Quantifying Logical Relations of Financial Risks with BERT and Mutual Information
Jia Minghua1,2(),Wang Xiuli1,3
1School of Information, Central University of Finance and Economics, Beijing 102206, China 2Peking University Library, Beijing 100871, China 3Engineering Research Center of State Financial Security, Ministry of Education, Beijing 102206, China
[Objective] This paper tries to prevent and control financial risks by quantifying their logical relationship, which also improve the reliability of processing word frequency of financial events. [Methods] We proposed a quantitative analysis method for the logical relation of financial risks based on BERT and mutual information combined with domain knowledge. Then, we quantified the relations with COPA and financial data sets. [Results] The proposed model effectively addressed the issue of unreliable quantization of word frequency. Its accuracy reached 80.1%, which was 3.1%~37.4% higher than the benchmark models. [Limitations] More research is needed to examine our new model with non-financial and other corpora. [Conclusions] Our new method can reveal the evolutionary path of financial risk events and improve the effect quantitative presentation of their logical relationship.
贾明华, 王秀利. 基于BERT和互信息的金融风险逻辑关系量化方法[J]. 数据分析与知识发现, 2022, 6(10): 68-78.
Jia Minghua, Wang Xiuli. Quantifying Logical Relations of Financial Risks with BERT and Mutual Information. Data Analysis and Knowledge Discovery, 2022, 6(10): 68-78.
The man broke his toe. What was the CAUSE of this?
Alternative 1：
He got a hole in his sock.
Alternative 2：
He dropped a hammer on his foot.
Premise：
I tipped the bottle. What happened as a RESULT?
Alternative 1：
The liquid in the bottle froze.
Alternative 2：
The liquid in the bottle poured out.
Premise：
I knocked on my neighbor's door. What happened as a RESULT?
Alternative 1：
My neighbor invited me in.
Alternative 2：
My neighbor left his house.
Table 2 COPA数据示例
编号
抽象主题事件
泛化事件数
结果事件数
原因事件数
E1
货币超发
3
10
10
E2
股市大跌
3
10
10
E3
美联储加息
3
10
10
E4
人民币升值
3
10
10
E5
人民币贬值
3
10
10
E6
中美贸易摩擦
3
10
3
E7
英国脱欧
3
10
1
E8
股市上涨
3
10
10
Table 3 抽象主题事件
编号
一因多果
由果溯因
方法A
方法B
方法A
方法B
E1
0.73
1.00
0.59
1.00
E2
0.73
1.00
0.75
1.00
E3
0.95
1.00
0.92
1.00
E4
0.79
1.00
0.85
1.00
E5
0.88
1.00
0.79
1.00
E6
0.86
1.00
0.37
1.00
E7
0.59
1.00
0.04
1.00
E8
0.58
1.00
0.69
1.00
Table 4 关系量化值结果对比
Fig.5 一因多果关系量化值分布
Fig. 6 由果溯因关系量化值分布
模型
参数量
模型 层数
隐层 大小
批大小
词表 大小
BERT-base-uncased
108M
12
768
16
30 522
BERT-large-uncased
334M
24
1 024
4
30 522
RoBERTa-base
123M
12
768
16
30 522
RoBERTa-large
355M
24
1 024
4
50 265
ALBERT-base
12M
12
768
32
30 000
ALBERT-large
18M
24
1 024
12
30 522
Table 5 BERT模型参数设置
实验方法
Test Set
Dev Set
Dev + Test
协方差*
50.2%
49.0%
49.6%
共现频率
50.0%
51.8%
50.9%
互信息
57.8%
58.8%
58.3%
BERT-base-uncased+PMI
58.2%
62.0%
60.1%
BERT-large-uncased+PMI
71.6%
68.6%
70.1%
RoBERTa-base+PMI
71.4%
76.8%
74.1%
RoBERTa-large+PMI
68.8%
70.6%
69.7%
ALBERT-base+PMI
57.6%
58.4%
58.0%
ALBERT-large+PMI
78.8%
81.4%
80.1%
Table 6 在COPA数据集上关系量化推理任务的准确率
[1]
Singhal A. Introducing the Knowledge Graph: Things, Not Strings[EB/OL]. (2012-05-16). [2020-03-01]. https://www.blog.google/products/search/introducing-knowledge-graph-things-not/.
(Liu Zongtian, Huang Meili, Zhou Wen, et al. Research on Event-oriented Ontology Model[J]. Computer Science, 2009, 36(11): 189-192.)
[3]
Lee S. Simulation Modeling with Event Graphs[J]. Communications of the ACM, 1983, 26(11): 957-963.
doi: 10.1145/182.358460
[4]
Buss A H. Modeling with Event Graphs[C]// Proceedings of the 28th Conference on Winter Simulation. 1996: 153-160.
[5]
Yang C C, Shi X D. Discovering Event Evolution Graphs from Newswires[C]// Proceedings of the 15th International Conference on World Wide Web. 2006: 945-946.
[6]
Yang C C, Shi X D, Wei C P. Discovering Event Evolution Graphs from News Corpora[J]. IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, 2009, 39(4): 850-863.
doi: 10.1109/TSMCA.2009.2015885
[7]
Li Z Y, Zhao S D, Ding X, et al. EEG: Knowledge Base for Event Evolutionary Principles and Patterns[C]// Proceedings of the 6th National Conference on Social Media Processing. 2017: 40-52.
[8]
Li Z Y, Ding X, Liu T. Constructing Narrative Event Evolutionary Graph for Script Event Prediction[C]// Proceedings of the 27th International Joint Conference on Artificial Intelligence. 2018: 4201-4207.
[9]
Ding X, Li Z Y, Liu T, et al. ELG: An Event Logic Graph[OL].arXiv Preprint, arXiv: 1907.08015.
(Hu Yang, Yan Hongfei, Chen Chong. Joint Entity and Relation Extraction for Constructing Financial Knowledge Graph[J]. Journal of Chongqing University of Technology (Natural Science), 2020, 34(5): 139-149.)
(Li Jianglong, Lyu Xueqiang, Zhou Jianshe, et al. Event Sentence Extraction in Financial Field[J]. Application Research of Computers, 2017, 34(10): 2915-2918.)
[12]
Quinlan J R. C4.5: Programs for Machine Learning[M]. San Mateo, CA: Morgan Kaufmann, 1993.
(Cheng Xingguo, Xiao Nanfeng. Parallel Implementation for Co-occurrence Statistics with MapReduce Model[J]. Journal of Chongqing University of Technology (Natural Science), 2013, 27(11): 53-57.)
(Zhong Maosheng, Liu Hui, Liu Lei. Method of Semantic Relevance Relation Measurement Between Words[J]. Journal of Chinese Information Processing, 2009, 23(2): 115-122.)
(Huang Jin, Ruan Tong, Jiang Ruiquan. Sentiment Analysis in Financial Domain Based on SVM with Dependency Syntax[J]. Computer Engineering and Applications, 2015, 51(23): 230-235.)
(Zhang Hongkuan, Song Hui, Wang Shuiyi, et al. A BERT-Based End-to-End Model for Chinese Document-level Event Extraction[C]// Proceedings of the 19th Chinese National Conference on Computational Linguistics. 2020: 390-401.)
[17]
Devlin J, Chang M W, Lee K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[C]// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies, 2019: 4171-4186.
[18]
Yang Z L, Dai Z H, Yang Y M, et al. XLNet: Generalized Autoregressive Pretraining for Language Understanding[C]// Proceedings of the 33rd International Conference on Neural Information Processing Systems. 2019: 5753-5763.
[19]
Liu Y H, Ott M, Goyal N, et al. RoBERTa: A Robustly Optimized BERT Pretraining Approach[OL]. arXiv Preprint, arXiv: 1907.11692.
(Liu Huan, Zhang Zhixiong, Wang Yufei. A Review on Main Optimization Methods of BERT[J]. Data Analysis and Knowledge Discovery, 2021, 5(1): 3-15.)
[21]
Sun Y, Wang S H, Li Y K, et al. ERNIE: Enhanced Representation Through Knowledge Integration[OL]. arXiv Preprint, arXiv: 1904.09223.
[22]
Zhang Z Y, Han X, Liu Z Y, et al. ERNIE: Enhanced Language Representation with Informative Entities[C]// Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019: 1441-1451.
[23]
Lan Z Z, Chen M D, Goodman S, et al. ALBERT: A Lite BERT for Self-supervised Learning of Language Representations[C]// Proceedings of the 8th International Conference on Learning Representations. 2020: 1-17.
[24]
Do Q X, Chan Y S, Roth D. Minimally Supervised Event Causality Identification[C]// Proceedings of the Conference on Empirical Methods in Natural Language Processing. 2011: 294-303.
[25]
Hashimoto C, Torisawa K, Kloetzer J, et al. Generating Event Causality Hypotheses through Semantic Relations[C]// Proceedings of the 29th AAAI Conference on Artificial Intelligence. 2015: 2396-2403.
[26]
Luo Z Y, Sha Y C, Zhu K Q, et al. Commonsense Causal Reasoning Between Short Texts[C]// Proceedings of the 15th International Conference on Principles of Knowledge Representation and Reasoning. 2016: 421-430.
[27]
Staliūnaitė I, Gorinski P J, Iacobacci I. Improving Commonsense Causal Reasoning by Adversarial Training and Data Augmentation[C]// Proceedings of the 35th Conference on Innovative Applications of Artificial Intelligence. 2021: 13834-13842.
[28]
Sharma A, Kiciman E. Causal Inference and Counterfactual Reasoning[C]// Proceedings of the 7th ACM IKDD CoDS and 25th COMAD. 2020: 369-370.
[29]
Han M Y, Wang Y L.Doing Good or Doing Right? Exploring the Weakness of Commonsense Causal Reasoning Models[C]// Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing. 2021: 151-157.
(Zhang Dongdong, Peng Dunlu. ENT-BERT: Entity Relation Classification Model Combining BERT and Entity Information[J]. Journal of Chinese Computer Systems, 2020, 41(12): 2557-2562.)
[31]
Bai C Y, Pan L M, Luo S L, et al. Joint Extraction of Entities and Relations by a Novel End-to-End Model with a Double-Pointer Module[J]. Neurocomputing, 2020, 377: 325-333.
doi: 10.1016/j.neucom.2019.09.097
