Implicit Business Competitor Inference Using Heterogeneous Knowledge Graph

Abstract

Competitor inference is the task of identifying current or potential competitors given their primary markets and Business Scope. Previous methods have achieved remarkable success on explicit competitor inference using state-of-the-art natural language processing (NLP) techniques, mainly relying on comparative expressions. However, those methods lack interpretability and cannot identify implicit competitors without the explicit mentions of competitive relationships in the text. To remedy these problems, in this paper, we propose a probabilistic graphical model which leverages heterogeneous enterprise knowledge graph containing both structured information, e.g., Product Analysis, Sales Territory, and unstructured information, e.g., Business Scope. The model is defined with first-order logic rules using the declarative language of Probabilistic Soft Logic (PSL). As a result, our model enables predicting implicit competitors while provides pieces of interpretable evidence. Experimental results show that our approach is significantly superior to previous methods.