The Development and Improvement of Similar Case Retrieval Mechanism

Abstract： In the era of big data， combining case guidance with the technical means of AI to facilitate case retrieval in the process of implementing the guiding cases system in China will undoubtedly be beneficial to promote the quality of case retrieval and improve the recall and precision ratios of cases， which is conducive to realizing the goal of \"similar cases to be treated similarly\". The development of similar case retrieval mechanism has experienced three stages， namely the embryonic stage， the trial stage and the development stage. The similar case retrieval mechanism has gradually become a long-term mechanism to promote the standardization of trial procedures and encourage judges to make correct judgments. However， in the current judicial practice， the operation of the case retrieval mechanism still has some problems， such as the lack of a rigid constraint and incentive mechanisms， the insufficiency of the total number of cases， too few categories in the database， and the lack of professionalism when labeling guiding cases in the current intelligence system. To promote a comprehensive integration of AI and the guiding cases system and maximize the potential of AI， it is essential to proactively pursue countermeasures， explore various paths to solutions， and foster a culture of continuous innovation. These efforts will provide the case retrieval mechanism with further improvement and optimization， causing it to evolve in a satisfactory direction. In addition to the compulsory retrieval imposed on judges， corresponding punishment and accountability mechanisms must be set up. Relevant departments should strengthen the construction of a unified and complete judicial database to include all typical， instructive， and authoritative guiding cases and reference cases. Sufficient endeavors must be made to label and identify guiding cases so that more judicial workers can use AI to discover and successfully apply them. Labeling should represent the basic facts of cases and relevant statutory rules objectively， clearly， and accurately.

Keywords： artificial intelligence; guiding cases; similar case retrieval mechanism; judicial database

CLC： DF 62 " " " " " " " DC： A " " " " Article： 2096-9783（2024）02?0127?12

A fundamental requirement of formal justice in judicial adjudication is that similar cases be handled similarly. However， as often as not， similar cases are found to have different judgments. In the context of the reform of the judicial accountability system， the similar case retrieval mechanism has become a new measure of the SPC， intended to unify adjudication criteria. Artificial intelligence and its combination with big data provide unprecedented convenience for judges to access judgments of similar cases， thus laying a solid foundation for a similar case retrieval mechanism. Under this mechanism， ordinary cases gain the effect of \"being referred to\" and resemble stare decisis in a common law system. In the era of big data， the acquisition of similar cases has witnessed a quantum leap forward with the help of AI， which is widely recognized and used because it breaks through the barriers of the legal profession[1]. Based on pilot practices by courts in many places， Guiding Opinions on Uniform Application of Laws and Strengthening Similar Case Retrieval （hereinafter referred to as Guiding Opinions） follows the foregoing trends and formally affirms the validity of similar cases. In general， Guiding Opinions is relatively conservative and prudent; it provides preliminary provisions on the identification conditions， retrieval subjects， retrieval objects， and methods of similar cases. At the conceptual level， the foregoing general characteristics of Guiding Opinions reflect judicial practitioners' prudent attitudes toward the intervention of artificial intelligence in adjudication.

1 Necessity for the Similar Case Retrieval Mechanism

Since the formal implementation of the guiding cases system in China， the application of guiding cases in judicial practice has played a unique role by unifying the application of law and ensuring judicial justice to a certain extent. However， the application effect of guiding cases is not ideal in general， and their active application by judges only accounts for a lower proportion. Some scholars have calculated statistics： \"As of December 31， 2017， the Supreme People's Court had issued a total of 17 sets of 92 guiding cases， in which 60 guiding cases have been applied to judicial practice， and 32 cases have not yet been applied. There are 1 571 times be cited by judges， namely application cases， in which 580 cases are explicitly cited and 980 cases are implicitly cited\"[2]. These statistics show that few guiding cases have been applied in judicial trials. According to the work report of the Supreme People's Court in 2023， the Court accepted 18 547 cases and concluded 13 785 cases in 2022. In the same year， local people's courts at all levels accepted 33.7 million cases and concluded 30.81 million cases[3]. This reveals that the total number of cases tried by courts at all levels in China has reached nearly 30 million every year. When comparing the vast number of accepted cases to the limited amount applicable to guiding cases， it becomes evident that the effectiveness of guiding cases falls short of the system's initial expectations. This is further supported by the relatively low citation rate of guiding cases， highlighting the need for improvement in their application. The reason for this poor application effect is related to the low intelligence level of the similar case retrieval database， which makes it difficult for judges to find and cite cases similar to pending cases. To implement the guiding cases system， therefore， it is necessary to further develop and enrich the technical means， enhance intelligent case recognition capability， and improve the level of intelligent retrieval.

We know that the objective for \"intelligent information retrieval is to analyze， reason， and make decisions on the retrieved contents by using AI based on the traditional information retrieval， and present them to users in a satisfactory form. In addition to the traditional fast retrieval， relevance ranking， and other functions， it also provides the functions of user role registration， automatic recognition of user interests， semantic understanding of content， intelligent information filtering and push\". The same is true for judicial AI information retrieval. For example， Westlaw， an intelligent case retrieval database in the United States， is a comprehensive retrieval and analysis system for big data legal cases centering on the retrieval of American cases[4]; it can retrieve in accordance with the case number， the name of the trial judge， key words， the names of both parties， and the company name. Intelligent retrieval methods， created in Westlaw， can， for example， use of the West Key Number to limit the scope of the theme， refine the important points of dispute and the application of law in each precedent， and use the concise title to indicate the corresponding key code so that the case and the applicable law are systematically arranged and classified. Westlaw's system can carry out professional， accurate identification and intelligent push for similar cases， making it convenient for users to retrieve and search.

Therefore， combining case guidance with the technical means of AI to facilitate case retrieval in the process of implementing the guiding cases system in China will undoubtedly be beneficial to promote the quality of case retrieval and improve the recall and precision ratios of cases. At present， the overall application level of AI in China's judicial field is not high， and its boosting effect on guiding case retrieval has not been fully demonstrated. However， as far as existing relevant intelligent systems are concerned， the application of AI does help to improve the efficiency of guiding case retrieval. For instance， some intelligent systems developed during smart court construction have set up case codes， key words， and applicable laws in the text style of published guiding cases. Combined with systematic arrangement and classification of AI， these settings can help judges distinguish the difference between pending cases and guiding cases. Once judges have mastered the case database retrieval method， they can retrieve cases efficiently without complicated retrieval procedures. In addition， the guiding case's key points of judgment are set out， enabling the judge to see the formation process and grasp the key points of the whole case， and the judgment key points of different cases are intelligently refined and compared in a way that is convenient for judges to master. The system significantly reduces the time previously spent by judges on finding， comparing， and studying cases.

At present， few guiding cases are available， and the difficulty of retrieval has not yet been eased. However， as the system improves， the number of cases will gradually increase. Consequently， judges face challenges in swiftly and accurately searching for cases. To address these concerns， it is essential to enhance the role of AI in case guidance， ensuring efficient retrieval and referencing of guiding cases. By advancing AI technology， favorable outcomes can be achieved.

2 Development Periods of the Similar Case Retrieval Mechanism

In recent years， the Supreme People's Court has been paying close attention to the similar case retrieval mechanism， constantly refining provisions for its advancement. The similar case retrieval mechanism is at the core of the intelligent administration of justice. Its establishment helps achieve \"similar judgment for similar cases\" and \"uniform judicial application\". It promotes the application of cases and enables courts at all levels to make similar judgments when handling similar cases.

The main periods are as follows：

2.1 The Embryonic Stage

The period during which similar case retrieval was proposed. On the implementation of Rules on the Work of Case Guidance in 2010， the Supreme People's Court stipulated the validity of guiding cases. Article 7 stipulates， \"Guiding cases issued by the Supreme People's Court shall be referred by courts at all levels in handing similar cases\". In effect， the guiding cases' normative binding force is founded on institutional authority[5]. The use of the word \"shall\" means that judicial staff should first use similar case retrieval to identify whether similar cases exist. Any found shall be cited and applied proactively. Several Opinions of the Supreme People's Court on Improving the Judicial Accountability System of People's Courts， issued by the Supreme People's Court in 2015， mentioned that \"we should establish a legal symposium mechanism for trials and unify adjudication criteria through references to similar cases and case analysis etc\". The reference to similar cases is conducive to promoting similar cases with similar judgements and uniform adjudication criteria. Article 10 of the Detailed Rules for the Implementation of the Provisions of the Supreme People's Court on Case Guidance issued by the Supreme People's Court， provides that \"where a guiding case is referred by people's courts at all levels in handling similar cases， it shall be cited as the reason of the judgement\". From the perspective of methodology， Article 13 stipulates that \"the Supreme People's Court shall maintain paper archives and establish an electronic information database for guiding cases to guarantee the application mutatis mutandis， consultation， retrieval and compilation of guiding cases\". It can be seen that the SPC is attempting to establish an electronic database to facilitate the reference and application of guiding cases. This article also mentioned the term retrieval， the scope of which is mainly guiding cases. Although the normative binding force of guiding cases was established in this period without defining a similar case retrieval mechanism， the concept of retrieval was proposed.

2.2 The Trial Stage

The period during which the similar case retrieval mechanism was clarified. On August 1， 2017， Article 39 of Opinions of the Supreme People's Court on the Implementation of the Judicial Accountability System （for Trial Implementation） laid down the requirement that \"in the trial of a case， a judge handling the case shall， on the case handling platform， file system， China Judgments Online， www.faxin.cn， the Intelligent Trial System， etc.， comprehensively search the similar cases and correlated cases closed concluded or under trial and develop a report on the searching of similar cases and correlated cases\". The words \"shall all\" （均應(yīng)） indicates that the requirements for case retrieval are mandatory and that a case retrieval report should be prepared. In 2018， the Supreme People's Court issued Implementation Opinions on Judicial Accountability System1， in which it was clearly pointed out that it was necessary to \"establish a compulsory retrieval mechanism for similar cases and related cases\" to ensure uniform adjudication criteria for similar cases. In February 2019， the Supreme People's Court issued Opinions on Deepening the Comprehensive Supporting Reform of the Judicial System of the People's Courts， emphasizing \"improving the working mechanism for compulsory retrieval reports of similar and novel cases\". This was also the first time that compulsory retrieval became a requirement for \"novel cases\". It can be seen that the promulgation of the series of documents standardized similar case retrieval and proposed a preliminary plan for the establishment of a similar case retrieval system. At the same time， similar case retrieval， uniform adjudication criteria， and uniform application of laws were combined. In this period， similar case retrieval became compulsory.

2.3 The Development Stage

The period of improvement and practice of the similar case retrieval mechanism. In 2019， the Supreme People's Court issued Measures for Implementation of Establishing a Mechanism for Resolving Disagreement in the Application of Laws， emphasizing the strict implementation of a system of \"compulsory retrieval of similar cases and novel cases\". Article 3 provided that \"in organizing the special study on similar cases with the same judgment in people's courts\"， if any disagreement were to arise in the application of laws in a similar case， the law institute should apply to the review administration office to resolve it. On July 31， 2020， the Supreme People's Court issued Guiding Opinions. This further clarified the concept of similar cases， the scope of similar case retrieval， the available methods， and the construction of a case database. Opinions of the Supreme People's Court on Improving the Working Mechanism for Unifying the Application of Law， implemented in September 2020， improved the working mechanism for compulsory retrieval reports of similar cases and novel cases， and encouraged courts at all levels to use existing platforms to conduct similar case retrieval and analysis and unify the application of laws.

The underlying purpose for establishing a similar case retrieval mechanism is to unify the application of laws. Similar case retrieval is the \"golden key\" to achieving fairness and justice， which is conducive to the enhancement of judicial credibility. To a large extent， the retrieval of similar cases avoids the narrow and limited understanding of judicial staff because the reference to and absorption of others' opinions enables more comprehensive considerations， fairer and more reasonable judgments， and consistency with previous authoritative judgments， thus greatly promoting uniform adjudication. The phenomenon of \"similar cases with different judgments\" caused by the inconsistent application of laws will seriously damage judicial credibility. With accelerated information， different judgments by different courts in similar cases will be easily understood and recognized by the public. Particularly for parties closely related to the cases being handled， the inconsistent application of laws results in a more serious loss of judicial credibility. The similar case retrieval mechanism is about reference to previous similar cases and assists judges in ensuring the uniform application of legal norms when dealing with similar cases. This is completely consistent with the main purpose of the guiding cases system， at whose core is the identification and judgment of similar cases. If effective identification and judgment of similar cases are not available， reference to similar cases will naturally be unsustainable. In this sense， accurate retrieval and identification of similar cases are logical premises for reaching the same judgment in similar cases. Now that some guiding cases have been released， judges have started to cite the adjudication rules created by the Supreme People's Court to achieve uniform adjudication in practice out of various considerations. However， the very limited quantity and coverage of guiding cases have led to many problems in practice that are yet to be solved; more support for the similar case retrieval mechanism is needed[6]. This is an important background for the establishment of a similar case retrieval mechanism.

3 The Problem of Applying a Similar Case Retrieval Mechanism

The similar case retrieval mechanism aims to assist judges in finding highly matched similar cases and utilizes the adjudication rules from previous similar cases as a benchmark to caution judges in pending cases against excessive deviation from this standard[7]. The implementation of this mechanism has encountered many practical problems due to of the unclear legal effect of guiding cases， lack of obligatory provisions for similar case retrieval， and lack of a unified judicial database. Problems arising from the application of the similar case retrieval mechanism in practice are summarized in the subsections below.

3.1 Compulsory Retrieval Lacks Rigid Constraint and Incentive Mechanisms

Article 9 of Guiding Opinions on Uniform Application of Laws and Strengthening Similar Case Retrieval stipulates， \"Where a retrieved similar case is a guiding case， the people's court shall make a judgment by reference to the case， except for those which conflict with new laws， administrative regulations， or judicial interpretations， or those which have been replaced by new guiding cases\". The provision obliges legal staff to retrieve guiding cases. Without special causes or reasons， guiding cases should be drawn from or referred to as similar cases. To some extent， the retrieval of similar cases and analysis of their judgements will reduce the possibility of \"similar case with different judgment\"， thus satisfying the public's demand for \"similar cases with similar judgments\"[8].

The local courts have also taken the initiative to explore. The compulsory similar case retrieval mechanism and the mechanism for solving disagreements in the application of laws have now been established， forming a standard system for similar cases. However， the compulsory retrieval system does not work well in judicial practice for the following four reasons：

1. On the whole， similar case retrieval is a new obligatory provision for judges and courts that handle cases， and it currently demands more time and energy. Senior judges focus more on existing experience， social experience， and other extrajudicial factors， while their unfamiliarity with information technology also makes them fear or resist similar case retrieval[9]. Young judges without practical experience， on the other hand， are more willing to use similar case retrieval in reaching adjudication[10]. The large scope and quantity of retrieved similar cases， caused by poor authority and relevance， limits the guiding significance of similar cases， so judges still have to further screen and sort retrieved cases. However， judges' non-proficiency in system operation led to a great increase in their workload rather than the improvement in work efficiency that similar case retrieval was intended to accomplish[11].

2. The existing retrieval mechanism is highly subjective and lacks objective constraints or type-specific setting and rationality. Although mandatory retrieval obligations are stipulated， no effective safeguards and specific guidelines exist for the obligations that should be fulfilled; neither are there any provisions on the accountability for failure to carry out similar case retrieval in Guiding Opinions.

3. The existing system lacks an effective incentive mechanism. Guiding Opinions has no diverse incentive mechanism for encouraging judges to proactively perform the obligation of similar case retrieval; this means that judges' enthusiasm is difficult to mobilize.

4. There is no external supervision mechanism. The current similar case retrieval mechanism is limited to the internal judicial system. The active roles of parties， lawyers， experts， and scholars in the adjudication mechanism of similar cases has not been introduced. Therefore， a complete legal community system cannot be formed for the supervision of effectiveness.

3.2 The Defects of the Current Judicial Database

Unsatisfactory aspects remain concerning the data in the current similar case retrieval database. At the initial operation stage， in particular， the inadequate supply of guiding cases means that there are certain defects in the guiding cases in the database. These include inadequate total data， too few categories， a unitary data source， inaccuracy of some legal data， and incomplete records of decision making during the trial process. To be blunt， the data on guiding cases is inadequate， incomprehensive， and untrue.

Although current guiding cases published by the Supreme People's Courts are representative and typical， they are relatively few and lack diversity. Some scholars have discussed this problem， including Zhou Shaohua， who wrote， \"As of today， the guiding cases published by the supreme judicial organ are still extremely limited in terms of quantity and are far from enough to give all-round guidance to all cases\"[12]. Chen Huali noted that \"'guiding cases' are quite limited in terms of either quantity or categories and are far from satisfying the demand in trials of intellectual property rights\"[13].

In addition to the inadequate quantities and categories of existing guiding cases， there are also deficiencies in terms of their spatial and temporal distribution. Since the history of the guiding cases system in China is relatively short and the timespan of normative guiding cases is not long enough， it is difficult to draw relatively stable judicial experience under different historical conditions. This has led to a certain limitation in the experience of guiding cases in the time dimension. Without adequate quantity， the role of guiding cases in the similar case retrieval mechanism will be virtualized because a judge may directly read the whole text of corresponding cases to make comparisons with similar cases without relying on the retrieval of similar cases. At the same time， the excessive inadequacy of guiding cases cannot satisfy the demand in judicial practice and negatively influences the operation of the guiding cases system. \"If the Supreme People's Court is over-prudent and issues guiding cases whose quantity and categories are far from enough， their role in providing guidance to nationwide trial is limited\"[14]. It is worth mentioning that current research and development and application of legal artificial projects in China rely mostly on the legal database of China Judgments Online. This website was established in 2013， and all judgment documents were uploaded from that point on[15]. However， there are rare electronic versions of judgment documents created before 2013 that offer limited historical data for reference. This circumstance means that there is a time limitation on guiding cases （including guiding cases published by the Supreme People's Court and reference cases published by high people's courts） required by legal artificial intelligence. Viewed from the source of guiding cases， there is uneven distribution among provincial-level administrative regions of China. For example， the sources of the 211 guiding cases （as of September 2023） published by the Supreme People's Court are as follows： 65 cases from the Supreme People's Court， 28 from Jiangsu Province， 21 from Shanghai Municipality， 13 from Zhejiang Province， 12 from Beijing Municipality， 8 from Shandong Province， 8 from Jiangxi Province， 8 from Chongqing Municipality， 7 from Sichuan Province， 5 from Anhui Province， 5 from Hunan Province， 5 from Tianjin Municipality， 5 from Guangdong Province， 3 from Henan Province， 2 from Inner Mongolia， 2 from Shanxi Province， 2 from Guizhou Province， and one each from Heilongjiang Province， Jilin Province， Gansu Province， Fujian Province， and Hubei Province（some cases did not mention the specific province）[16]. This list shows significant regional differences among the sources of guiding cases， with the regional distribution so unbalanced that some administrative regions contributed no guiding cases at all.

Inadequate quantities and categories， unbalanced spatial and temporal distribution-and sometimes an inaccurate database algorithm-all limit the legal data resources on which AI relies. This not only directly influences in-depth learning and computational analysis by artificial intelligence， but also directly impedes AI intervention in case guidance， causing difficulties in coping with new， difficult， and complicated cases to be heard and resulting in a reduction in the guiding function of cases.

3.3 Lack of Professionalism When Labeling Guiding Cases in the Current Intelligence System

The main problems include unsatisfactory extraction and labeling of elements in some cases， which， together with low professionalism， make guiding case retrieval and use inconvenient.

The essence of \"similar judgment for similar cases\" is to seek definite and unified judgment rules by reference to which the application of law may be completed where there are substantial similarities between the pending case and the similar case. This makes for true unification of the judgment scale. In practice， however， there is usually more than one similar case， and a certain number of those will have diverse judgments， so judges face tough problems when trying to decide which of the similar cases should be referred[17]. Identification of similarities between cases requires deliberation between facts in the case and legal norms， so the basic elements of guiding cases should cover both factors. According to Opinions on Style for Compiling and Submitting Guiding Cases and the Format for Guiding Cases issued by the Supreme People's Court， the basic elements of guiding cases include key words， key judgment points， basic facts， judgments， and reasons for judgments. When identifying pending and guiding cases in the database， the AI-aided judicial system will， following an algorithm designed by a programmer， automatically identify and match the main facts and focus of disputes in the indictment with the basic elements in guiding cases. The identification of similarities relies on prior labeling of the basic elements of the guiding cases; this is the first step of information identification in artificial intelligence， and directly affects the identification speed and accuracy of the AI-aided judicial system. Labeling should therefore represent the basic facts of cases and relevant statutory rules objectively， clearly， and accurately; fuzzy or even incorrect labeling of basic elements will surely cause incorrect identification.

An examination of the current 211 guiding cases （as of September 2023） published by the Supreme People's Court and included in the intelligent database shows that some problems with the labeling of basic elements remain.

In the list below， I discuss two labeling problems of guiding cases included in the intelligent database， taking key words that directly affect the retrieval， reference， and application of guiding cases as examples.

1. The extracted key words are too abstract and have no pertinence or directivity. It is well known that accurate and appropriate extraction and labeling of key words will help identify the main points of cases， revealing the core disputes and law application involved in the whole case. In this way， judicial workers can develop a general idea about the main focuses of disputes， legal relationships to be adjusted， and so on without reading the whole judgment document. Since guiding cases are generally typical and novel， the key words should contain complete， clear， correct， pertinent， and representative information and point directly to the case's core disputed issues so as to facilitate automatic identification by the intelligence system and retrieval by judicial personnel. However， the key words in some cases are sparse and have inclusive connotation and poor distinctiveness; they are thus not helpful for the identification of guiding cases. For example， in administrative guiding cases， most key words are \"administration，\" \"administrative procedure\"， and other abstract words， making it difficult for users to identify relevant cases quickly and accurately. For instance， No. 22 guiding case has only three key words： \"administrative procedure\"， \"case acceptance range\"， and \"reply\". In fact， the case's key point is that the reply of the concerned party was an internal administrative behaviour， not subject to litigation， which became a specific administrative behaviour when actual influences were caused to the rights and interests of the administrative counterpart after implementation. Accuracy and clarity of directivity would have been better served if the key words of the guiding case had included \"internal administrative behaviours\" and \"specific administrative behaviours\".

2. The extracted key words are excessively concrete， lacking conciseness and generalization. To facilitate retrieval， key words should be concise， recapitulative， and immediately understandable. Redundant key words with poor generalization will cause retrieval difficulties or even reduce the rate of guiding cases' references or applications. At present， the key words in some guiding cases are long， narrative phrases or sentences， resulting in poor identification. For example， in No. 2 guiding case， Wu Mei v. Sichuan Meishan Xicheng Paper Co.， Ltd over sales contract disputes， the key words are \"civil procedure\"， \"enforcement\"， \"reconciliation\"， \"appeal denied\"， \"non-performance of reconciliation agreement\"， and \"application for enforcement of judgment of the first instance\". The six key words merely represented the case by stating its development. Furthermore， the sixth key word， \"application for enforcement of judgment of the first instance\"， was too wordy and overlapped with the second key word， \"enforcement\"， indicating inappropriate extraction and labeling of key words. In another example， No. 90 guiding case， HE Huifeng v. Traffic police brigade of Haining Police Bureau over administrative punishment in traffic management， the four key words are \"administration\"， \"administrative punishment\"， \"motor vehicles give way to pedestrians\"， and \"crossing the road at that moment\". Obviously， the fourth key word， \"crossing the road at that moment\"， should be a short phrase but was too long， and the statement is a more specific matter， the connotation was not generally applicable. Redundant and trivial key words like these overstress the particularity of guiding cases and do not favour identification and matching with other similar cases; their use thus reduces guiding cases' retrieval and citation rates. Guiding cases should focus on generalization of application of judgment for the purpose of providing reference to help solve similar general problems， not to pursue justice in an individual case. For guiding cases to be identified， cited， and widely promoted， key words should demonstrate generality， comprehensiveness， and universal significance; there is no need to demonstrate peculiarity.

As the above examples illustrate， certain defects exist in guiding case labeling， in the current intelligent database that cause inconvenience to enquiry and identification. Judicial workers and legal researchers making retrievals often find that intelligent judgment retrieval engines， such as China Judgment Online or the PKULAW database （http：//www.pkulaw.cn/）， return a very large number of cases in response to inputs of terms concerned with cause of action， key words， theme， and focus of disputes. Further manual screening， consuming much time and energy， is needed to obtain guiding cases with closely matching peculiarity and relevance. To add to these difficulties， the intelligent system can neither rank search results by relevance nor screen out cases with similar or conflicting judgments. These problems indicate a lack of professionalism in the labeling of cases （including guiding cases）， causing inconvenience during retrieval and identification. At present， AI and big data analysis offer technological support for retrieval of guiding cases that is still primary and rough[18]. Professionalism in the labeling of basic elements of cases and intelligent correlation with the electronic portfolio are important desiderata for the in-depth combination of artificial intelligence and big data with judicial practice.

4 The Path to Optimization of the Similar Case Retrieval Mechanism

Currently， the application of the similar case retrieval mechanism has become a practical attempt that has attracted much attention in the judicature. Some achievements have been made， and greater development is to be expected. There are also some difficulties and problems with the introduction that make it necessary to seek countermeasures against them， explore the paths to solutions and make constant innovations， thus promoting in-depth fusion of AI and the guiding cases system and giving AI the fullest play possible.

4.1 A Shift from Accountability to Incentive

In addition to the compulsory retrieval imposed on judges， corresponding punishment and accountability mechanisms must be set up. At present， the accountability mechanism is not just an investigation into wrong judgment in terms of results; it is a \"whole-process supervision\" over judicial activities， achieved through a series of normalized appraisal indicators. The reformers transplanted the means of organizational management in enterprises and administrative agencies into the judicial system， expecting to monitor judges' handling of cases and master the operation of litigation procedures using a quantitative performance appraisal system that carries \"premodern\" and \"administrative\" characteristics， giving judicial organs and their personnel a long-term tension of \"double structuring\"[19]. Although the indicators for the adjudication of similar cases are currently not yet complete， intelligent judicial management reform has been promoted in many places. Factors such as \"same cases with same judgment\" and \"similar case retrieval\" may be added to the evaluation system soon. Bearing this in mind， a scientific and reasonable incentive mechanism should be established to encourage judges to constantly carry out compulsory retrieval. For instance， Article 11 of the Rules of the High People's Court of Jiangsu Province on Establishing a Compulsory Similar Case Retrieval Report System （Trial） stipulates that \"the courts at all levels in the province shall include the implementation of the compulsory similar case retrieval report system into the case quality assessment and the judge's trial performance appraisal\". This combines the compulsory retrieval system with the judge's trial performance appraisal and directly grants judges awards and bonuses.

Article 3 of Opinions on Similar Case Retrieval stipulates that \"the judge shall carry out similar case retrieval relying on China Judgments Online and the database of judicial precedents， etc. and be responsible for the authenticity and accuracy of the retrieval\". Judges take ultimate responsibility for compulsory similar case retrieval， mastering the first-hand information about cases， and playing a decisive role from determination of the scope of similar cases to the final collective decision-making. Because similar case retrieval is carried out by a collegial panel or a single judge， the \"existence of disagreement in the application of laws\" should apply to circumstances where the collegial panel disagrees or the single judge identifies or considers the existence of disagreement. The formal requirement for compulsory initiation of similar case retrieval is that the case should be submitted to a professional judges' meeting for discussion. This generally means that a similar case retrieval report must be prepared for the case to be discussed. Therefore， judge assistants or trial teams should be authorized to carry out similar case retrieval because， when cases outnumber judges， judge assistants can help deal with the technical affairs and ensure that the judges will have adequate time and energy to handle the problems directly related to the adjudication of cases. Other members of the trial team can also ensure overall work efficiency by playing a similar role. It is already common practice to assign the task of similar case retrieval to judge assistants or other members of the trial team. For instance， Article 3 of the Rules of the High People's Court of Jiangsu Province on Establishing a Compulsory Similar Case Retrieval Report System （Trial） stipulates that \"a judge who handles a case may carry out similar case retrieval in person or assign the task to a judge assistant， but he/she shall be responsible for analysis and application of the retrieval result\". Based on the primary retrieval conducted by judge assistants or other members of the trial team， the judge handling the case may review， summarize， and simplify to provide effective options at the professional judges' meeting or during discussion of the review committee， thus improving the overall efficiency of the trial team and reducing the pressure on judges. Furthermore， the presiding judge should strengthen supervision and guidance over major cases; this is a reasonable option between respect for judicial laws and objective reality， as well as an important part of promoting the standardized implementation of similar cases with similar judgements from the perspective of internal supervision.

4.2 Constructing a Unified and Complete Judicial Database

According to Article 15 of Implementation Measures of the Supreme People's Court on the Uniform Application of Laws， the Supreme People's Court should establish a unified platform for the application of law and its database. The Office of Trial and Administration， the Research Office， the China Institute of Applied Jurisprudence， and the Information Technology Service Center of the People's Court are responsible for the planning， construction， research and development， operation and maintenance， and upgrading of the platform and its database according to their respective functions. Therefore， it is of great importance to construct a unified database.

The introduction of artificial intelligence into the guiding cases system with the objective of improving its actual effect requires support from a strong and complete judicial database， which shall not only collect a complete range of various guiding cases but also absorb a useful quantity of judicial judgments from local courts （this is mainly intended to facilitate judges' learning about judgments in similar cases at courts in different regions， especially judgments in cases where the same guiding case is adopted in such regions）. However， the relevant judicial database is not yet ideal and needs further construction， a task that is critical for optimization of the guiding cases system.

To overcome the defects of the present intelligent database， relevant departments should strengthen the construction of a unified and complete judicial database to include all typical， representative， instructive， and authoritative guiding cases and reference cases. In this way， the judgment rules and experience contained in guiding cases will be generally applicable to other cases and will satisfy the demand for depth and accuracy of stored information data， thus providing correct and accurate AI assistance to the judicature. To accomplish this， some have appealed for the \"construction of a guiding case database and cases retrieval auxiliary system. It's suggested that the Supreme People's Court should clear existing guiding case publishing platforms， promote construction of guiding case database， construct a nationwide unified and convenient guiding case retrieval technology platform and improve the construction of the legal database\"[20].

In the field of artificial intelligence， the important theoretical concept \"GARBAGE IN GARBAGE OUT\" （nonsense input data produce nonsense output）[21] indicates the critical nature of input data quality in an AI system. Whether full play will be given to the role of AI in its intervention into case guidance depends to some extent on the quality of the judicial database. Hence， the construction and improvement of the judicial database should focus on the four aspects listed below.

1. The guiding cases in the database should be constantly enriched to achieve a full range of coverage. A significant increase in the quantity of guiding cases is an urgent priority so that they will play a greater role in the similar case retrieval mechanism. The guiding cases should cover the main causes of action， crimes， and so on in the civil， criminal， and administrative fields; they should be regularly published to help cope with new， difficult， and complicated problems. Guiding case numbers should be increased by expanding their supply channels. To do this， the most effective and feasible way at present is to expand the creators of guiding cases. Specific improvement measures may also take advantage of the fact that local high people's courts quickly and accurately identify and cultivate candidate guiding cases， recognize recommendations of guiding cases as an important index for judges' performance appraisal， and re-sort and recognize more cases in the gazette as guiding cases. They also encourage law schools， law experts， representatives of the sessions-the Chinese People's Political Consultative Conference （CPPCC） and the National People's Congress （NPC） -and lawyers to recommend guiding cases. In view of the special status of the Supreme People's Court and its circuit courts， where time is appropriate， their future cases may be swiftly converted into guiding cases through a simplified procedure. This is an effective way to significantly increase guiding cases and is worth adopting to improve the guiding cases system in the future.

2. More judgment documents with improved quality should be available online. An effective online supervision mechanism may be established to safeguard the quality and transparency of judgment documents. This will promote \"similar judgment for similar cases\" and prevent undisclosed and nontransparent judicial judgment. Some local courts have already worked on this aspect and issued the necessary documents. For instance， Jiangsu Nantong Intermediate People's Court published its Implementation Scheme for Disclosure of Judgment Documents， under which all people's courts in Nantong must disclose judgment documents online except in four circumstances where such disclosure is inappropriate. The court also paid great attention to the quality of judgment documents disclosed online and publicized 46 defective judgment documents in its 2014 judgment document quality report[22].

3. The construction of the judicial database should be constantly refined. The quality of data in guiding cases and judgment documents should be guaranteed; there are some current problems with guiding cases and judgment documents， such as unclear reasons for judgment， inconsistency between key words， focus of disputes， and reasons for judgment that affect the quality of data in the database. Therefore， the accuracy of judicial data should be further improved. In addition， from the angle of actual operations， professional training should be strengthened to improve the quality of the database. The input and maintenance of data in judicial databases ultimately rely on operations conducted by judicial personnel. Accurate operations are fundamental to the construction of a high-quality database， and this places higher requirements on the professional ethics and business skills of judicial workers. Therefore， it is necessary to improve relevant judicial workers' awareness through professional training and learning， so they will rationally accept the inevitability and importance of intellectualization in judicial work and consciously endeavour to contribute to the construction of a unified， complete judicial database of high quality; professional training should also enable them to acquire new knowledge and skills about legal AI.

4. Hardware equipment should be provided for people's courts in underdeveloped regions to ensure that people's courts at all levels can submit judgment documents online and retrieve guiding cases. This will not only enrich the data of the judicial database， but also improve guiding case awareness and citation rates.

4.3 Improvement of Accuracy in Labeling Elements of Guiding Cases

In past judicial judgments in China， the approach was generally \"seeking application of law based on laws and regulations\"[23]. Today， however， under the guiding cases system， judicial judgment takes the approach of \"seeking application of a guiding case based on cases\"， enriching the means to seek application of law. When \"seeking a guiding case based on cases\"， comparisons and confirmations must be made concerning the similarities between cases. Therefore， against the background of the guiding cases system， sufficient endeavours must be made to label and identify guiding cases so that more judicial workers can use AI to discover and successfully apply them.

These problems make it necessary to carry out further technological optimization of the data labeling of guiding cases and demonstrate the characteristics of elements in cases， thus correctly and clearly reflecting the basic facts of a case and its relevant statutory rules.

The application of AI to case guidance requires the system to compare similarities between a pending case and guiding cases. However， the identification of such similarities requires comparison and analysis of cases' basic elements. Therefore， when compiling intelligent data， the elements of each case should be extracted and labeled accurately and clearly to enable the AI system to perform accurate and effective automatic identification of similarities and allow its users to retrieve desired cases rapidly and conveniently.

Some scholars have pointed out that， in the practice of case guidance， the labeling technique for base cases involves three templates for screening， selection， and application to similar cases： （1） the template for extracting elements of cases， including interpretation of judgment documents， integration of similar cases， and matching of \"fact-conclusions\"; （2） the template for interpreting key points of judgment， which requires linkage with legal norms， judgment norms， and specific facts; and （3） the template for application to similar cases， which integrates inductive reasoning of base cases， analogical reasoning of similar cases， and deductive reasoning of pending cases， then summarizes， extracts， and applies the rules of precedents. It can be seen that for accurate extraction and labeling of elements in the text of guiding cases， such as key words， key points of judgment， judgment， and reasons for judgment， the premise is to facilitate automatic identification by the intelligent system and legal reasoning by its users. The labeling of elements of cases should be like adding bookmarks that will provide clarity of directivity and convenience of identification.

Taking the key words among the basic elements as examples， since they are key elements of cases， the labeling should follow the principle of facilitating the identification and retrieval of cases. They should be concise and suggest the essence， the core， and the focus without linguistic repetition or long phrases or sentences. Extracted and generalized appropriately， neither too concrete nor too abstract， they should also be accurate in language to avoid ambiguity， have reasonable logic， and be appropriately ordered. They should reflect the core of disputes in a case so that other legal workers retrieving similar cases can match them from the judicial database through identification of key words. Keywords should neither be so few that they fail to reflect the focus of the disputes in the case nor so many that they replace the key points of judgment.

Taking the basic facts of cases for another instance， the factual aspect of guiding cases provides legal fact support for judgment reasoning and results. The narration of basic facts should be correct and brief， with clear clues， distinct gradation， conciseness， highlights， and appropriate details and omissions. It should neither omit facts related to key points of judgment nor appear trivial by copying the facts in the judgment document.

For correct application of the Notice of the Supreme People's Court on Issuing the Provisions on Case Guidance and to standardize and unify the style of guiding cases， the Research Office of the Supreme People's Court issued normative documents， such as Issuing the Opinions on Style for Compiling and Submitting Guiding Cases and the Format for Guiding Cases， which undoubtedly offer practical guidance on improving the accurate labeling of cases in an intelligent data system. Nonetheless， the author believes that many of the technical provisions in such normative documents are ambiguous and rough; they should be refined and improved to accommodate actual needs in judicial practice. In addition， technical specifications applicable to AI should be formulated and implemented based on the development and application of legal AI. Legal language is well known for its strong logic and irreplaceability. In particular， the connotations of abstract legal terminologies， such as \"liabilities\"， \"obligations\"， and \"rights\"， are abstract， uncertain， and extremely flexible. When they are converted into computer language， their original meaning may be lost， resulting in deviation of comprehension. The state judicial administration department therefore needs to publish relevant documents for regulating the conversion of legal language into computer language. Some countries have carried out such work， and their experience may be used for reference. Germany is one of these： \"In Germany， legal terminologies in guide specifications are divided into nine categories based on different legal functions， namely， 'liabilities'， 'compensation'， 'license'， 'prohibition'， 'objection'， 'additional condition'， 'legal consequence'， 'definition'， and 'reference'. Legal provisions and cases are classified according to the nine categories of legal terminologies. A modal verb is the label of a verb， while a verb indicates an action and plays an important role in defining semantic type. In German， modal verbs of 'dürfen/m?gen'（may）， 'k?nnen'（can）， 'müssen' （must）， 'sollen' （will） and 'wollen' （want） are considered to be included in this classification\"[24]. If specific provisions are made for the legal language to be applied in AI， it can be believed that the accuracy in labeling the elements of cases will be improved.

References：

[1] CAO Y Y. The balance between mobile phone forensics and privacy protection-from the perspective of principle of proportionality under modern public law[J]. Science Technology and Law（Chinese-English Version） ， 2019（6）： 68-76.

[2] GUO Y， SUN M. 2017 annual report on the judicial application of the guidance case from the Supreme People's Court[J]. China Journal of Applied Jurisprudence， 2018（3）： 108?133.

[3] Supreme People's Court's Work Report in 2023[EB/OL].（2023-03-17）[2023-09-21]lt; https：//www.court.gov.cn/zixun/xiangqing/393751.htmlgt;.

[4] DENG J. The comparing and empirical study on case guidance system[M]. Beijing： China Renmin University Press， 2015： 94.

[5] LEI L. Rethinking on the status of legal source concerning guiding case[J]. China Legal Science， 2015（1）： 272?290.

[6] GAO Y F. Design principle and implementation evaluation of similar case retrieval mechanism[J]. Presentday Law Science， 2021，19（01）： 1?11.

[7] SUN H B. In what sense does similar case retrieval help with the same case with same judgement？[J]. Tsinghua University Law Journal， 2021， 15（1）： 79?97.

[8] SUN Y. On the judgment rules of similar cases and their refining methods[J]. Hubei Social Sciences，2021（8）： 123?132.

[9] AN F D. Research on normalization and standardization of judgments in similar cases[J]. China Journal of Applied Jurisprudence， 2021（3）： 62?76.

[10] SUN G N. Operation of similar case retrieval and its improvement-taking the guiding opinions on uniform application of laws and strengthening similar case retrieval （trial） as analysis object[J]. Journal of Nantong University （Social Science Edition）， 2022， 38（1）： 90?100.

[11] ZHU F Y. A study on improving the mechanism of \"similar case with same judgement\" in the perspective of intelligent administration of justice[J]. Applied Law Review， 2021（1）： 3?20.

[12] ZHOU S H. Differentiated judgment in criminal cases and its reasonability[J]. China Legal Science， 2019（4）： 145?164.

[13] CHEN H L. A discussion on core disputes in intellectual property rights guiding case system with Chinese characteristics[J]. Intellectual Property， 2018（8）： 15?23.

[14] SONG X. Generation of precedents and Chinese guiding case system[J]. Chinese Journal of Law， 2011， 33（4）： 58?73.

[15] YANG J J. China's practice of disclosing judgment documents online： progress， problems and improvement[J]. China Law Review， 2019（6）： 125?147.

[16] PKULAW database[DB/OL]. （2023-09-01）[2023-09-06]. http：//www.pkulaw.cn/.

[17] CHEN X L. Chinese system of directive cases[M]. Beijing： Peking University Press， 2014： 195.

[18] CAO Y Y. Opportunities， challenges and solutions of judicial practice in the era of artificial intelligence[J]. Nomocracy Forum， 2019（3）： 278?290.

[19] HUANG W Z. Business appraisal system and rule of law in criminals[J]. Social Science Research， 2006（2）： 85?94.

[20] ZHANG H. Optimization of generation of guiding cases[J]. Legal Method， 2018， 25（3）： 220?241.

[21] DARRYL S. Garbage in： garbage out. （from the editor）[J]. Quality， 2018， 57（5）： 6.

[22] MA C. Wrong Chinese characters frequently witnessed in court's judgments： Nantong intermediate people's court gives criticism on 46 defective judgments[N]. Legal Daily， 2018-10-14（5）.

[23] ZHAO X Z. Legal discovery： a new interpretation of the rules of \"finding law\" by judges[J]. Journal of Shanghai University of Political Science and Law（The Rule of Law Forum） ， 2019， 34（04）： 52?60.

[24] WALTL B， BONCZEK G. Semantic types of legal norms in German laws： classification and analysis using local linear explanations[J]. Artificial Intelligence Law， 2019， 27（8）： 1?29.

人工智能類案檢索機制的發(fā)展與完善

曹奕陽

（華中師范大學(xué) 法學(xué)院，武漢 430079）

摘 " "要：在大數(shù)據(jù)時代，以人工智能技術(shù)服務(wù)于類案檢索，有助于提升案件的查全率和查準率，促進類案類判。人工智能類案檢索機制的發(fā)展經(jīng)歷了三個階段，即萌芽期、試行期與發(fā)展期，逐步成為推動審判流程標準化的一種長效機制，促進法官作出正確裁判。但在目前司法實踐中，存在著諸如檢索缺乏剛性約束機制與激勵機制，數(shù)據(jù)庫案例總量偏低、種類過少、數(shù)據(jù)來源較單一，智能系統(tǒng)的案例標識技術(shù)尚欠專業(yè)等問題，影響著類案檢索制度的實際運行效果。因此，人工智能類案檢索機制還有待于進一步完善：應(yīng)實現(xiàn)從問責為主到激勵為先的轉(zhuǎn)變，在賦予法官強制檢索義務(wù)的同時，探索和建立科學(xué)合理的激勵機制，鼓勵法官在司法實踐中積極開展類案檢索工作；應(yīng)建設(shè)統(tǒng)一完備的司法數(shù)據(jù)庫，將具有典型性、指導(dǎo)性、權(quán)威性的指導(dǎo)性案例和參考案例收入司法數(shù)據(jù)庫；應(yīng)提升案例要素標識的精準度，凸現(xiàn)案件的要素特征，從而推進類案檢索智能化、精準化。

關(guān)鍵詞：人工智能；指導(dǎo)性案例；類案檢索；司法數(shù)據(jù)庫