Recommended Method of Legal Articles for Legal Judgment
Documents
Piqiang Xiao
Jiangxi Institute of Applied Science and Technology, Nanchang, Jiangxi, 330038, China
Keywords: Judgment Documents, Law Recommendations, Smart Judiciary, Model Fusion.
Abstract: With the advancement of artificial intelligence and big data, my country's judicial organs have already
proposed to use intelligent judicial services as the purpose, and then build a period of "smart justice". Under
the continuous leadership of the court, the Judgment Document Network and the China Law Application
Digital Network service platform were launched. At the same time, such a plan can support the construction
of smart courts and propose the next step to promote the development of laws and regulations. In 2018, the
Ministry of Science and Technology issued a new plan, focusing on the task of research and development of
judicial topics. Intelligent services will promote the basic scientific issues of intelligent justice in my
country's law, procuratorate, and division. Intelligent services can promote justice. Favorable goal of
informatization to intelligent development. This article mainly takes "comprehensive rule of law" as a
strategic partner, actively strengthens the technology and application of artificial intelligence, and
encourages the realization of artificial intelligence laws. At the same time, during the same period, the
national smart judicial construction system should be used as a foundation, Legal documents can record the
trial results of the people's court, mainly information about the case, the name of the case, and the
development of the case. Legal provisions are the legal responsibilities that can limit the people's right to
use and personal obligations, and are very important for the court to judge when making a judgment.
Zaogen drama takes criminal law as an example of relevant documents as experimental data. When it comes
to using the recommended model of laws and regulations, there will be many problems, which is how to
solve them. For example, there are relatively small differences between different laws and regulations. A
document will involve a lot of laws and regulations, so it will be difficult to determine how many
corresponding laws and regulations are there. For example, the issues described in different adjudication
documents are also different, but the similarity in content and structure between them is also very high.
1 INTRODUCTION
In recent years, many tasks and deep learning
techniques have been greatly affected. For a
vocabulary and SKip-gram model that can be trained
from a large amount of text, these phrases have
certain problems in meaning and sentence meaning,
in the English classification has also obtained a
more excellent effect. The characteristic of obtaining
the full text by Kalchbrenner technology is
proposed, which can achieve better results in
English classification. At this stage, in order to be
able to use a variety of neural network systems for
the task of text classification, the legal judgment
prediction is verified after the machine input, and
the final result of the legal case can be output. This
work is a work that can be realized for a long time at
home and abroad, but these works are limited to a
specific case and can be regarded as generalization
problems that can be encountered elsewhere. In
2021, scholar Li Ru defined judgment documents in
an article on legal judgment prediction based on
legal judgment documents. In 2019, scholar Zhang
Hu came up with the prediction results on the
recommended method of legal articles for legal
judgment documents. In terms of statute
recommendation, we can classify and discuss
various issues through a fixed combination of
statutes, and then convert the results according to
different goals. The recommendation of legal
articles plays a very important role in the law. When
dealing with problems in key parts, the use of
artificial intelligence as the point of support is also
the field of law in terms of deep learning, and it is
736
Xiao, P.
Recommended Method of Legal Articles for Legal Judgment Documents.
DOI: 10.5220/0012043200003620
In Proceedings of the 4th International Conference on Economic Management and Model Engineering (ICEMME 2022), pages 736-741
ISBN: 978-989-758-636-1
Copyright
c
2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
also one of the important factors in legal
construction.
In these studies, we can clearly know that the
neural network model combined with the
adjudication document data recommendation
method improves the accuracy of this paper to a
certain extent. There are also certain problems in the
task of recommending legal provisions. For
example, the traditional method is prone to local
problems, thus ignoring the relationship between the
structures of sentences; the data in the process of
using the legal provisions recommendation task
belongs to the law. The differences between
different laws and regulations will also be different,
which will make the traditional classification
method inconsistent with our ideas; on the issue of
legal classification, one problem corresponds to the
situation of multiple laws and regulations. In order
to solve the problem smoothly, the model fusion
method of neural network must be used.
2 NEURAL NETWORK-BASED
LAW RECOMMENDATION
MODEL
2.1 Convolutional Neural Networks
Convolutional neural networks play their own roles
in both image processing and speech, using their
own convolution and pooling structures to operate.
In language processing, convolutional neural
networks can be used to achieve desirable results in
retrieval, text classification, etc. The processing rule
recommendation problem of convolutional neural
network mainly includes three aspects:
convolutional layer, pooling layer, and fully
connected layer. Following a top-to-bottom
principle, the input legal documents can be
identified from this, and the output is the
corresponding legal provisions for the facts. As
shown in Figure 1.
Figure 1: Convolutional Neural Network Architecture Diagram.
2.2 Character-Level Convolutional
Neural Networks
In the case of combining deep learning and language
problems, there are mainly two ways to process this
data, word-based and word-based. When developing
from the perspective of characters, it is characterized
by being able to know the large-scale data for
training without needing to know the meaning of
words in the text in advance. In this neural network
system, the convolution model is the most basic.
Input functions and then perform operations to
obtain results. Specifically as shown in Figure 2.
Recommended Method of Legal Articles for Legal Judgment Documents
737
Figure 2: Schematic diagram of character-level convolutional neural network
3 FUSION OF MULTIPLE
MODELS
First of all, the idea of multi-model fusion is used
for research, and the fusion of different
convolutional neural networks and probability can
achieve the effect of model complementarity and
promote the security of legal recommendations.
Then train the entire model. After basic training, all
the probabilities are sorted in the Softmax layer,
which is the law of prediction. The fusion diagram
of the model is shown in Figure 3.
Figure 3: Fusion structure diagram of the model.
A single case in the recommendation of articles
will have multiple articles. This situation is a
“one-to-many” legal article problem, and traditional
methods should be used to solve this type of
problem. This paper adopts the strategy of
multi-model fusion and result value combination to
carry out the corresponding process. This method is
used to identify which one is better for
"one-to-many" application. As shown in Figure 4.
ICEMME 2022 - The International Conference on Economic Management and Model Engineering
738
Figure 4: Flowchart of the "one-to-many" law problem
4 SYSTEM ANALYSIS
At present, the relevant research on legal judgment
prediction has not been collected. On the issue of
legal recommendation, it is selected through the data
set to realize the authenticity of the case. In order to
verify the validity of the model, each data is sorted
together and further validated and tested by scale.
With regard to the processing of data, the facts of
the various statutes are classified. This approach is
to expand the data set, and then involve multiple
laws, and then experiment with "one-to-many" laws.
There will also be many laws and regulations on the
same thing, but there will be gaps in description. For
example, Article 347 of the Criminal Law involves
many crimes, such as drug crimes, which is different
from Article 348 of the Criminal Law. The offence
involved is the offence of unlawful possession of
drugs. The corresponding statutes between them are
the same, but the meaning of the events is indeed
different. The descriptions between the two cases are
so similar that traditional classification models are
difficult to discern. For this kind of problems, the
fusion model is used to solve the problems of
different differences in the classification of legal
articles.
4.1 Comparison of Model Fusion and
Single Model
The CNN model with the best model fusion effect in
CAIL2018 is compared with CNN\SVM. The
comparison results on the database are shown in
Figure 5.
Recommended Method of Legal Articles for Legal Judgment Documents
739
Figure 5: Experimental comparison results.
The experimental results in Figure 5 show that
the fusion model is dominant in comparison with
other models, indicating that the fusion of the
convolutional neural network model proposed in this
paper is beneficial in the task of law
recommendation.
The analysis of the experiment can know that the
use of the fusion model has a more repeated usage
rate, which is an advantageous advantage of the
convolutional neural network. Combining the fusion
model as the first policy of legal recommendation is
to promote the further development of legal
recommendation.
5 CONCLUSION
In this paper, the method of model fusion is used in
the research on legal recommendation. Its main
functions are: (1) The use of character-level
convolutional neural network for legal
recommendation can produce different effects in
different environments, it can be seen that it is very
helpful for convolutional neural networks. (2) The
use of model fusion can resolve the differences in
legal recommendations. (3) The method of result
value can solve the problem of "one-to-many", and
better solution to the problem page generated by the
model also improves the application value of the
model.
The experimental results can show that it is
necessary to use the right method, and it can better
solve the problem of legal recommendation. It needs
to be improved on this basis, so as to better solve the
problem and develop the next key research goals. .
To achieve the success of the task of recommending
legal articles, the knowledge of legal articles can be
fully utilized, the information of legal articles can be
familiarized, and the experimental performance can
be improved. The significance of this study lies in
ICEMME 2022 - The International Conference on Economic Management and Model Engineering
740
the better use of legal recommendations, to provide
legal assistance to the people, and to help people
who do not understand. The next step of
development is to carry out the next step of
improvement on the usage method recommended by
the law, so as to achieve a more perfect state.
REFERENCES
Chen Haiwen, Wang Shouxiang, Wang Shaomin, Wang
Dan. Load aggregate prediction method based on
gated recurrent unit network and model fusion [J].
Automation of Electric Power Systems, 2019, 43(01):
65-72.
Fan Shixiong, Liu Xingwei, Yu Yijun, Zhang Wei, Li
Lixin. Ultra-short-term bus load prediction method
based on multi-source data and model fusion [J].
Power Grid Technology, 2021, 45(01): 243-250. DOI:
10.13335/j.1000-3673.pst.2020.1167.
Liu Yunting, Yu Qingsong, Li Shenke, Liu Xiaoyu.
Research on intelligent detection method of
multi-model fusion images based on deep learning [J].
Electronic Measurement Technology, 2021, 44(20):
168-174. DOI: 10.19651/ j.cnki.emt.2107217.
Lian Zhipeng, Xu Yong, Fu Sheng, Chen Lixia, Liu Lei.
Using multi-model fusion method to evaluate
landslide susceptibility: a case study of Wufeng
County, Hubei Province [J]. Geological Science and
Technology Bulletin, 2020, 39(03):
178-186.DOI:10.19509/j.cnki.dzkq.2020.0319.
Liu Bo, Qin Chuan, Ju Ping, Zhao Jingbo, Chen Yanxiang,
Zhao Jian. Short-term bus load prediction based on the
fusion of XGBoost and Stacking models [J]. Electric
Power Automation Equipment, 2020, 40(03): 147-153.
DOI:10.16081/j.epae.202002024.
Liang Zhu, Shen Si, Ye Wenhao, Wang Dongbo. Research
on automatic recommendation of judgment documents
based on structural content characteristics [J]. Journal
of Information Science, 2022, 41(02): 167-175.
Pan Guobing, Gong Mingbo, He Min, Wu Chenghuan,
Tang Xiaoqi, Yang Lv, Ouyang Jing. Risk
identification method of electricity bill recovery based
on Stacking model fusion [J]. Electric Power
Automation Equipment, 2021,41(01):152-
160.DOI:10.16081/j.epae.202010022.
Sun Wenqing, Deng Aidong, Deng Minqiang, Liu Yang,
Cheng Qiang. Fault Diagnosis of Wind Turbine
Gearbox Based on Model Fusion [J]. Journal of Solar
Energy, 2022, 43(01): 64-72. DOI: 10.19912/j.0254-
0096.tynxb.2020-0181.
Xu Hongxue, Wang Anqi, Che Weiwei, Du Yingkui, Sun
Wanyou, Wang Yangyang. Microblog text sentiment
analysis model based on multi-model fusion [J].
Journal of Shenyang University (Natural Science
Edition), 2022, 34(02): 112-118+133.
DOI:10.16103/j.cnki.21-1583/n.2022.02.005.
Yin Zhangzhi, Li Xinzi, Huang Degen, Li Jiuyi. Research
on Chinese Named Entity Recognition Fusion Word
Model [J]. Chinese Journal of Information, 2019,33
(11):95-100+106.
Yang Ke, Fang Cheng, Duan Liming. Automatic detection
of casting defects based on deep learning model fusion
[J]. Journal of Instrumentation, 2021, 42(11): 150-159.
DOI: 10.19650/j.cnki.cjsi .J2108170.
Zhou Wei, Wang Zhaoyu, Wei Bin. A generative automatic
summary model for legal judgment documents [J].
Computer Science, 2021, 48(12): 331-336.
Zhang Hu, Pan Bangze, Tan Hongye, Li Ru. Legal
judgment prediction based on legal judgment
documents [J]. Big Data, 2021, 7(05): 164-175.
Zhang Hu, Wang Xin, Wang Chong, Cheng Hao, Tan
Hongye, Li Ru. A method for recommending legal
articles for legal judgment documents [J]. Computer
Science, 2019, 46(09): 211-215.
Zheng Zhicong, Wang Hong, Qi Linhai. Voltage sag
source identification method based on deep learning
model fusion [J]. Chinese Journal of Electrical
Engineering, 2019, 39(01): 97-104+324. DOI:
10.13334/j .0258-8013.pcsee.181337.
Recommended Method of Legal Articles for Legal Judgment Documents
741
