A dynamic knowledge graph and knowledge service model are constructed to address the risk management needs of enterprises. By extracting, integrating, and processing knowledge entities, a complete knowledge graph is established, and then a knowledge service model is designed to achieve intelligent and dynamic risk management. The experiment shows that in terms of entity extraction, the F1 value of traditional TextRank is only 73.8%. Through the three-level enhancement of weighted information entropy, sentence length features, and adjustable fusion weights, the research improved F1 to 81.8% (A-class dataset, 80,000 entries)/82.7% (B-class dataset, 42,000 entries), with a net increase of about 8%, surpassing mainstream algorithms such as TF-IDF, LSTM-CRF, GCN-CRF, etc. In terms of relationship fusion, the proposed BERT-BiLSTM model converges in 16 epochs with an accuracy of 87.8% (Class A dataset), and achieves the best balance between accuracy and efficiency with 112 M parameters. In addition, in 1000 concurrent transaction tests, the average response time of the research system is 678 s, 23% faster than similar systems, with a peak throughput of 1918 TPS. Meanwhile, the CPU usage is only 54.7%, the memory is 3 GB, and the P99 tail latency is 2.95 s, all of which are better than existing benchmarks. The above results indicate that the dynamic knowledge graph and enterprise risk management knowledge service model perform well in multiple core indicators, and can effectively improve the intelligence and dynamic level of enterprise risk management.
Jia-yin et al. (Thu,) studied this question.