Company Relation Extraction from Web News Articles for Analyzing Industry Structure

Ayana Yamamoto, Yuichi Miyamura, Kouta Nakata, Masayuki Okamoto

Abstract: It is indispensable to understand and analyze industry structure and company relations from documents, such as news articles, in order to make management decisions concerning supply chains, selection of business partners, etc. Analysis of company relations from news articles requires both a macro-viewpoint, e.g., overviewing competitor groups, and a micro-viewpoint, e.g., grasping the descriptions of the relationship between a specific pair of companies collaborating. Research has typically focused on only the macro-viewpoint, classifying each company pair into a specific relation type. In this paper, to support company relation analysis from both macro-and micro-viewpoints, we propose a method that extracts collaborative/competitive company pairs from individual sentences in Web news articles by applying a Markov logic network and gather extracted relations from each company pair. By this method, we are able not only to perform clustering of company pairs into competitor groups based on the dominant relations of each pair (macro-viewpoint) but also to know how each company pair is described in individual sentences (micro-viewpoint). We empirically confirmed that the proposed method is feasible through analysis of 4,661 Web news articles on the semiconductor and related industries.