A Dataset for Inter-Sentence Relation Extraction using Distant Supervision

A Dataset for Inter-Sentence Relation Extraction using Distant Supervision Angrosh Mandya, Danushka Bollegala, Frans Coenen, Katie Atkinson Department of Computer Science, University of Liverpool, UK {angrosh,danushka,coenen,katie}@liverpool.ac.uk Abstract This paper presents a benchmark dataset for the task of inter-sentence relation extraction. The paper explains the distant supervision method followed for creating the dataset for inter-sentence relation extraction, involving relations previously used for standard intra- sentence relation extraction task. The study evaluates baseline models such as bag-of-words and sequence based recurrent neural network models on the developed dataset and shows that recurrent neural network models are more useful for the task of intra-sentence relation extraction. Comparing the results of the present work on iner-sentence relation extraction with previous work on intra-sentence relation extraction, the study suggests the need for more sophisticated models to handle long-range information between entities across sentences. Keywords: Inter-sentence Relation Extraction, Relation Extraction, Inter-sentence Relation Extraction Dataset, Distant Supervi- sion for Inter-sentence Relation Extraction 1. Introduction In recent times, the field of relation extraction has received significant research attention due to its importance in in- formation retrieval (Culotta and Sorensen, 2004; Mintz et al., 2009; Banko et al., 2007; Etzioni et al., 2011). The key task in relation extraction is to recognise the semantic relation that exist between two given entities. Depending on the scope of the co-occurrences of the two entities, re- lation extraction methods can be broadly categorised into two groups: (a) intra-sentence relation extraction meth- ods (Brin, 1998; Banko et al., 2007; Mintz et al., 2009; Riedel et al., 2010), and (b) inter-sentence relation extrac- tion methods (Swampillai and Stevenson, 2010; Gu et al., 2017; Quirk and Poon, 2016; Peng et al., 2017). While intra-sentence relation extraction attempts to extract rela- tions between two entities that co-occur within the same sentence, inter-sentence relation extraction methods con- sider entities that might not necessarily co-occur in the same sentence. In more detail, the distinction between intra and inter- sentence relation extraction tasks can be illustrated as fol- lows. Let us assume that a relation r takes e1 as the first argument and e2 as the second argument. Further, let us also assume that e1 is included in a sentence si and e2 is included in a sentence sj. Then, we define intra-sentence relation extraction as the task of extracting relation r from si and sj, when i = j. We define inter-sentence relation extraction as the task of extracting relation r from si and sj, when i ̸= j. In this work, we limit i and j such that |i − j| = 1. Examples of intra-sentence and inter-sentence relation extraction are provided below in Listings 1 and 2, respectively. LISTING 1: EXAMPLE OF INTRA-SENTENCE RELATION EX- TRACTION S1: In 1957, <e1>Ayn Rand</e1> published her best- known work, the novel <e2>Atlas Shrugged</e2>. LISTING 2: EXAMPLE OF INTER-SENTENCE RELATION EX- TRACTION S1: <e1>Ayn Rand</e1> (born <e1>Alisa Zi- novyevna Rosenbaum</e1>, March 6, 1982) was a Russian-born American novelist, philosopher, playwright, and screenwriter. S2: She is known for her two best-selling …

Content truncated to 500 words