Chinese Character Relationship Extraction Method Based on BERT

With the continuous development of information technology and the advent of the era of big data, today's society has entered the era of artificial intelligence. A variety of artificial intelligence application products continue to appear, playing various important roles in many fields, such as AI + agriculture, AI + medical, AI + autonomous driving, AI + education, etc. The combination of artificial intelligence and traditional industries The emergence of artificial intelligence has made these traditional industries shine. In the field of artificial intelligence, there is a very important sub field--Natural Language Processing (NLP). In the development of various industries, the scale of data generated by various industries is increasing, and the problem of information overload is becoming more and more serious, so how to quickly and accurately Obtaining key information in the data is of great significance. However, the text data generated by all walks of life is often unstructured and diverse in types. Therefore, in recent years, information extraction in the field of natural language has become a research hotspot, and person relationship extraction is an important branch of information extraction. Person relationship extraction is the core task of text mining and information extraction, and its task is to identify sentences. The semantic relationship between two person name entities, in recent years, person relationship extraction has been widely promoted and applied in various aspects, such as intelligent semantic search, character knowledge graph construction, question and answer systems, etc. It performs well. Therefore, it is of great significance to design and develop a set of Chinese character relationship extraction system. In response to this problem, this paper proposes a BERT-based Chinese character relationship extraction architecture. Its fusion model architecture is BERT+BI-LSTM+Multi-head-Self-Attention+FC. The current popular deep learning for Chinese character relationship extraction the model of has a better effect on the relationship extraction of a single character relationship pair. However, when it is extended to single-sentence multi-person relationship pairs and document-level semantic complexity, the evaluation index data of this model is not high. Through comparison, it is found that BERT has a significant improvement in the pre-training effect of the word2vec model, and the BI-LSTM+Multi-head-Self-Attention model has a significant improvement in the effect of a single LSTM model. Therefore, the proposed BERT+BI-LSTM +Multi-head-Self-Attention+ FC fusion model has certain practical value in Chinese character relationship extraction.