A Multi-strategy Learning Approach to Competitor Identification

Abstract

Competitor identification tries to find competitors of some entity in a given field, which is the key to the success of market intelligence. Manually collecting competitors is labor-intensive and time consuming. So automatic approaches are proposed for this purpose. However, these approaches suffer from the following two main challenges. Competitor information might not only be contained in semi-structured sources like lists or tables, but also be mentioned in free texts. The diversity of its sources make competitor identification quite difficult. Also, these competitors might not always occur in form of their full names. The occurrences of name variants further increase the diversity, and make the task more challenging. In this paper, we propose a novel unsupervised approach to identify competitors from prospectuses based on a multi-strategy learning algorithm. More precisely, we first extract competitors from lists using some predefined heuristic rules. By leveraging redundancies among competitor information in lists, tables, and texts, these competitors are fed as seeds to distantly supervise the learning process to find table columns and text patterns containing competitors. The whole process is iteratively performed. In each iteration, the newly discovered competitors of high confidence from various sources are treated as new seeds for bootstrapping. The experimental results show the effectiveness of our approach without human intentions and external knowledge bases. Moreover, the approach significantly outperforms traditional named entity recognition approaches.