Clustering for unsupervised relation identification

Abstract

Unsupervised Relation Identification is the task of automatically discovering interesting relations between entities in a large text corpora. Relations are identified by clustering the frequently cooccurring pairs of entities in such a way that pairs occurring in similar contexts end up belonging to the same clusters. In this paper we compare several clustering setups, some of them novel and others already tried. The setups include feature extraction and selection methods and clustering algorithms. In order to do the comparison, we develop a clustering evaluation metric, specifically adapted for the relation identification task. Our experiments demonstrate significant superiority of the singlelinkage hierarchical clustering with the novel threshold selection technique over the other tested clustering algorithms. Also, the experiments indicate that for successful relation identification it is important to use rich complex features of two kinds: features that test both relation slots together ("relation features"), and features that test only one slot each ("entity features"). We have found that using both kinds of features with the best of the algorithms produces very high-precision results, significantly improving over the previous work.

Discovery of Predicate-Oriented Relations among Named Entities Extracted from Thai Texts

“…In the past, several works [2], [3], [17], [18], [21] focused on how to identify relevant entities (e.g., person, location and organization) and their relations (e.g., work for, live in and kill) such as work for (person-organization), live in (personlocation) and kill (person-person). In contrast to these previous works, this work focuses on predicate-oriented relations among relevant entities (e.g., person, location and action) and their relationships (e.g., related and unrelated) are identified such as related (person-action), unrelated (personaction), related (action-person), unrelated (action-person), related (action-location) and unrelated (action-location).…”

“…Recently several information extraction (IE) approaches have been proposed to transform an unstructured text into knowledge base, such as those in [1]- [3]. The extracted knowledge is often used in question-answering systems for response to 4W1H (who, what, when, where and how) questions.…”

“…In addition, this method also provided a resolution of anaphora phenomenon between these entities using natural language processing (NLP) techniques. To discover relations among two named entities, a number of works [2], [3], [22] proposed methods to identify relations using context words between them. In [23], Agichtein and Cucerzan claimed that relation extraction from text documents was a harder task than named entity recognition.…”

“…Interaction, movement, transfer, creation and destruction events were five types of events identified and characterized in the Event Detection and Characterization (EDC) task. In previous works [2], [3], [17], [18], [21], the detected NE relations are mostly of specific actions, such as work for (person-organization), live in (person-location) and kill (person-person). NE extraction or NE relation extraction from texts in English and other languages with sufficient processing tools, syntactic information can be provided to analyze the sentence structure and extract NEs, and semantic information is usually applied to detect relations among such NEs.…”

Topic identification for fine-grained opinion analysis

“…Hasegawa et al (2004), Rosenfeld and Feldman (2007)). This work, however, considers named entities and heads of proper noun phrases rather than topic spans, and the relations learned are those commonly held between NPs (e.g.…”