Research On Knowledge Graph Embedding And Construction Technology For Manufactory Industry

With the development of Industrial Internet+,the digital transformation of manufacturing is in full swing.Knowledge plays a vital role in industry,and it is the cornerstone of many intelligent decision-making and resource optimization.But now the manufacturing knowledge is distributed independently,and many decision rely on human experience.The knowledge graph has made great strides in recent years.It can integrate data from different sources and has good reasoning ability.Therefore,establishing a knowledge graph of manufacturing will promote the intelligentization of manufacturing.Most of the existing knowledge graphs only have qualitative knowledge,but there are many quantitative knowledge and evolutionary knowledge in the manufacturing industry.Therefore,this paper proposes a knowledge graph construction and representation model for manufacturing industry based on the knowledge characteristics of manufacturing industry.We have improved the manufacturing knowledge graph from three aspects.The main work of this paper is as follows:1)Combine quantitative knowledge with qualitative knowledgeThe extraction of quantitative knowledge is transformed into named entity recognition and attribute extraction.Attribute extraction usually translates into relationship classification problems.It is necessary to define categories and tag a large amount of training data in advance.However,there are many kinds of attributes in manufacturing,which are difficult to define in advance.This chapter proposes a combination of personalized Page Rank and Bi-LSTM-CRF for attribute extraction.It is not necessary to specify attribute categories in advance.This algorithm combines with unsupervised methods,which makes the method is portable and reduces the cost of manual labeling.2)Combine conceptual knowledge with affair knowledgeThe extraction of evolutionary knowledge is transformed into event trigger identification,event description completion,and event relationship classification.Event relationship classifications typically use traditional machine learning and only classify events in the same sentence.There are many events in the manufacturing industry that are in different sentences,but there exists relationships.This chapter uses the Bi-LSTM method to classify events,not only for events in the same sentence,but also for events across sentences.3)Combining quantitative and evolutionary knowledge for joint reasoningThe knowledge graph completion problem is transformed into a sorting problem,and we improved ProjE algorithm in this chapter.The quantitative knowledge is transformed into a vector and it was combined with event embedding.At the result,we obtain the final knowledge graph embedding about entity and relationship.The knowledge embedding not only addresses the relationship between entities and entities,but also enhances the ability of entity links.In summary,this paper establishes a knowledge graph for manufacturing,which is different from the existing knowledge graph.It combines quantitative knowledge and evolutionary knowledge.And in the case of knowledge inference,its knowledge graph embedding combines the attribute knowledge of quantitative knowledge,not only for the relationship between entities or evolutionary.The experimental results show that our method has improved compared to other methods.