Numerous digitisation activities in manufacturing led to an enormous increase in available, accessible data. Knowledge graphs (KGs) become increasingly popular in this domain as they show strengths in integrating different data sources and serve as a basis for downstream tasks. Yet, constructing a KG is still a challenging and time-consuming process. Neuro-symbolic AI approaches, especially with powerful LLMs, have shown promising potential in research and industry and can support KG construction. Nevertheless, KG construction with neural methods must be aware of, or ideally even handle, the inexplicability of results when applying the KG to downstream manufacturing tasks, e.g., tasks of reliability- or safety-relevance. This makes it interesting to evaluate the utilisation of neuro-symbolic AI and LLMs in KG construction in manufacturing. To the best of our knowledge, there is no systematic literature review on neuro-symbolic AI and LLMs in KGs in manufacturing to date. Hence, this paper conducts a systematic literature review on neuro-symbolic AI and LLMs in KG construction in manufacturing. We show a solid increase of relevant publications on manufacturing KG construction and further show that BERT embeddings, RNN encodings, especially BiLSTM, CRF decodings, and, recently, LLMs, are common components of knowledge extraction from text documents to build KGs in manufacturing. With this systematic review, we support both further research and industry application in this field. The main question to guide this review is “Which role play neuro-symbolic AI, especially LLM approaches in knowledge graph construction for manufacturing?”.
Schmidt et al. (Thu,) studied this question.