intelligence Research Center (AIRC), National Institute of Advanced Industrial Science and Technology (AIST), Japan {efstathia.christopoulou, sophia.ananiadou}@manchester.ac.uk makoto-miwa@toyota-ti.ac.jp Abstract Document-level relation extraction is a com- plex human process that requires logical infer- ence to extract relationships between named entities in text. Existing approaches use graph- based neural models with words as nodes and edges as relations between them, to encode relations across sentences. These models are node-based, i.e., they form pair representa- tions based solely on the two target node rep- resentations. However, entity relations can be better expressed through unique edge repre- sentations formed as paths between nodes. We thus propose an edge-oriented graph neural model for document-level relation extraction. The model utilises different types of nodes and edges to create a document-level graph. An inference mechanism on the graph edges en- ables to learn intra- and inter-sentence rela- tions using multi-instance learning internally. Experiments on two document-level biomed- ical datasets for chemical-disease and gene- disease associations show the usefulness of the proposed edge-oriented approach.1 1 Introduction The extraction of relations between named enti- ties in text, known as Relation Extraction (RE), is an important task of Natural Language Processing (NLP). Lately, RE has attracted a lot of attention from the field, in an effort to improve the inference capability of current methods (Zeng et al., 2017; Christopoulou et al., 2018; Luan et al., 2019). In real-world scenarios, a large amount of rela- tions are expressed across sentences. The task of identifying these relations is named inter-sentence RE. Typically, inter-sentence relations occur in 1Source code available at https://github.com/ fenchri/edge-oriented-graph The case of a 40 - year - old patient who underwent an unsuccessful Bilateral optic neuropathy due to combined and treatment . ethambutol and isoniazid is reported . A with an unusual central was found . cadaver kidney transplantation and was treated with scotoma isoniazid ethambutol bilateral retrobulbar neuropathy bitemporal hemianopic Figure 1: Example of document-level, inter-sentence relations adapted from the CDR dataset (Li et al., 2016a). The solid and dotted lines represent intra- and inter-sentence relations, respectively. textual snippets with several sentences, such as documents. In these snippets, each entity is usu- ally repeated with the same phrases or aliases, the occurrences of which are often named entity men- tions and regarded as instances of the entity. The multiple mentions of the target entities in differ- ent sentences can be useful for the identification of inter-sentential relations, as these relations may depend on the interactions of their mentions with other entities in the same document. As shown in the example of Figure 1, the en- …

Content truncated to 500 words