Coherence and Salience-Based Multi-Document Relationship Mining

Yongpan Sheng,Zenglin Xu

DOI: https://doi.org/10.1007/978-3-030-26072-9_30

2019-01-01

Abstract:In today’s interconnected world, there is an endless 24/7 stream of new articles appearing online. Faced with these overwhelming amounts of data, it is often helpful to consider only the key entities and concepts and their relationships. This is challenging, as relevant connections may be spread across a number of disparate articles and sources. In this paper, we propose a unified framework to aid users in quickly discerning salient connections and facts from a set of related documents, and presents the resulting information in a graph-based visualization. Specifically, given a set of relevant documents as input, we firstly extract candidate facts from above sources by exploiting Open Information Extraction (Open IE) approaches. Then, we design a Two-Stage Candidate Triple Filtering (TCTF) approach based on a self-training framework to maintain only coherent facts associated with the specified document topic from the candidates and connect them in the form of an initial graph. We further construct this graph by a heuristic to ensure the final conceptual graph only consist of facts likely to represent meaningful and salient relationships, which users may explore graphically. The experiments on two real-world datasets illustrate that our extraction approach achieves 2.4% higher on the average of F-score over several OpenIE baselines. We also further present an empirical evaluation of the quality of the final generated conceptual graph towards different topics on its coverage rate of topic entities and concepts, confidence score, and the compatibility of involved facts. Experimental results show the effectiveness of our proposed approach.