Law Article-Enhanced Legal Case Matching a Causal Learning Approach

2025-05-03 0 0 2.21MB 10 页 10玖币

侵权投诉

Law Article-Enhanced Legal Case Matching:

a Causal Learning Approach

Zhongxiang Sun

Gaoling School of Articial

Intelligence

Renmin University of China

Beijing, China

sunzhongxiang@ruc.edu.cn

Jun Xu∗

Gaoling School of Articial

Intelligence

Renmin University of China

Beijing, China

junxu@ruc.edu.cn

Xiao Zhang

Gaoling School of Articial

Intelligence

Renmin University of China

Beijing, China

zhangx89@ruc.edu.cn

Zhenhua Dong

Noah’s Ark Lab, Huawei

Shenzhen, China

dongzhenhua@huawei.com

Ji-Rong Wen

Gaoling School of Articial

Intelligence

Renmin University of China

Beijing, China

jrwen@ruc.edu.cn

ABSTRACT

Legal case matching, which automatically constructs a model to

estimate the similarities between the source and target cases, has

played an essential role in intelligent legal systems. Semantic text

matching models have been applied to the task where the source and

target legal cases are considered as long-form text documents. These

general-purpose matching models make the predictions solely based

on the texts in the legal cases, overlooking the essential role of the

law articles in legal case matching. In the real world, the matching

results (e.g., relevance labels) are dramatically aected by the law

articles because the contents and the judgments of a legal case

are radically formed on the basis of law. From the causal sense,

a matching decision is aected by the mediation eect from the

cited law articles by the legal cases, and the direct eect of the key

circumstances (e.g., detailed fact descriptions) in the legal cases.

In light of the observation, this paper proposes a model-agnostic

causal learning framework called Law-Match, under which the legal

case matching models are learned by respecting the corresponding

law articles. Given a pair of legal cases and the related law arti-

cles, Law-Match considers the embeddings of the law articles as

instrumental variables (IVs), and the embeddings of legal cases as

treatments. Using IV regression, the treatments can be decomposed

into law-related and law-unrelated parts, respectively reecting the

mediation and direct eects. These two parts are then combined

with dierent weights to collectively support the nal matching

prediction. We show that the framework is model-agnostic, and a

∗

Jun Xu is the corresponding author. Work partially done at Engineering Research Cen-

ter of Next-Generation Intelligent Search and Recommendation, Ministry of Education.

Permission to make digital or hard copies of all or part of this work for personal or

classroom use is granted without fee provided that copies are not made or distributed

for prot or commercial advantage and that copies bear this notice and the full citation

on the rst page. Copyrights for components of this work owned by others than the

author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or

republish, to post on servers or to redistribute to lists, requires prior specic permission

and/or a fee. Request permissions from permissions@acm.org.

SIGIR ’23, July 23–27, 2023, Taipei, Taiwan

ACM ISBN 978-x-xxxx-xxxx-x/YY/MM. . . $15.00

https://doi.org/10.1145/nnnnnnn.nnnnnnn

number of legal case matching models can be applied as the under-

lying models. Comprehensive experiments show that Law-Match

can outperform state-of-the-art baselines on three public datasets.

CCS CONCEPTS

•Applied computing →Law

;

•Information systems →Con-

tent analysis and feature selection.

KEYWORDS

Legal Case Matching, Causal Inference, Law

ACM Reference Format:

Zhongxiang Sun, Jun Xu, Xiao Zhang, Zhenhua Dong, and Ji-Rong Wen.

2023. Law Article-Enhanced Legal Case Matching: a Causal Learning Ap-

proach. In Proceedings of the 46th International ACM SIGIR Conference

on Research and Development in Information Retrieval (SIGIR ’23), July

23–27, 2023, Taipei, Taiwan. ACM, New York, NY, USA, 10 pages. https:

//doi.org/10.1145/nnnnnnn.nnnnnnn

1 INTRODUCTION

Legal case matching has played an important role in intelligent legal

systems. For example, in legal case retrieval, the matching models

help the system to determine the relevance between the query

cases and the candidate cases. Traditionally, the task is formalized

as matching two long-form text documents at the semantic level.

General-purpose document matching models have been adapted

to tackle the problem, including the heuristic methods [

network-based methods [7, 8], and text-based methods [29, 41].

Though eective, simply considering the legal cases as general

long-form text documents [

] still has spaces for improvement.

One striking dierence between legal cases and general documents

is that legal cases usually cite a number of law articles

. These

law articles are selected from the law book (e.g., Chinese Criminal

Law) by the judges and provide essential knowledge of the legal

case’s context and judgments. Existing studies have shown that law

articles are benecial to a number of legal-related tasks [

Law articles are the foundation of statutes or written laws which are usually enacted

by the administration of justice (e.g., Criminal Law in China).

arXiv:2210.11012v2 [cs.IR] 26 Apr 2023

SIGIR ’23, July 23–27, 2023, Taipei, Taiwan Zhongxiang Sun, Jun Xu, Xiao Zhang, Zhenhua Dong, & Ji-Rong Wen

Article 163:

[The crime of Bribery]

Company and enterprise

work personnel, who, in the

course of economic contacts,

receive personal kick-backs

and commissions in various

forms in violation of state

rules …….

Article 246:

[The crime of Libel]

Those openly insulting

others using force or other

methods or those fabricating

stories to slander others, if

the case is serious, are to be

sentenced to three years in

prison ......

Case A:

Fact Description: The defendant, Zhou **，slandered Li ** for taking

bribes out of revenge which made a huge negative impact on Li ** ‘s life

and work ……

Cited law article: PRC Criminal Law, Article 246.

Case B:

Fact Description: The defendant, Hua ** took advantage of his position to

take bribes, making a profit of more than 1,000,000 YUAN during the

period ......

Cited law article: PRC Criminal Law, Article 163.

Case C:

Fact Description: The defendant, Le **, for the purpose of profit,

created chat groups to attract gambling participants to join, gambling

activities using the games on Xingyue Mahjong APP ……

Cited law article: PRC Criminal Law, Article 303.

Case D:

Fact Description: The defendant , Xu ** and Xu ** opened a gambling

shop in the form of "small village" in Xu ** convenience store in Wenling

City, and called on gamblers to gamble in the shop ……

Cited law article: PRC Criminal Law, Article 303.

Label: mismatch

Label: match

Article 303:

[The crime of Gambling]

Whoever, for the purpose of

reaping profits, assembles a

crow to engage in gambling,

opens a gambling house, or

makes an occupation of

gambling is to be sentenced

to not more than three

years …...

PRC Criminal Law

Legal case pair

Figure 1: Left: two pairs of legal cases; Right: three cited law

articles in the legal cases. (translated from Chinese)

Analysis shows that the law articles are also important in legal

case matching. Figure 1 shows the snapshots of two real legal case

pairs

. Contents of the cited law articles are listed in the right part of

the gure. In the rst legal case pair, Case A and Case B share a large

number of words in their fact descriptions. However, the judges’

decisions are: Case A is libel crime (PRC Criminal Law, Article

246) while Case B is bribery crime (Article 163). The associated law

articles are helpful in identifying the key information (highlighted)

in the two cases [

]. By comparing the key information in the two

cases, experts annotate the matching label as “mismatch” though

they have relatively high semantic text similarity (measured by

Lawformer [

]). In the second example of Figure 1, Case C and

Case D have relatively lower semantic text similarity than the

previous pair. However, both of them are judged as the gambling

crime (Article 303). The law article helps to identify similar key

information (highlighted) in these two legal cases. So the expert-

annotated matching label is “match”.

Usually, the key constitutive elements and the key circumstances

provide important signals for the matching of two legal cases [

The key constitutive elements are highly summative texts written

in light of some law articles. The key circumstances, on the other

hand, are detailed fact descriptions and are usually very dierent

from case to case. They are not directly related to any law articles.

Therefore, it is possible that law articles can help the matching

model to identify and decompose these key information.

From the causal sense, the matching of two legal cases is aected

by the mediation eect from the law articles and the direct eect

from the key circumstances part of legal cases. More specically,

the key constitutive elements in the legal cases mediate the law

articles’ eect on the matching decision (i.e., the mediation eect).

In contrast, the key circumstances directly aect the matching

decision (i.e., the direct eect). As a result, the embedding of a legal

case actually consists of two parts: the law-related part, which is

the mediator of the mediation eect, and the law-unrelated part,

which has direct eect. These two parts reect dierent association

2Crawled from http://faxin.cn and translated to English.

mechanisms between the legal cases and the matching decisions. It

is necessary to identify and treat them dierently.

To address the issue, this paper proposes a causal representa-

tion learning framework tailored for legal case matching, called

Law-Match. Specically, Law-Match considers the legal cases as

treatment and the corresponding law articles as instrument variables

(IVs) [

]. In the matching phase, after getting the embed-

dings of the legal cases (i.e., treatments) and the related law articles

(i.e., IVs), Law-Match rst uses the IVs to regress the treatments,

resulting in the tted vector (law-related part) and the residuals

(law-unrelated part). These two parts have dierent eects on the

nal matching. Law-Match then combines them into a newly recon-

structed treatment vector with the attention mechanism. Finally, the

reconstructed treatment is fed to the underlying matching model

for making the nal matching prediction. In the training phase,

an alternative optimization procedure is developed to learn the

parameters in the IV regression and matching models.

We summarize the major contributions of the paper as follows:

(1) We analyze the essential role of law articles in legal case

matching from a causal view: the matching decisions are aected

by the mediation eect of the law articles and the direct eect of

the key circumstances in the legal cases.

(2) We propose a novel model-agnostic causal learning frame-

work which introduces the law articles into the process of legal case

matching in a theoretically sound way. IV regression is adopted

to decompose the mediation eect and direct eect from the legal

case embeddings by considering law articles as IVs and legal cases

as the treatments.

(3) We conducted extensive experiments on three public datasets.

Experimental results demonstrated that Law-Match could signi-

cantly improve the underlying models’ performance and outper-

form the baselines, verifying the importance of the law articles in

legal case matching.

2 RELATED WORK

2.1 Legal case matching

Conventionally, legal case matching can be addressed with manual

knowledge engineering (KE) [

]. The methods include the Boolean

search technology and manual classication [

]. With the devel-

opment of NLP, deep learning has been adapted to realize seman-

tic level matching of legal cases. According to [

], these studies

can be categorized as network-based and text-based methods. The

network-based methods are tailored for common law and use the ci-

tations of dierent cases to construct a Precedent Citation Network

(PCNet). For example, [

] use PCNet-based Jaccard similarity to

infer the paired legal cases’ similarity. Bhattacharya et al

. [8]

use

Node2vec to map the nodes of the graph to a vector space and then

compute the legal cases’ cosine similarity. See also [7, 21].

The text-based methods compute the semantic similarity be-

tween legal cases. Shao et al

. [29]

utilize BERT to capture the

semantic relationships at the paragraph level and then infer the

relevance between two cases by aggregating the paragraph-level

interactions. Xiao et al

. [41]

release the longformer-based [

] pre-

trained language model to get a better representation of long legal

documents. Yu et al

. [46]

propose a three-stage explainable legal

case matching model. Law articles have shown their eects on a

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

LawArticle-EnhancedLegalCaseMatching:aCausalLearningApproachZhongxiangSunGaolingSchoolofArtificialIntelligenceRenminUniversityofChinaBeijing,Chinasunzhongxiang@ruc.edu.cnJunXu∗GaolingSchoolofArtificialIntelligenceRenminUniversityofChinaBeijing,Chinajunxu@ruc.edu.cnXiaoZhangGaolingSchoolofArtificialI...

展开>> 收起<<

Law Article-Enhanced Legal Case Matching a Causal Learning Approach.pdf

共10页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Law Article-Enhanced Legal Case Matching a Causal Learning Approach

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: