Learning Knowledge-Enriched Company Embeddings for Investment Management

Relationships between companies serve as key channels through which the effects of past stock price movements and news events propagate and influence future price movements. Such relationships can be implicitly found in knowledge bases or explicitly represented as knowledge graphs. In this paper, we propose Knowledge-Enriched Company Embedding (KECE), a novel multi-stage attention-based dynamic network embedding model combining multimodal information of companies with knowledge from Wikipedia and knowledge graph relationships from Wikidata to generate company entity embeddings that can be applied to a variety of downstream investment management tasks. Experiments on an extensive set of real-world stock prices and news datasets show that the proposed KECE model outperforms other state-of-the-art models on key investment management tasks.