The Second Half of Knowledge Graphs: Opportunities and Challenges

The report provides a deep analysis of the development of knowledge graphs since their emergence in 2012, describing how they have become a core technology for knowledge representation in the era of big data and cognitive intelligence.

It reviews the achievements of the first half of knowledge graph research, including large‑scale simple applications such as search, recommendation, and question answering, where massive entity and relation graphs have been successfully deployed by leading internet companies.

The second half presents new challenges: moving from massive, data‑rich scenarios to small‑scale, complex domains that require dense expert knowledge, limited data, and deep semantic understanding, especially in industries like energy, manufacturing, healthcare, and law.

Key challenges identified include the vocabulary gap between users and domain experts, missing causal chains in data, fragmented data integration, and the scarcity of high‑quality labeled data for specialized tasks.

Opportunities arise from advances in machine learning, such as low‑resource learning, unsupervised and weakly supervised methods, and the integration of symbolic knowledge with statistical models to reduce dependence on large labeled datasets.

The report proposes three strategic directions: (1) Knowledge representation – enhancing graph structures with temporal, multimodal, and personalized representations; (2) Knowledge acquisition – developing low‑cost, multi‑granular, large‑scale common‑sense and domain‑specific extraction methods, including implicit crowdsourcing; (3) Knowledge application – improving transparency of graph‑based services, enabling explainable AI, and guiding machine learning with symbolic knowledge.

It emphasizes the importance of multimodal knowledge graphs that link symbols to images, audio, and video to achieve cross‑modal understanding, and calls for research on personalized graph representations that reflect different user perspectives.

Finally, the report highlights that the future of knowledge graphs lies in bridging big data with big knowledge, advancing both the theoretical foundations and practical deployments to support complex, expert‑driven intelligent systems.