A legal case intelligent matching and recommendation device

Abstract

This invention relates to the fields of artificial intelligence and robotics, and discloses a legal case intelligent matching and recommendation device. The device involves deconstructing judgment documents into a dynamic legal argument causal graph containing node position attributes and edge argument validity scores, and achieving strategic matching by calculating the path similarity between the query intent and the case graph. The system also provides a counterfactual path inference function, performing attribution analysis on losing cases and generating strategy optimization suggestions. The invention further includes an autonomous mobile device equipped with this system. This device, through an obstacle perception and decision-making module, can autonomously avoid large obstacles and perform physical obstacle removal on small obstacles. This invention solves the problem of insufficient relevance in traditional keyword retrieval strategies and significantly improves the operational reliability of physical terminals in complex environments.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and robotics, specifically to a legal case intelligent matching and recommendation device. Background Technology

[0002] In the legal services field, the ability to quickly and accurately retrieve similar cases with reference value from a vast amount of court documents is fundamental for legal professionals to conduct case research and formulate litigation strategies. Existing legal case retrieval technologies typically rely on methods such as keyword matching and natural language processing, starting from surface-level text features to search for legal documents in databases relevant to the user's input query. Meanwhile, in certain service scenarios, it is necessary to deploy information retrieval systems on autonomous, mobile physical terminals to provide convenient on-site information services to users in environments such as office buildings and service halls. These mobile terminals are usually equipped with sensors and navigation systems to enable autonomous movement and obstacle avoidance in complex environments.

[0003] However, existing technologies still have shortcomings in practical applications. Current case retrieval systems mainly rely on the literal similarity of texts. When users' search needs involve specific argumentation ideas or strategies, such as wanting to understand how a particular type of evidence is used to successfully support a particular claim, search results based solely on keywords often do not match the user's intent in terms of argumentation logic, resulting in low recommendation accuracy. Users need to spend a lot of time manually screening the results.

[0004] Furthermore, existing technologies typically present case content in raw text form, failing to delve into the adversarial relationships between the parties' claims or quantify the strength of a judge's acceptance or rejection of a particular argument (e.g., the strength of evidence supporting a fact). Users must read the entire text, independently summarize and judge the logical connections and key factors behind the judgment, resulting in inefficient analysis and subject to subjective bias.

[0005] At the physical device level, existing autonomous mobile terminals also have relatively limited capabilities in environmental interaction. Their obstacle handling mechanisms are usually a simple obstacle avoidance and detour mode. When the device encounters small obstacles that can be easily removed, such as crumpled paper or small packaging boxes, it still performs complex path replanning and detour operations. This indiscriminate handling reduces the device's operating efficiency and its ability to continuously execute tasks in real physical environments. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a legal case intelligent matching and recommendation device, which solves the problems of low relevance of legal case retrieval strategies, difficulty in conducting in-depth quantitative analysis of case argumentation logic, and low operating efficiency and reliability of autonomous mobile devices in physical environments due to the single obstacle handling method.

[0007] To address the aforementioned technical problems, the first aspect of this invention provides a legal case intelligent matching and recommendation device, comprising: a base, a computer fixedly connected to the top of the base, the computer having a built-in legal case intelligent matching and recommendation system, an operating screen electrically connected to one side of the computer, a camera electrically connected to the top of the computer, multiple casters rotatably connected to the bottom of the base, a sensor electrically connected to one side of the base, a rotating shaft rotatably connected inside the base, and a clearing component provided on the outer wall of the rotating shaft.

[0008] The obstacle clearing assembly includes a support shell, an extension rod, and two soft pads. One end of the support shell is fixedly connected to the outer wall of the rotating shaft. An electric push rod is fixedly connected inside the support shell, and the output end of the electric push rod is connected to the interior of the extension rod. The outer wall of the extension rod is slidably connected to the interior of the support shell, and a limit strip is fixedly connected to the outer wall of one end of the extension rod. Connecting strips are fixedly connected to both sides of the extension rod, and one side of each soft pad is fixedly connected to the outer wall of the connecting strip.

[0009] In one specific embodiment, the computer's built-in intelligent legal case matching and recommendation system includes:

[0010] The core argument element extraction module is used to extract multiple categories of core argument elements from the input judgment document text through a sequence labeling task.

[0011] The dynamic legal argument causal graph construction module is connected to the core argument element extraction module. Based on the extracted core argument elements and their logical relationships in the original text, it constructs a legal argument causal graph containing nodes and directed edges, where nodes correspond to core argument elements and edges correspond to the argumentative relationships between elements.

[0012] The Argument Validity and Stance Quantification module is connected to the Dynamic Legal Argument Causal Graph Construction module. It is used to label the nodes in the legal argument causal graph with stance attributes, calculate the argument validity score of the edges connecting the nodes, and finally generate a dynamic legal argument causal graph with quantified attributes.

[0013] The case matching and recommendation module is connected to the argument validity and position quantification module. It receives a structured query graph paradigm and, in a dynamic legal argument causal graph database, matches and outputs legal cases that conform to the query graph paradigm by calculating path similarity.

[0014] In one specific embodiment, the intelligent legal case matching and recommendation system further includes:

[0015] An obstacle perception and decision-making module communicates with the sensors. Its function is to receive and process obstacle data collected by the sensors to determine the size of the obstacles. When the obstacle size is determined to be larger than a preset size threshold, the obstacle perception and decision-making module generates an obstacle avoidance control command to drive the multiple omnidirectional wheels to perform an obstacle avoidance maneuver. When the obstacle size is determined to be less than the size threshold, the obstacle perception and decision-making module generates an obstacle removal control command to drive the obstacle removal component to perform an obstacle removal action.

[0016] Preferably, the core argument elements extracted by the core argument element extraction module specifically include: factual nodes, evidence nodes, points of contention nodes, legal provisions nodes, reasoning nodes, and judgment nodes.

[0017] Furthermore, the argument validity and position quantification module labels the position attributes of the nodes, including: plaintiff's claim, defendant's claim, and whether the court adopts or rejects it. This labeling is determined based on the structural region of the core argument elements in the original text of the judgment.

[0018] In one specific embodiment, the process by which the argument validity and position quantification module calculates the argument validity score of the edges connecting nodes is as follows:

[0019] This calculation is based on the positional attributes of the two nodes connected by an edge, combined with the linguistic features of the argument nodes directly related to the argumentative relationship represented by that edge. A specific calculation method can be expressed by the following formula:

[0020]

[0021] Wherein, ε(e ij _ is edge e ij Argument validity score; S)n i _ and S)n j _ represents the position attributes of the starting and ending nodes connected by the edge; B)S(n i ),S(n j )) is a function that returns a base score based on the position attributes of two nodes; T r It is the text of the logical nodes related to this argument; C k (T r ) is used to extract text T r The function to extract the k-th linguistic feature; w kis the preset weight corresponding to the k-th linguistic feature; K represents the total number of categories of linguistic features extracted from the argumentative text; tanh is an "activation function" or "compression function"; ∑ means to accumulate all terms from k=1 to k=K.

[0022] Preferably, the path similarity calculated by the case matching recommendation module is obtained by weighting the structural similarity of the candidate paths with their aggregation effectiveness. The aggregation effectiveness is calculated from the argument effectiveness scores of all edges on the candidate path. The query graph paradigm is a graph structure containing a preset argumentation strategy. Specifically, this graph structure includes: at least one query node with a predetermined node type; at least one query edge with a predetermined edge type for connecting the query node; and expected attributes associated with the query node or the query edge, the expected attributes including expected stance attributes or expected ranges of argument effectiveness scores.

[0023] In one implementation, the intelligent legal case matching and recommendation system further includes:

[0024] The counterfactual path inference module is connected to the argument validity and position quantification module. It receives a user-specified case, identifies the invalid paths in the dynamic legal argument causal graph of the case, and generates a counterfactual graph based on the invalid paths.

[0025] Furthermore, the specific processing steps of the counterfactual path inference module are as follows: First, based on whether the aggregation power of a path is lower than a preset power threshold, it is determined whether the path is a failed path. Then, using the context information of the identified failed path as an index, successful paths with similar contexts but ultimately leading to positive results are retrieved from the case database. Finally, the missing path is determined by calculating the difference between the nodes and edges in the successful path and the failed path.

[0026] Furthermore, the counterfactual path inference module also includes: constructing the counterfactual graph by performing a graph union operation on the dynamic legal argument causal graph of the specified case and the determined missing path, so as to present a reinforced argument path structure.

[0027] This invention provides a legal case intelligent matching and recommendation device. It has the following beneficial effects:

[0028] 1. This invention deconstructs judgment documents into a dynamic legal argumentation causal graph containing core argumentation elements such as facts, evidence, and points of contention, and matches them based on path similarity including node type, position attribute, and validity score. This eliminates the reliance on surface text and makes the recommended cases highly consistent with the user's strategic needs in terms of argumentation structure and logical thinking, rather than just overlapping in wording.

[0029] 2. By introducing a quantitative mechanism of position attributes and argument validity scores, this invention explicitly and calculably marks the judge's adoption or rejection opinions implicit in the judgment documents, as well as the adversarial results claimed by the plaintiff and defendant, on the nodes and edges of the graph. This allows users to intuitively identify key winning or losing points, providing users with objective evidence for in-depth and quantitative analysis of cases.

[0030] 3. This invention, by incorporating sensors and an obstacle perception and decision-making module, can proactively identify the size of obstacles ahead. For large, immovable obstacles, the system controls the omnidirectional wheels to autonomously navigate around them; while for small, movable obstacles, the system automatically activates the obstacle removal component to physically remove them. This tiered processing strategy ensures that the device will not be interrupted by common obstacles during task execution, enhancing its operational reliability and environmental adaptability as a physical terminal. Attached Figure Description

[0031] Figure 1 This is a perspective view of the present invention;

[0032] Figure 2 This is a schematic diagram of the internal structure of the base of the present invention;

[0033] Figure 3 for Figure 2 A magnified view of the structure at point A in the middle;

[0034] Figure 4 for Figure 2 A magnified schematic diagram of the structure at point B in the middle;

[0035] Figure 5 This is a functional module block diagram of the legal case intelligent matching and recommendation system of the present invention;

[0036] Figure 6 This is a schematic diagram illustrating the workflow of the device of the present invention, which enables autonomous movement and interaction with the environment.

[0037] The components include: 1. Base; 2. Computer; 3. Control panel; 4. Camera; 5. Casters; 6. Sensor; 7. Rotary shaft; 8. Support shell; 9. Electric push rod; 10. Extension rod; 11. Limiting strip; 12. Connecting strip; 13. Soft pad. Detailed Implementation

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Please see the appendix Figure 1 - Appendix Figure 4 This invention provides a legal case intelligent matching and recommendation device, including a base 1, a computer 2 fixedly connected to the top of the base 1, the computer 2 having a built-in legal case intelligent matching and recommendation system, an operation screen 3 electrically connected to one side of the computer 2, a camera 4 electrically connected to the top of the computer 2, multiple casters 5 rotatably connected to the bottom of the base 1, a sensor 6 electrically connected to one side of the base 1, a rotating shaft 7 rotatably connected inside the base 1, and a clearing component provided on the outer wall of the rotating shaft 7;

[0040] The obstacle clearing assembly includes a support shell 8, an extension rod 10, and two soft pads 13. One end of the support shell 8 is fixedly connected to the outer wall of the rotating shaft 7. An electric push rod 9 is fixedly connected inside the support shell 8. The output end of the electric push rod 9 is connected to the inside of the extension rod 10. The outer wall of the extension rod 10 is slidably connected to the inside of the support shell 8. A limit strip 11 is fixedly connected to the outer wall of one end of the extension rod 10. Connecting strips 12 are fixedly connected to both sides of the extension rod 10. One side of each soft pad 13 is fixedly connected to the outer wall of the connecting strip 12.

[0041] Specifically, the base 1 serves as the support platform for the entire device. Multiple casters 5 are rotatably connected to the bottom of the base 1. These casters 5 support the base 1 and the overall weight of the device, allowing the device to move freely on a flat surface.

[0042] A computer 2 is fixedly connected to the top of the base 1. This computer 2 is the data processing and control core of this device, and it has a built-in intelligent legal case matching and recommendation system to perform the various data processing tasks that will be detailed in subsequent chapters.

[0043] Electrically connected to computer 2 are an operation screen 3 located on one side of computer 2 and a camera 4 located on top of computer 2. The operation screen 3 serves as a human-computer interaction interface, used to display system output information and receive user touch commands. The camera 4 is used to capture external visual information.

[0044] To enable the device to perceive its environment during movement, a sensor 6 is electrically connected to one side of the base 1. This sensor 6 is used to detect obstacles in front of the device's path and send the detected data to the computer 2 for processing.

[0045] Inside the base 1, there is a rotating shaft 7. On the outer wall of the rotating shaft 7, there is a clearing component.

[0046] In one specific embodiment, the obstacle clearing assembly includes a support shell 8, an electric push rod 9, an extension rod 10, a limiting strip 11, two connecting strips 12, and two soft pads 13.

[0047] One end of the support shell 8 is fixedly connected to the outer wall of the rotating shaft 7 and rotates together with the rotating shaft 7. Inside the support shell 8, an electric push rod 9 is fixedly connected. The output end of the electric push rod 9 is connected to the inside of the extension rod 10.

[0048] The outer wall of the extension rod 10 is slidably connected to the inner wall of the support housing 8. This connection allows the output end of the electric actuator 9 to push the extension rod 10 to extend or retract along its axial direction when the electric actuator 9 is activated. A limiting strip 11 is fixedly connected to the outer wall of the end of the extension rod 10 away from the electric actuator 9. The limiting strip 11 is larger than the opening of the support housing 8 to prevent the extension rod 10 from dislodging from the support housing 8.

[0049] A connecting strip 12 is fixedly connected to both sides of the extension rod 10. One side of each pad 13 is fixedly connected to the outer wall of the corresponding connecting strip 12. The pad 13 is made of elastic material and is used to provide cushioning and apply thrust when in contact with obstacles.

[0050] With the above structure, the computer 2 can drive the electric push rod 9 to move by control commands, so that the extension rod 10 drives the soft pad 13 to extend to the predetermined working position; then, by driving the rotating shaft 7 to rotate, the support shell 8 and the entire obstacle clearing assembly rotate around the axis of the rotating shaft 7, thereby using the soft pad 13 to remove small obstacles from the travel path.

[0051] See attached document Figure 5 , attached Figure 5 This is a functional block diagram of a legal case intelligent matching and recommendation system according to an embodiment of the present invention. The legal case intelligent matching and recommendation system, installed inside computer 2, may include multiple interconnected functional modules for performing the conversion from original judgment documents to structured, quantifiable argument graphs, and based on this, realizing case matching and analysis.

[0052] In one specific embodiment, the intelligent legal case matching and recommendation system includes: a core argument element extraction module, a dynamic legal argument causal graph construction module, an argument validity and stance quantification module, a case matching and recommendation module, and a counterfactual path inference module. These modules communicate and transmit data through a pre-defined data interface.

[0053] The core argument element extraction module receives the original electronic text of the judgment document as input. First, it preprocesses the input text, including text cleaning and sentence segmentation. Then, it performs structured analysis, dividing the document into different functional areas based on paragraph headings or keywords such as "Plaintiff's Claim," "Defendant's Defense," "Findings of the Court," and "The Court's Opinion." Finally, using a sequence labeling model, it performs entity recognition on the text within the document, extracting six predefined core argument elements: factual nodes, evidentiary nodes, points of contention nodes, legal provision nodes, reasoning nodes, and judgment nodes, and outputting them in structured data format.

[0054] The Dynamic Legal Argument Causal Graph Construction Module receives discrete core argument elements output by the core argument element extraction module. The function of this module is to organize these elements into a directed acyclic graph.

[0055] Specifically, each core argument element is instantiated as a node in a graph. The edges between nodes, i.e., the argument relationships, are established through a set of pre-defined heuristic rules. These rules include, but are not limited to: analyzing syntactic dependencies between sentences, identifying specific phrases indicating argument relationships (e.g., "sufficient to conclude", "insufficient evidence to prove"), and determining the co-occurrence relationships of elements in the text. The generated graph structure is defined as G = (N, E), where N is the set of nodes; E is the set of directed edges; and G represents the data structure of the entire graph.

[0056] The Argument Validity and Stance Quantification module receives the initial graph structure generated by the Dynamic Legal Argument Causal Graph Construction module and enhances its attributes to generate a dynamic legal argument causal graph. This process includes two steps. The first step is node stance attribute labeling. Based on the text region corresponding to each node, the Argument Validity and Stance Quantification module assigns a stance attribute to each node. Stance attributes include: plaintiff's claim, defendant's claim, court adoption, or court rejection. For example, the element node extracted from the "Plaintiff's Claim" section has the stance attribute labeled as "Plaintiff's Claim".

[0057] The second step is to calculate the argument validity score for each edge. The Argument Validity and Stance Quantification module calculates a quantified validity score for each edge in the graph, which represents the strength of the argument it represents. This calculation process is defined by the following formula:

[0058]

[0059] Wherein, ε(e ij ) is edge e ij Argument validity score; S(n) i ) and S(n j) represent the position attributes of the starting and ending nodes connected by the edge, respectively; B(S(n) i ),S(n j )) is a function that returns a base score based on the position attributes of two nodes; T r It is the text of the logical nodes related to this argument; C k (T r ) is used to extract text T r The function to extract the k-th linguistic feature; w k is the preset weight corresponding to the k-th linguistic feature; K represents the total number of categories of linguistic features extracted from the argumentative text; tanh is an "activation function" or "compression function"; ∑ means to accumulate all terms from k=1 to k=K.

[0060] The case matching and recommendation module receives a query graph paradigm and performs a matching operation in the generated dynamic legal argument causal graph database. The query graph paradigm is a structured graph query generated by the user through input or selection on the operation screen 3. Its content includes: at least one query node of a specified type, at least one query edge of a specified type, and expected attributes attached to the node or edge, such as the expected position attribute being "court acceptance," or the expected argument validity score being greater than 0.5.

[0061] The case matching and recommendation module calculates the path similarity between the query graph paradigm and candidate paths in the case graph in the database. This similarity is calculated by weighting the structural similarity and aggregation power of the candidate paths. Aggregation power is calculated by multiplying or weighting the argumentative power scores of all edges on the candidate path. Finally, the case matching and recommendation module outputs a list of legal cases sorted by path similarity.

[0062] The intelligent legal case matching and recommendation system also includes a counterfactual path inference module. This module receives a specific case specified by the user and its corresponding dynamic legal argument causal graph. First, it calculates the aggregation power of each path in the graph and identifies paths with aggregation power below a preset threshold as invalid paths. Then, using the context of the invalid path (i.e., the path's start node, end node, and neighboring node information) as an index, the module retrieves successful paths with similar contexts but aggregation power above the threshold from the case database.

[0063] By calculating the difference between the node set and edge set between successful and failed paths, a set of missing paths is identified. Finally, the counterfactual path inference module constructs and outputs a counterfactual graph by performing a union operation on the original dynamic legal argument causal graph and the execution graph of the missing paths.

[0064] See attached document Figure 6 , attached Figure 6 This is a schematic diagram illustrating the workflow of an autonomous movement and environmental interaction of a device according to an embodiment of the present invention. The autonomous movement and environmental interaction function of the intelligent legal case matching and recommendation device in this embodiment is controlled by an obstacle perception and decision-making module located within computer 2.

[0065] As the device moves using multiple omnidirectional wheels 5, sensors 6 continuously scan the area ahead of the travel path and send the collected raw obstacle data to the obstacle perception and decision module in real time. The raw obstacle data is a set of data points representing the presence, distance, and outline of obstacles.

[0066] The obstacle perception and decision-making module receives the raw obstacle data and processes it to calculate the obstacle's geometric dimensions, such as width and height. This calculation is accomplished through cluster analysis and bounding box fitting of the data point set.

[0067] The obstacle perception and decision-making module then compares the calculated obstacle size with a preset size threshold stored in the system. This threshold defines the upper limit of obstacle size that can be processed by the obstacle clearance component.

[0068] When an obstacle is determined to be larger than a preset size threshold, the obstacle perception and decision-making module generates an obstacle avoidance control command. This command contains local path planning data for bypassing the obstacle. The obstacle perception and decision-making module converts this path planning data into specific motion control signals for the multiple omnidirectional wheels 5. The drive unit then bypasses the obstacle along the newly planned path and resumes the original route.

[0069] When the obstacle size is determined to be no larger than a preset size threshold, the obstacle perception and decision-making module generates an obstacle clearance control command. This obstacle clearance control command is a sequence of action instructions executed in a time sequence.

[0070] Specifically, the command sequence first drives the electric push rod 9 within the obstacle removal assembly, causing its output end to push the extension rod 10 to extend to a predetermined working length. Next, the command sequence drives the rotating shaft 7 to rotate at a predetermined angle and speed, causing the support shell 8 and soft pad 13 fixed thereon to laterally displace the obstacle. After the obstacle is cleared, the command sequence then sequentially drives the rotating shaft 7 to rotate in the opposite direction to reset, and drives the electric push rod 9 to retract the extension rod 10 to its initial position, completing one obstacle removal operation.

[0071] In a specific work scenario, a user initiates a legal case search request to the device via the operating screen 3. This request is a query intent containing a specific argumentation strategy, such as searching for cases where "the plaintiff successfully proved the existence of a contractual relationship by using WeChat chat records as evidence and ultimately won the case."

[0072] The built-in case matching and recommendation module of Computer 2 receives the query intent and parses it into a structured query graph paradigm. This paradigm defines the expected node types (evidence nodes, fact nodes, judgment nodes), node attributes (evidence node content is related to "WeChat chat history", fact node content is related to "contractual relationship established"), position attributes (evidence node position is "plaintiff's claim", fact node position is "court adoption", judgment node is "victory"), and the attributes of the connecting edges (the argumentative power score from evidence to fact is positive).

[0073] The case matching and recommendation module then executes a graph matching algorithm in a pre-built dynamic legal argument causal graph database generated from massive amounts of cases. By calculating the path similarity between the query graph paradigm and the candidate paths in the database, cases that match the query intent in terms of argument structure, adversarial relationship, and argumentative effectiveness are selected.

[0074] After the matching results are sorted by similarity, they are presented to the user on operation screen 3. The user can select one case for in-depth analysis, such as a case with a similar argument structure but ultimately lost. At this time, the user can activate the counterfactual path inference function. The counterfactual path inference module is invoked, which first identifies the invalid path in the lost case (for example, the path where WeChat chat record evidence was not accepted by the court), then determines the missing key argument by performing difference calculation with the successful path, and finally generates and displays a counterfactual diagram containing the reinforcing path on operation screen 3.

[0075] While the aforementioned information processing and interaction are underway, or as the device moves to a designated location according to instructions, its environmental interaction function continues to operate. Sensor 6 continuously detects the travel path. When an obstacle smaller than a preset threshold is detected, the obstacle perception decision module generates and executes a sequence of obstacle removal instructions, driving the obstacle removal component to remove the obstacle. When an obstacle larger than the preset threshold is detected, the obstacle perception decision module generates and executes an obstacle avoidance instruction, controlling multiple omnidirectional wheels 5 to replan and execute an alternative path.

[0076] Through the collaborative work of the aforementioned hardware components and software modules, the device in this embodiment fully realizes a complete closed-loop operation, from receiving complex policy queries, executing deep matching recommendations, performing counterfactual attribution analysis, to autonomously moving in physical space and handling obstacles.

Claims

1. A legal case intelligent matching and recommendation device, characterized in that, include: A base (1) is fixedly connected to a computer (2) at the top of the base (1). The computer (2) has a built-in intelligent matching and recommendation system for legal cases. An operating screen (3) is electrically connected to one side of the computer (2). A camera (4) is electrically connected to the top of the computer (2). Multiple casters (5) are rotatably connected to the bottom of the base (1). A sensor (6) is electrically connected to one side of the base (1). A rotating shaft (7) is rotatably connected inside the base (1). A clearing component is provided on the outer wall of the rotating shaft (7). The obstacle clearing assembly includes a support shell (8), an extension rod (10), and two soft pads (13). One end of the support shell (8) is fixedly connected to the outer wall of the rotating shaft (7). An electric push rod (9) is fixedly connected inside the support shell (8). The output end of the electric push rod (9) is connected to the inside of the extension rod (10). The outer wall of the extension rod (10) is slidably connected to the inside of the support shell (8). A limit strip (11) is fixedly connected to the outer wall of one end of the extension rod (10). Connecting strips (12) are fixedly connected to both sides of the extension rod (10). One side of each soft pad (13) is fixedly connected to the outer wall of the connecting strip (12).

2. The intelligent matching and recommendation device for legal cases according to claim 1, characterized in that, The aforementioned intelligent legal case matching and recommendation system includes: The core argument element extraction module is used to extract multiple categories of core argument elements from judicial documents. A dynamic legal argument causal graph construction module is connected to the core argument element extraction module and is used to construct a legal argument causal graph containing nodes and edges based on the core argument elements and logical relationships. The argument validity and position quantification module is connected to the dynamic legal argument causal graph construction module. It is used to label the nodes in the legal argument causal graph with position attributes and calculate the argument validity score of the edges connecting the nodes, thereby generating the dynamic legal argument causal graph. The case matching and recommendation module, connected to the argument validity and position quantification module, is used to receive query graph paradigms and match and recommend legal cases based on the path similarity in the dynamic legal argument causal graph.

3. The intelligent matching and recommendation device for legal cases according to claim 1, characterized in that, The intelligent matching and recommendation system for legal cases also includes: The obstacle perception and decision-making module, which is communicatively connected to the sensor (6), is used for: Receive and process obstacle data collected by the sensor (6) to determine the size of the obstacle; When the size of an obstacle is determined to be greater than a preset threshold, an obstacle avoidance control command is generated to drive the multiple omnidirectional wheels (5) to go around the obstacle; When it is determined that the size of the obstacle is not greater than the preset threshold, a clearing control command is generated to drive the clearing component to remove the obstacle.

4. The intelligent matching and recommendation device for legal cases according to claim 2, characterized in that, The core elements of the argument include: factual points, evidentiary points, points of contention, legal provisions, logical arguments, and judgment points.

5. The intelligent matching and recommendation device for legal cases according to claim 2, characterized in that, The position attributes include the plaintiff's claim, the defendant's claim, and whether the court adopts or rejects it.

6. The intelligent matching and recommendation device for legal cases according to claim 2, characterized in that, The calculation of the argument validity score of the edge connecting the node is specifically as follows: Based on the position attributes of the two nodes connected by an edge, and combined with the linguistic features of the argument nodes related to the argumentation relationship of the position attributes, the argumentation effectiveness score of the edge connecting the nodes is calculated.

7. The intelligent matching and recommendation device for legal cases according to claim 2, characterized in that, The path similarity calculated by the case matching recommendation module is obtained by weighting the structural similarity and aggregation power of the candidate paths. The aggregation power is calculated from the argument power score of all edges on the candidate paths. The query graph paradigm is a graph structure containing a preset argumentation strategy, and the graph structure includes: At least one query node with a predefined node type; At least one query edge of a predetermined edge type is used to connect the query node; The expected attributes associated with the query node or the query edge, including expected position attributes or expected range of argument validity scores.

8. The intelligent matching and recommendation device for legal cases according to claim 2, characterized in that, The intelligent matching and recommendation system for legal cases also includes: The counterfactual path inference module, connected to the argument validity and position quantification module, is used to identify the failure paths in the dynamic legal argument causal graph of a specified case, and generate a counterfactual graph based on the failure paths.

9. A legal case intelligent matching and recommendation device according to claim 8, characterized in that, The counterfactual path inference module is specifically as follows: Whether a path is a failed path is determined based on whether the aggregation efficiency of a path is lower than a preset threshold. Using the context of the failed path as an index, retrieve successful paths with similar contexts from the case database; The missing path is determined by calculating the difference between the successful path and the failed path.

10. A legal case intelligent matching and recommendation device according to claim 9, characterized in that, The counterfactual path inference module also includes: The counterfactual graph is constructed by performing a graph union operation on the dynamic legal argument causal graph of the specified case and the missing path.

Images