| Popis: |
Online textual content constitutes a valuable source of information for market stakeholders, enabling them to unveil their business network's most important operations and interactions , and to gain insights about their customers, business partners, and competitors, in order to make well-informed strategic decisions. Due to the problem of information overload , manually extracting this information remains a laborious task for professionals, making the use of Information Extraction technologies a powerful asset. In this context, this paper concerns discovering business relations between companies (e.g. company-partner) from French content on the web. We present a new dataset for business relation extraction at the sentence level and develop a set of deep learning experiments to distinguish between business vs. non-business relations , as well as identify five types of business relations according to a predefined taxonomy. Our results are encouraging , showing that multitask architectures are the most productive beating several strong state of the art baselines. |
