The knowledge graph is one of the essential infrastructures of artificial intelligence. It is a challenge for knowledge engineering to construct a high-quality domain knowledge graph for multi-source heterogeneous data. We propose a complete process framework for constructing a knowledge graph that combines structured data and unstructured data, which includes data processing, information extraction, knowledge fusion, data storage, and update strategies, aiming to improve the quality of the knowledge graph and extend its life cycle. Specifically, we take the construction process of an enterprise knowledge graph as an example and integrate enterprise register information, litigation-related information, and enterprise announcement information to enrich the enterprise knowledge graph. For the unstructured text, we improve existing model to extract triples and the F1-score of our model reached 72.77%. The number of nodes and edges in our constructed enterprise knowledge graph reaches 1,430,000 and 3,170,000, respectively. Furthermore, for each type of multi-source heterogeneous data, we apply corresponding methods and strategies for information extraction and data storage and carry out a detailed comparative analysis of graph databases. From the perspective of practical use, the informative enterprise knowledge graph and its timely update can serve many actual business needs. Our proposed enterprise knowledge graph has been deployed in HuaRong RongTong (Beijing) Technology Co., Ltd. and is used by the staff as a powerful tool for corporate due diligence. The key features are reported and analyzed in the case study. Overall, this paper provides an easy-to-follow solution and practice for domain knowledge graph construction, as well as demonstrating its application in corporate due diligence.

In the process of medical diagnosis and treatment, medical laboratory sheets are an important basis for doctors to diagnose diseases. However, paper medical laboratory sheets have many formats and complex contents, which brings many challenges to the existing text recognition systems, such as low recognition accuracy and serious adhesion of recognition results. To this end, this paper proposes a medical laboratory sheet layout analysis model that integrates regional features and table line features, innovatively introducing layout analysis into the field of medical laboratory sheet recognition. Considering that the sheets are in the form of a table with few lines, the model is dedicated to identifying the table lines and regional features, and realizing layout analysis. Experimental results show that in the medical laboratory sheet layout analysis, the mean AP on patient items and full items achieve 0.712 and 0.706, respectively. It proves that our model can significantly improve the accuracy of medical laboratory sheet layout recognition.

In knowledge-intensive fields such as medicine, the text often contains numerous professional terms, specific text fragments, and multidimensional information. However, most existing text representation methods ignore this specialized knowledge and instead adopt methods similar to those used in the general domain. In this paper, we focus on developing a learning module to enhance the representation ability of knowledge-intensive text by leveraging a graph-based cross-granularity message passing mechanism. To this end, we propose a novel learning framework, the M ulti- G ranularity G raph N eural N etwork (MG-GNN), to integrate fine-grained and coarse-grained knowledge at the character, word, and phase levels. The MG-GNN performs learning in two stages: 1) inter-granularity learning and 2) intra-granularity learning. During inter-granularity learning, semantic knowledge is extracted from character, word, and phrase granularity graphs, whereas intra-granularity learning focuses on fusing knowledge across different granularity graphs to achieve comprehensive message integration. To enhance the fusion performance, we propose a context-based gating mechanism to guide cross-graph propagation learning. Furthermore, we apply MG-GNN to address two important medical applications. Experimental results demonstrate that our proposed MG-GNN model significantly enhances the performance in both diagnosis prediction and medical named entity recognition tasks.