A Financial Knowledge Q&A Model for Power Enterprise Based on ChatGLM2-6B

doi:10.3969/j.issn.1006-2475.2024.11.009

Abstract

Abstract: With the continuous expansion of the scale of the power system， a significant amount of repetitive and complex tasks emerge in daily financial operations. Traditional methods of organizing and managing financial knowledge are no longer sufficient to satisfy the requirements of the current power system. With the consideration of this， the paper constructs a financial knowledge graph using the large-scale language model called ChatGLM2-6B. This method aims to standardize financial and project management processes and assist in financial decision-making. Firstly， the ChatGLM2-6B model should be optimized through instruction fine-tuning and prompt learning in order to extract event and event relationship pairs from financial contracts and invoice data， respectively. Then， the event relationship pairs are then stored as a local knowledge base using the FAISS vector database， additionally， a FAISS-ERNIE similarity evaluation model is trained to enhance the capability of knowledge retrieval， which could improve the question-answering ability of ChatGLM2-6B. Finally， hierarchical clustering algorithm is employed to generalize event relationship pairs， aiming to obtain contract knowledge graph and invoice knowledge evolutionary graph. These two graphs could be utilized to provide standardized guidance and supervision for real-time financial operations， achieving transparency in financial execution. The experimental results demonstrate that the method proposed in this paper exhibits excellent performance in event extraction， event relationship pair extraction， and similarity retrieval. The constructed contract and invoice knowledge evolutionary graphs hold significant implications for financial management in power enterprises， contributing to enhance the level of corporate management.

Key words: , ChatGLM2-6B, ERNIE, FAISS vector database, instruction fine-tuning, prompt learning

CLC Number:

TP3-05

YE Xue, YANG Sheng, CHENG Kai, ZHU Feng. A Financial Knowledge Q&A Model for Power Enterprise Based on ChatGLM2-6B[J]. Computer and Modernization, 2024, 0(11): 54-63.

References

［1］李明节，陶洪铸，许洪强，等. 电网调控领域人工智能技术框架与应用展望［J］. 电网技术， 2020，44（2）：393-400.
［2］高海翔，苗璐，刘嘉宁，等. 知识图谱及其在电力系统中的应用研究综述［J］. 广东电力， 2020，33（9）：66-76.
［3］ ZOU X H. A survey on application of knowledge graph［C］// Journal of Physics： Conference Series， 4th Interational Conference on Control Engineering and Artificial Intelligence. IOP， 2020. DOI: 10.1088/1742-6596/1487/1/012016.
［4］陶睿，吴继春，谢胜强，等. 深度学习和知识图谱在智能监管中的应用研究［J］. 金融纵横， 2019（8）：56-66.
［5］王倩玉. 领域知识图谱构建及其在财务审计的应用——以A高校为例［D］. 南京：南京审计大学， 2019.
［6］ CUI Y P， YANG Z P， YAO X. Efficient and effective text encoding for Chinese LLaMa and Alpaca［J］. arXiv preprint arXiv：2304.08177， 2023.
［7］ CHEN Z， WANG Y H， ZHAO B， et al. Knowledge graph completion： A review［J］. IEEE Access， 2020，8：192435-192456.
［8］ GUAN S P， CHE X Q， BAI L， et al. What is event knowledge graph： A survey［J］. IEEE Transactions on Knowledge and Data Engineering， 2023，35（7）：7569-7589.
［9］沈文斌，张家瑞. 事理图谱应用于电力安全事故预控策略研究［J］. 现代工业经济和信息化， 2022，12（12）：208-210.
［10］林穿，徐启峰，黄奕钒. 基于事理图谱的电力安全事故预控方法［J］. 中国安全生产科学技术， 2021，17（10）：39-45.
［11］温学兵，宋雨泽，王秋萍. 事理图谱研究进展［J］. 沈阳师范大学学报（自然科学版）， 2023，41（2）：153-157.
［12］ LIU X， ZHENG Y N， DU Z X， et al. GPT understands， too［J］. arXiv preprint arXiv：2103.10385， 2023.
［13］ KASNECI E， SESSLER K， KÜCHEMANN S， et al. ChatGPT for good? On opportunities and challenges of large language models for education［J］. Learning and Individual Differences， 2023，103. DOI: 10.1016/j.lindif.2023.102274.
［14］ CAO Y H， LI S Y， LIU Y X， et al. A comprehensive survey of AI-generated content （AIGC）： A history of generative AI from GAN to ChatGPT［J］. arXiv preprint arXiv：2303.04226， 2023.
［15］ DU Z X， QIAN Y J， LIU X， et al. GLM： General language model pretraining with autoregressive blank infilling［J］. arXiv preprint arXiv：2103.10360， 2021.
［16］FAN Z L C， TIAN J. Chinese-vicuna： A Chinese Instruction-following LLaMa-based Model［EB/OL］. （2023-06-12）［2023-09-05］. https：//github. com/Facico/Chinese-Vicuna.
［17］ XIONG H L， WANG S， ZHU Y T. et al. DoctorGLM： Fine-tuning your Chinese doctor is not a herculean task［J］. arXiv preprint arXiv：2304.01097v2， 2023.
［18］ HUANG Q Z， TAO M X， ZHANG C， et al. Lawyer LLaMA technical report［J］. arXiv preprint arXiv：2305.15062， 2023.
［19］贾骏，杨强，付慧，等. 基于电力设备大数据的预训练语言模型构建和文本语义分析［J］. 中国电机工程学报， 2023， 43（3）：1027-1036.
［20］孙洁，李能飞，赵梦茹. 媒体负面报道与公司财务困境预警——基于文本分析和机器学习［J］. 财经论丛， 2023，39（9）：80-90.
［21］ LIU X， JI K X， FU Y C， et al. P-tuning： Prompt tuning can be comparable to fine-tuning across scales and tasks［C］// Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics. ACL， 2022，2：61-68.
［22］ AGRAWAL M， HEGSELMANN S， LANG H， et al. Large language models are few-shot clinical information extractors［C］// Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. ACL， 2022：1998-2022.
［23］ WANG H F， LI J W， WU H， et al. Pre-trained language models and their applications ［J］. Engineering， 2023，25：51-65.
［24］ WEI J， BOSMA M， ZHAO V Y， et al. Finetuned language models are zero-shot learners［J］. arXiv preprint arXiv.2109.01652， 2021.
［25］ LI X Y， LI F Y， PAN L， et al. DuEE： A large-scale dataset for Chinese event extraction in real-world scenarios［C］// Proceedings of the 9th CCF International Conference on Natural Language Processing and Chinese Computing（NLPCC 2020）. Springer， 2020：534-545.
［26］ DENG H L， ZHANG Y N， ZHANG Y F. et al. Title2Event： Benchmarking open event extraction with a large-scale Chinese title dataset［J］. arXiv preprint arXiv：2211.00869， 2022.
［27］ LIU X， JI K X， FU Y C， et al. P-Tuning v2： Prompt tuning can be comparable to fine-tuning universally across scales and tasks［J］. arXiv preprint arXiv：2110.07602， 2021.
［28］ ZHOU D， SCHÄRLI N， HOU L， et al. Least-to-most prompting enables complex reasoning in large language models［J］. arXiv preprint arXiv：2205.10625， 2022.
［29］ SUN Y， WANG S H， LI Y K， et al. ERNIE： Enhanced representation through knowledge integration［J］. arXiv preprint arXiv：1904.09223， 2019.
［30］ DEVLIN J， CHANG M W， LEE K， et al. BERT： Pre-training of deep bidirectional transformers for language understanding［J］. arXiv preprint arXiv：1810.04805， 2018.
［31］ CUI J X， NING M N， LI Z J， et al. ChatLaw： Open-source legal large language model with integrated external knowledge bases［J］. arXiv preprint arXiv：2306.16092， 2023.
［32］ BISWA S. Role of ChatGPT in education［J］. SSRN， 2023： 4369981.
［33］ HOSSIN M， SULAIMAN M N. A review on evaluation metrics for data classification evaluations［J］. International Journal of Data Mining & Knowledge Management Process， 2015，5（2）. DOI: 10.5121/ijdkp.2015.5201.
［34］ NI J M， ÁBREGO G H， CONSTANT N， et al. Sentence-T5： Scalable sentence encoders from pre-trained text-to-text models［J］. arXiv preprint arXiv：2108.08877， 2021.
［35］ LAGLER K， SCHINDELEGGER M， BÖHM J， et al. GPT2： Empirical slant delay model for radio space geodetic techniques［J］. Geophysical Research Letters， 2013，40（6）：1069-1073.
［36］ TOUVRON H， MARTIN L， STONE K， et al. Llama 2： Open foundation and fine-tuned chat models［J］. arXiv preprint arXiv：2307.09288， 2023.
［37］ HUANG A H， WANG H， YANG Y. FinBERT： A large language model for extracting information from financial text［J］. Contemporary Accounting Research， 2023，40（2）：806-841.
［38］ LAN Z Z， CHEN M D， GOODMAN S， et al. ALBERT： A lite BERT for self-supervised learning of language representations［J］. arXiv preprint arXiv：1909.11942， 2019.
［39］ LIN C Y. Looking for a Few Good Metrics： ROUGE and Its Evaluation［Z］. Working Notes of NTCIR-4， 2004.

[1]	ZHENG Jiuchao, ZHAO Xinyuan. Entity Linking Method Based on Topics and Description Information [J]. Computer and Modernization, 2024, 0(12): 10-14.
[2]	ZHAO Chenyang, XUE Tao, LIU Junhua. Fashion Clothing Pattern Generation Based on Improved Stable Diffusion [J]. Computer and Modernization, 2024, 0(12): 15-23.
[3]	WANG Xiaohang1, LI Yongjie1, YU Lei1, FAN Xiao2. A Method of Using Compound Event Probability Operation to Solve Problem of Negative Information Blocking Maximization [J]. Computer and Modernization, 2024, 0(12): 24-33.
[4]	ZHANG Xiaodong1, BAI Guangzhi1, LI Min1, LI Haoyang2. Oil and Gas Well Production Prediction Model Based on Empirical Wavelet Transform [J]. Computer and Modernization, 2024, 0(12): 53-58.
[5]	LIU Yunhai1, Feng Guang1, WU Xiaoting2, YANG Qun2 . Safety Helmet Wearing Detection Algorithm for Complex Construction Scenes [J]. Computer and Modernization, 2024, 0(12): 66-71.
[6]	LIU Baobao, YANG Jingjing, TAO Lu, WANG Heying . DSMSC Based on Attention Mechanism for Remote Sensing Image Scene Classification [J]. Computer and Modernization, 2024, 0(12): 72-77.
[7]	GU Yue, DENG Songfeng, SHEN Ji, MU Wentao, ZHAO Enqi. SAR Ship Detection Algorithm Based on Improved YOLOv8 [J]. Computer and Modernization, 2024, 0(12): 78-83.
[8]	WU Xiuling1, ZHOU Sheng1, WANG Chunjuan1, YU Cuizhuo2, LIU Hao3. Research Progress in Ultra Short-term Power Load Forecasting Technology [J]. Computer and Modernization, 2024, 0(12): 108-115.
[9]	LI Deyou1, 2, YU Jinsongdi1, 2, WEI Dandan1, 2, LUO Yuan1, 2, TONG Ruiju3. Abstract Tree Model for Gridded Cube Metadata [J]. Computer and Modernization, 2024, 0(11): 1-6.
[10]	GONG Yicheng1, 2, LIU Qing1, 2. Beijing Opera Binary Classification Based on RF-LCE-BiLSTM-Attention-AMSSA Model [J]. Computer and Modernization, 2024, 0(11): 7-12.
[11]	LI Taoying, LI Meng, WU Mengqiao. Taxi Passenger Flow Prediction Based on Heterogeneous Spatiotemporal Graph#br# Convolutional Networks [J]. Computer and Modernization, 2024, 0(11): 13-18.
[12]	ZHANG Tai1, YAN Zihao2, DUAN Jie2, ZHANG Zhihong2. Information Forwarding Strategy of Internet of Vehicles in Named Data Network [J]. Computer and Modernization, 2024, 0(11): 19-27.
[13]	YUAN Qingle, MU Li. Inventory Forecasting Method Based on Improved Elman Neural Network [J]. Computer and Modernization, 2024, 0(11): 28-33.
[14]	ZHANG Kun1, ZHANG Yongwei1, WU Yongcheng1, ZHANG Xiaowen2, ZHAI Shichen2. An LLM-based Method for Automatic Construction of Equipment Failure Knowledge Graphs [J]. Computer and Modernization, 2024, 0(11): 46-53.
[15]	WAN Hongwei, CHEN Pinghua. Polyp Segmentation Based on Involution and Coordinate Reverse Attention [J]. Computer and Modernization, 2024, 0(11): 84-90.