Legal aid resource intelligent allocation and scheduling method, device and equipment and medium

Abstract

The application relates to a legal aid resource intelligent allocation and scheduling method, device, equipment and medium. The method comprises the following steps: acquiring consultation data; the consultation data comprises original consultation text, voice recording and video file; legal entities are extracted based on the consultation data through a multi-task learning model to obtain three-dimensional classification labels; the three-dimensional classification labels comprise geographical location codes, case type codes and emergency level codes; the three-dimensional classification labels and lawyer characteristics are matched through a deep forest model to obtain a matching probability matrix; the lawyer load is monitored based on the current state of the lawyer to obtain a resource state heat map; and a matching scheme set is obtained through a reinforcement learning model based on the matching probability matrix and the resource state heat map; the matching scheme set comprises a legal aid scheme. The method can provide real-time feedback on the state of lawyers and increase the flexibility of recommended legal aid schemes.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of resource scheduling, and particularly relates to a legal aid resource intelligent allocation and scheduling method, device, equipment and medium. BACKGROUND

[0002] With the increasing demand of the public for legal aid, a traditional legal aid resource allocation method appears. The traditional legal aid resource allocation method is usually realized based on a structured database and a rule engine. The system establishes a static mapping rule by manually labeling the lawyer practice field and case type keywords. When matching, the same field lawyers are first screened by keyword matching, and then filtered by geographical location according to administrative division codes. The distance between the lawyer and the case is calculated by using a fixed distance threshold, and finally the lawyers are sorted and recommended according to the historical winning rate. There are problems of lack of real-time feedback, long update cycle and inflexible recommended scheme. SUMMARY

[0003] Therefore, it is necessary to provide a legal aid resource intelligent allocation and scheduling method capable of providing real-time lawyer real-time state and flexibly generating a recommended scheme.

[0004] In a first aspect, the application provides a legal aid resource intelligent allocation and scheduling method, comprising:

[0005] obtaining consultation data; the consultation data includes original consultation text, voice recording and video file;

[0006] extracting legal entities based on the consultation data by a multi-task learning model to obtain three-dimensional classification labels; the three-dimensional classification labels include geographical location code, case type code and urgency code;

[0007] matching the three-dimensional classification labels and lawyer features by a deep forest model to obtain a matching probability matrix;

[0008] monitoring the lawyer load based on the current state of the lawyer to obtain a resource state heat map;

[0009] obtaining a matching scheme set by a reinforcement learning model based on the matching probability matrix and the resource state heat map; the matching scheme set includes a legal aid scheme.

[0010] Further, the three-dimensional classification labels obtained by extracting legal entities based on the consultation data by the multi-task learning model include:

[0011] generating context embedding vectors by a BERT-base model based on the consultation data;

[0012] performing graph neural network calculation on the context embedding vectors based on a legal knowledge graph to obtain a legal entity list according to the following formula:

[0013] e ij = LeakyReLU(W g [h i ||h j ])

[0014]

[0015] P i = Softmax(W c h′ i + b c )

[0016] LegalEntityList = Node i arg max(P i ) e ValidClasses, max(P i ) ≥ τ

[0017] where LegalEntityList is a list of legal entities, ValidClasses is a set of predefined valid legal entity classes, τ is a confidence threshold, P i is the class probability distribution of node i, W c is the classification weight matrix, b c is the bias term of the classification layer, h i , h j are the initial feature vectors of the i-th and j-th legal entities, carrying semantic information of legal entities, W g is the weight matrix, LeakyReLU is the leaky rectified linear unit activation function, e ij is the attention coefficient of edge (i, j), β ij is the normalized attention weight, W v is the linear transformation matrix, N i is the neighbor set of node i, h′ i is the updated node feature vector.

[0018] Based on the context embedding vector, a case type probability distribution is obtained.

[0019] Based on the case type probability distribution and the legal entity list, a case type code is obtained.

[0020] The context embedding vector is input into the BiLSTM-CRF model to obtain a geographic location code.

[0021] The consultation data is calculated through a hybrid model to generate an urgency code; the hybrid model includes a Wav2Vec2.0 unit and an NLP unit.

[0022] The case type code, geographic location code, and urgency code are integrated to obtain a three-dimensional classification label.

[0023] Further, the lawyer features are obtained by the following method:

[0024] Based on the lawyer practice database, the professional field set, the practice time and the historical success rate of the lawyer are extracted;

[0025] Based on the professional field set, the professional field vector is obtained through the legal knowledge graph;

[0026] Based on the practice time and the historical success rate, the seniority vector is calculated through the following formula:

[0027]

[0028] q=[Y norm ,W′]

[0029] Wherein, q is the seniority vector, indicating the quantitative representation of the lawyer's seniority, Y is the practice time of the lawyer, Y norm is the normalized practice time, W is the historical success rate, N is the total number of historical cases, and W' is the adjusted success rate;

[0030] Based on the current state of the lawyer, the current latitude and longitude coordinates and service radius of the lawyer are extracted to obtain the spatial identifier;

[0031] The professional field vector, seniority vector and spatial identifier features are spliced to obtain the lawyer features.

[0032] Further, the three-dimensional classification label and the lawyer features are matched through the deep forest model to obtain the matching probability matrix, including:

[0033] Based on the geographical position code in the three-dimensional classification label and the spatial identifier in the lawyer features, the spatial matching degree of the consultation case and the lawyer position is obtained;

[0034] Based on the lawyer features and the case type code, the professional similarity between the consultation case and the lawyer field is calculated through the following formula:

[0035]

[0036] Wherein, S pro is the professional similarity; P l is the professional field vector of the lawyer, t c is the case type code, and γ is the professional weight adjustment coefficient, SpecialtyWeight is the professional field weight;

[0037] Based on the seniority vector in the lawyer features and the urgency code, the time limit baseline of the lawyer is calculated through the following formula:

[0038]

[0039] wherein, T base is the efficiency baseline, the smaller the efficiency baseline, the higher the efficiency of the lawyer handling the case, Y norm is the normalized years of practice, W' is the adjusted win rate, and EmergencyLevel is the emergency level code;

[0040] Based on the spatial matching degree, professional similarity and efficiency baseline, the following formula is used for cascade forest training to obtain a matching probability matrix:

[0041]

[0042] wherein, is an element in the matching probability matrix, w space is the spatial matching degree, S pro is the professional similarity, T base is the efficiency baseline, and α, β, γ are weight coefficients, P init is the initial matching probability matrix, F (i,j) is the feature vector, CascadeForest is the cascade forest model, C is the cascade forest configuration parameter, 1e-5 is the scientific notation, representing 1×10 -5 , the minimum value, to avoid the denominator being zero.

[0043] Further, based on the current state of the lawyer, the lawyer's load is monitored to obtain a resource state heat map, including:

[0044] Based on the current state of the lawyer, the real-time load is calculated by the following formula:

[0045]

[0046] wherein, LoadRate i is the real-time load, is the number of cases being handled by the lawyer i, is the maximum number of cases handled by the lawyer i;

[0047] Based on the current state of the lawyer, the remaining in-transit time of the lawyer is determined;

[0048] The remaining in-transit time is equivalent to the case load to obtain the in-transit load;

[0049] Based on the historical case handling time sequence, the future case volume is predicted by the Prophet prediction model to obtain the future load; the historical case handling time sequence is obtained from the lawyer practice database;

[0050] Based on the real-time load, in-transit load and future load, the resource health degree is calculated by the following formula:

[0051]

[0052] Among them, Health i is the resource health, the closer to 1, the more idle the lawyers are, and negative values ​​indicate overload; is the total predicted case volume in the future time period a, For real-time equivalent caseload, TravelLoad i is the equivalent caseload in transit, is the lawyer’s total processing capacity in time period a;

[0053] Generate a resource status heat map based on resource health and lawyers' current location.

[0054] Furthermore, based on the matching probability matrix and resource status heat map, a matching solution set is obtained through the reinforcement learning model, including:

[0055] The matching probability matrix and resource status heat map features are integrated to obtain the status matrix;

[0056] Based on the matching probability matrix and resource status heat map, the reward matrix is ​​calculated using the following formula:

[0057]

[0058] Among them, R ij is the element in the reward matrix, P ij is the element in the matching probability matrix, is the predicted processing time for lawyer i to handle case j, α, β, θ are weight coefficients, and HealthLevel i The resource health level of lawyer i, HealthLevel i -5 means that a penalty is triggered when the monitoring degree exceeds the threshold of 5, to prevent overloaded lawyers from being over-allocated;

[0059] Based on the state matrix and reward matrix, the Q value is updated using the following formula to obtain the optimized Q value matrix:

[0060]

[0061] Among them, Q new (s,a) is the updated Q value, s is the current state, a is the current action, indicating the match between lawyer i and case j, s′ is the next state, α is the learning rate, γ is the discount factor, and Q(s,a) is the current Q value;

[0062] The matching probability matrix is ​​optimized based on the optimized Q value matrix to obtain the optimized matching probability matrix;

[0063] Based on the optimized matching probability matrix, a matching solution set is determined.

[0064] Further, based on the matching probability matrix and the resource state heat map, after obtaining the matching scheme set through the reinforcement learning model, the method further includes:

[0065] receiving a case closing report;

[0066] comparing the closing report with the matching scheme set to obtain a deviation set;

[0067] based on the deviation set, comparing the legal knowledge graph to obtain an association rule set;

[0068] based on the association rule set, adjusting the weight to obtain a model correction instruction set; the model correction instruction set is used to update the reinforcement learning model.

[0069] In a second aspect, the present application further provides a legal aid resource intelligent allocation and scheduling device, comprising:

[0070] a consultation module configured to obtain consultation data; the consultation data includes original consultation text, voice recording and video file;

[0071] an encoding module configured to extract legal entities based on the consultation data through a multi-task learning model to obtain three-dimensional classification labels; the three-dimensional classification labels include geographic location encoding, case type encoding and emergency degree encoding;

[0072] a matching module configured to match the three-dimensional classification labels and lawyer features through a deep forest model to obtain a matching probability matrix;

[0073] a state module configured to monitor lawyer load based on current state of the lawyer to obtain a resource state heat map;

[0074] a decision module configured to obtain a matching scheme set based on the matching probability matrix and the resource state heat map through a reinforcement learning model; the matching scheme set includes a legal aid scheme.

[0075] In a third aspect, the present application further provides a computer device comprising a memory and a processor; the memory stores a computer program; when the processor executes the computer program, the steps of the method provided in the first aspect of the present application are realized.

[0076] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon; when the processor executes the computer program, the steps of the method provided in the first aspect of the present application are realized.

[0077] The legal aid resource intelligent allocation scheduling method, device, equipment and medium obtain consultation data; the consultation data includes original consultation text, voice recording and video file; legal entities are extracted based on the consultation data through a multi-task learning model to obtain three-dimensional classification labels; the three-dimensional classification labels include geographic location codes, case type codes and emergency level codes; the three-dimensional classification labels and lawyer characteristics are matched through a deep forest model to obtain a matching probability matrix; the resource state heat map is obtained by monitoring the lawyer load based on the current state of the lawyer; and the matching scheme set is obtained through a reinforcement learning model based on the matching probability matrix and the resource state heat map; the matching scheme set includes a legal aid scheme. Through the above steps, the lawyer state can be obtained, and the legal aid scheme can be flexibly given based on the case situation. BRIEF DESCRIPTION OF DRAWINGS

[0078] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0079] Figure 1 The legal aid resource intelligent allocation scheduling method flowchart of the present application;

[0080] Figure 2 The legal aid resource intelligent allocation scheduling device diagram of the present application. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0082] The legal aid resource intelligent allocation scheduling method provided by the embodiments of the present application can be applied in the application environment of legal aid resource intelligent recommendation.

[0083] In one embodiment, as shown in Figure 1 A legal aid resource intelligent allocation scheduling method is provided. In this embodiment, the method is applied to a processor for illustration. It should be understood that the method can also be applied to a server, and can also be applied to a system including a processor and a server, and can be realized through the interaction of the processor and the server. In this embodiment, the method includes the following steps:

[0084] Step 101, obtaining consultation data; the consultation data includes original consultation text, voice recording and video file.

[0085] Specifically, the consultation data refers to the original information carrier submitted by the user in the legal aid process, including conversation records in text form, conversation recordings in audio format, and video data containing body language and expressions. Through the front-end interface of the legal service platform, the user's consultation request is collected in real time, and the user uploads the text file, call recording and video consultation recording. A complete initial data pool is formed to provide multi-modal data support for subsequent analysis. For example, the anxious expression in the video can assist in judging the emergency level, and the dialect voice can assist in positioning the geographical area.

[0086] Step 102, based on the consultation data, the legal entity is extracted by the multi-task learning model, and the three-dimensional classification label is obtained; the three-dimensional classification label includes geographical location coding, case type coding and emergency level coding.

[0087] Specifically, the multi-task learning model refers to a neural network architecture that shares underlying features but outputs multiple prediction targets; the three-dimensional classification label is a structured label group. The model processes text, speech and video in parallel. Among them, the geographical location is mapped through the address entity recognition and positioning system, the case type is calculated through the legal text matching degree, and the emergency level is determined through the semantic sentiment intensity and biological features. Unstructured consultation information is converted into standardized labels that can be processed by machines.

[0088] Step 103, match the three-dimensional classification label and the lawyer's characteristics through the deep forest model to obtain the matching probability matrix.

[0089] Specifically, the deep forest (gcForest) is a multi-level ensemble learning framework that realizes deep feature interaction through cascaded random forest combinations; the lawyer's characteristics can include practice years, win rate, professional field matching degree, historical response speed, etc. Feature vector, project the three-dimensional classification label and the lawyer's characteristics into the embedding space, and construct feature cross through multi-granularity scanning. Each decision tree learns the association mode of the lawyer's characteristics and the case label, and finally outputs the matching probability matrix.

[0090] Step 104, monitor the lawyer's load based on the current state of the lawyer to obtain the resource state heat map.

[0091] Specifically, the lawyer's load is constructed by a load evaluation function based on the current number of cases handled, the estimated settlement period, the density of the schedule, etc.; the heat map is a two-dimensional spatiotemporal visualization matrix, with the horizontal axis representing the time slice and the vertical axis representing the lawyer ID. The color depth represents the work intensity. Through the interface of the law firm OA system, the lawyer's schedule data is obtained in real time, and the load is predicted by combining the historical case processing time. Calculate the available case time window of each lawyer within the next 72 hours, and generate a dynamically updated resource distribution map.

[0092] Step 105, based on the matching probability matrix and the resource state heat map, a matching scheme set is obtained through the reinforcement learning model; the matching scheme set contains a legal aid scheme.

[0093] Specifically, through the reinforcement learning method of Q-learning, based on the matching probability matrix and the resource state heat map, the state matrix is generated by fusing the two through the reinforcement learning model, the reward matrix is calculated, the Q value matrix is iteratively updated to optimize the strategy, the optimized matching probability matrix is output, and finally the matching scheme set is generated.

[0094] The legal aid resource intelligent allocation and scheduling method provided in this embodiment obtains consultation data; the consultation data includes original consultation text, voice recording and video file; legal entities are extracted based on the consultation data through a multi-task learning model to obtain three-dimensional classification labels; the three-dimensional classification labels include geographic location encoding, case type encoding and urgency encoding; the three-dimensional classification labels and lawyer features are matched through a deep forest model to obtain a matching probability matrix; the current state of the lawyer is monitored based on the current state of the lawyer to obtain a resource state heat map; based on the matching probability matrix and the resource state heat map, a matching scheme set is obtained through a reinforcement learning model; the matching scheme set contains a legal aid scheme.

[0095] In one of the embodiments, the legal entities are extracted based on the consultation data through a multi-task learning model to obtain three-dimensional classification labels, including:

[0096] Step 201, the consultation data is generated through a BERT-base model to generate context embedding vectors.

[0097] Specifically, the BERT-base model is a pre-trained natural language processing model based on the Transformer architecture, which generates semantic vectors through bidirectional context understanding. The context embedding vector is a text semantic vector containing lexical context information. The pre-trained BERT-base model is used to encode the consultation data, converting unstructured text into structured vectors, providing a semantic basis for subsequent legal entity extraction.

[0098] Step 202, based on the legal knowledge graph, the context embedding vectors are calculated through the following formula to obtain a legal entity list:

[0099] e ij =LeakyReLU(W g [h i ||h j ])

[0100]

[0101] P i = Softmax(W c h′ i + b c )

[0102] LegalEntityList = Node i |arg max(P i )∈ValidClasses,max(P i )≥τ

[0103] where LegalEntityList is the legal entity list, ValidClasses is a pre-defined set of valid legal entity classes, τ is a confidence threshold, P i is the class probability distribution of node i, W c is the classification weight matrix, b c is the bias term of the classification layer, h i , h j is the initial feature vector of the i,j-th legal entity, carrying semantic information of the legal entity, W g is the weight matrix, LeakyReLU is the leaky rectified linear unit activation function, e ij is the attention coefficient of edge (i,j), β ij is the normalized attention weight, W v is the linear transformation matrix, N i is the neighbor set of node i, h′ i is the updated node feature vector.

[0104] Specifically, the legal knowledge graph is a structured knowledge base containing legal entities and their relationships. The legal entity list is a collection of key legal elements extracted from legal consultation data, which can represent specific legal provisions, case types, legal subjects and other related legal concepts involved in the case. The graph neural network is a deep learning model for processing graph-structured data, which updates node features by aggregating neighbor node information.

[0105] Step 203, based on the context embedding vector, obtain the case type probability distribution.

[0106] Specifically, the case type probability distribution is a probability vector of the case belonging to a pre-defined type. The context embedding vector generated by BERT is input into a fully connected layer, and the probability distribution is output by the Softmax function.

[0107] Step 204, based on the case type probability distribution and the legal entity list, obtain the case type encoding.

[0108] Specifically, the case type code is a unique identifier of the case type, and the probability of the highest category is selected from the case type probability distribution and mapped to a predefined coding rule.

[0109] Step 205, input the context embedding vector into the BiLSTM-CRF model to obtain the geographic location code.

[0110] Specifically, BiLSTM-CRF is a bidirectional long short-term memory network (captures context dependence) + conditional random field, which is used for sequence labeling tasks. The geographic location code is a standardized administrative division code.

[0111] Step 206, generate the urgency code by calculating the consultation data through the hybrid model; the hybrid model includes a Wav2Vec 2.0 unit and an NLP unit.

[0112] Specifically, the Wav2Vec 2.0 unit is a voice pre-training model that extracts acoustic features from audio. The NLP unit is a text analysis model that extracts keywords from text. The urgency code is a hierarchical code that identifies the urgency of the case.

[0113] Step 207, integrate the case type code, geographic location code and urgency code to obtain a three-dimensional classification label.

[0114] Specifically, the structured label is generated by concatenating in a fixed format, and the core attributes of the case are uniformly represented.

[0115] This embodiment converts unstructured consultation data into a three-dimensional classification label through multi-modal data processing, knowledge graph enhancement and hybrid model decision-making, providing accurate and structured input for subsequent lawyer resource matching, and ultimately achieving efficient, fair and interpretable legal aid allocation.

[0116] In one of the embodiments, the lawyer features are obtained by the following method:

[0117] Step 301, extract the professional field set, practice time and historical win rate of the lawyer based on the lawyer practice database.

[0118] Specifically, the lawyer practice database is a structured database that stores basic information of lawyers, including professional fields, practice time, historical case records, etc. The professional field set is a list of legal fields that the lawyer is good at. The practice time is the time from when the lawyer obtains the qualification to the present. The historical win rate is the proportion of winning cases in the past cases handled by the lawyer.

[0119] Step 302, based on the professional field set, obtain the professional field vector through the legal knowledge graph.

[0120] Specifically, the legal knowledge graph is a knowledge network containing nodes in the legal field and their relationships. The professional field vector is an embedded representation of the lawyer's professional field in the graph. The vector representation of the node is generated using a graph embedding method.

[0121] In step 303, the seniority vector is calculated based on the years of practice and the historical win rate using the following formula:

[0122]

[0123] q = [Y norm ,W′]

[0124] where q is the seniority vector, representing a quantitative representation of the lawyer's seniority, Y is the years of practice of the lawyer, Y norm is the normalized years of practice, W is the historical win rate, N is the total number of historical cases, and W' is the adjusted win rate.

[0125] Specifically, the seniority vector is a two-dimensional vector that combines years of practice and win rate. Bayesian smoothing is applied to the win rate of small sample cases to avoid extreme values.

[0126] For example, not only the number of lawsuits a lawyer has handled, but also the change in win rate over the past three years, W' can increase the weight of the win rate in the past three years.

[0127] In step 304, based on the current state of the lawyer, the current latitude and longitude coordinates and service radius of the lawyer are extracted to obtain the spatial identifier.

[0128] Specifically, the current latitude and longitude coordinates are the real-time geographic location of the lawyer. The service radius is the geographic range within which the lawyer can accept cases. The spatial identifier is a code or vector representing the lawyer's geographic service capability. The location is encoded as a string through geographic hashing.

[0129] In step 305, the professional field vector, seniority vector and spatial identifier features are concatenated to obtain the lawyer's features.

[0130] Specifically, the lawyer's features are a comprehensive vector that integrates professional field, seniority and spatial information.

[0131] This embodiment quantifies the core attributes of lawyers, solves the multi-objective optimization problem in legal aid matching, and has objectivity, scalability and adaptability.

[0132] In one embodiment, the three-dimensional classification label and the lawyer's features are matched through a deep forest model to obtain a matching probability matrix, including:

[0133] In step 401, based on the geographic location code in the three-dimensional classification label and the spatial identifier in the lawyer's features, the spatial matching degree between the consulting case and the lawyer's location is obtained.

[0134] Specifically, the geographic location encodes the standardized code of the case occurrence place. The spatial identifier represents the longitude and latitude coordinates of the lawyer and the service radius. The spatial matching degree represents the matching degree of the case and the lawyer's geographic location.

[0135] For example, a Beijing lawyer going to Shanghai to handle a case will be given a discount (considering the legal differences between the two places).

[0136] Step 402, based on the lawyer's characteristics and case type coding, the professional similarity between the consulting case and the lawyer's field is calculated by the following formula:

[0137]

[0138] Where S is the professional similarity; P is the professional field vector of the lawyer, t is the case type coding, γ is the professional weight adjustment coefficient, and SpecialtyWeight is the professional field weight. pro l c

[0139] Specifically, the professional similarity is an index that quantifies the matching degree between the lawyer's professional ability and the legal needs of the case. The larger the value, the more matched the lawyer's professional ability and the case needs.

[0140] Step 403, based on the seniority vector in the lawyer's characteristics and the emergency level coding, the time efficiency baseline of the lawyer is calculated by the following formula:

[0141]

[0142] Where T is the time efficiency baseline, the smaller the time efficiency baseline, the higher the efficiency of the lawyer handling the case, Y is the normalized years of practice, W' is the adjusted win rate, and EmergencyLevel is the emergency level coding. base norm

[0143] Specifically, the longer the years of practice, the higher the efficiency, and the higher the win rate, the efficiency is positively enhanced. The emergency level contains three levels.

[0144] Step 404, based on the spatial matching degree, the professional similarity, and the time efficiency baseline, the cascade forest training is carried out by the following formula to obtain the matching probability matrix:

[0145]

[0146]

[0147] Where, w is the spatial matching degree, S is the professional similarity, and T is the time efficiency baseline. space pro ​​​​​​For professional similarity, T base For time-based baseline, a, b, g are weight coefficients, P init For initial matching probability matrix, F (i,j) For feature vector, CascadeForest is a cascade forest model, C is a cascade forest configuration parameter, 1e-5 is a scientific notation, indicating 1x10 -5 , minimum value, avoid denominator zero.

[0148] Specifically, CascadeForest is an ensemble model composed of multiple layers of cascaded random forests, and each layer uses the prediction results of the previous layer to enhance the features. The matching probability matrix Pij represents the matching probability of the ith case and the jth lawyer.

[0149] The embodiment converts the matching process of lawyers and cases into a quantifiable and interpretable optimization problem by fusing and modeling the three-dimensional features of space, profession, and time, and adapts to the diversified needs of legal services.

[0150] In one of the embodiments, the lawyer load is monitored based on the current state of the lawyer, and a resource state heat map is obtained, including:

[0151] Step 501, based on the current state of the lawyer, the real-time load is calculated by the following formula:

[0152]

[0153] Where LoadRate i is the real-time load, is the number of cases being handled by lawyer i, is the maximum number of cases handled by lawyer i.

[0154] Specifically, the real-time load is the ratio of the current workload of the lawyer to the maximum processing capacity, which quantifies the current workload saturation of the lawyer, and the value closer to 1 indicates closer to full load.

[0155] Step 502, based on the current state of the lawyer, determine the remaining in-transit time of the lawyer.

[0156] Specifically, the remaining in-transit time of the lawyer is the traffic time consumed by the lawyer for investigating cases outside.

[0157] Step 503, equivalent the remaining in-transit time to case load to obtain in-transit load.

[0158] For example, assuming that the average time consumed by a lawyer to handle a single case is 2 hours, and the remaining in-transit time is 8 hours, which is equivalent to 4 cases of load.

[0159] At step 504, the future case load is predicted based on the historical case processing time sequence through the Prophet prediction model, and the future load is obtained. The historical case processing time sequence is obtained from the lawyer practice database.

[0160] Specifically, the future load is the predicted case load that needs to be processed by the lawyer i in the future time period. The Prophet prediction model is a time series prediction model.

[0161] At step 505, based on the real-time load, the in-transit load and the future load, the resource health degree is calculated through the following formula:

[0162]

[0163] Wherein, Health i is the resource health degree, and the closer to 1 indicates that the lawyer is more idle, and the negative value indicates overload; is the total predicted case load in the future time period a, is the real-time equivalent case load, TravelLoad i is the in-transit equivalent case load, is the total processing capacity of the lawyer in the time period a.

[0164] Specifically, the resource health degree is the comprehensive load state of the lawyer i in the future time period a. The real-time, in-transit and future loads are comprehensively evaluated to assess the health status of the lawyer.

[0165] At step 506, the resource state heat map is generated based on the resource health degree and the current position of the lawyer.

[0166] Specifically, the resource state heat map is a visual map that encodes the health degree of the lawyer with color as the coordinate of geographical position. The distribution and load state of the regional lawyer resources are intuitively displayed to guide task allocation.

[0167] The embodiment can realize dynamic early warning and global optimization by real-time monitoring, future prediction, multi-dimensional load fusion, converting the lawyer state into resource health degree, and generating a geographical heat map.

[0168] In one of the embodiments, based on the matching probability matrix and the resource state heat map, a matching scheme set is obtained through a reinforcement learning model, including:

[0169] At step 601, the matching probability matrix and the resource state heat map features are fused to obtain a state matrix.

[0170] Specifically, the state matrix is a feature matrix that fuses the matching probability matrix and the resource state heat map.

[0171] At step 602, based on the matching probability matrix and the resource state heat map, the reward matrix is calculated through the following formula:

[0172]

[0173] wherein R ij is an element in the reward matrix, P ij is an element in the matching probability matrix, is the predicted processing time of case j by lawyer i, a, b, q are weight coefficients, HealthLevel i is the resource health level of lawyer i, HealthLevel i -5 means that when the monitoring degree exceeds the threshold 5, the punishment is triggered to prevent the overloading lawyer from being excessively allocated.

[0174] Specifically, the reward matrix is the immediate reward of assigning case i to lawyer j, which is used to guide the reinforcement learning policy optimization. The predicted processing time is the estimated time consumption of lawyer i processing case j. The matching quality, processing efficiency and resource health are balanced to prevent overloading allocation.

[0175] Step 603, based on the state matrix and the reward matrix, update the Q value by the following formula to obtain the optimized Q value matrix:

[0176]

[0177] wherein Q new (s, a) is the updated Q value, s is the current state, a is the current action, represents the matching of lawyer i and case j, s' is the next state, a is the learning rate, g is the discount factor, and Q(s, a) is the current Q value.

[0178] Specifically, the Q value matrix is the long-term expected reward of executing action a (assigning case i to lawyer j) in state s. The learning rate is used to control the Q value update step size. The discount factor weighs the current reward and future reward.

[0179] Step 604, based on the optimized Q value matrix, optimize the matching probability matrix to obtain the optimized matching probability matrix.

[0180] Specifically, the optimized matching probability matrix is the adjusted matching probability of cases and lawyers according to the Q value matrix. For each case i, the Q value of lawyer j is normalized to a probability distribution, and the Q value is converted to a probability using the Softmax function.

[0181] Step 605, based on the optimized matching probability matrix, determine the matching scheme set.

[0182] Specifically, the matching scheme set is the final allocation list of cases and lawyers, which represents the detailed allocation strategy for a specific case. For each case i, the lawyer j with the highest probability can be selected. Random sampling can also be performed according to the probability distribution.

[0183] The embodiment combines static matching probability and dynamic resource state through reinforcement learning to generate a sustainable optimization matching scheme set, and realizes an adaptive allocation and long-term maximum profit decision method.

[0184] In one of the embodiments, after obtaining the matching scheme set through the reinforcement learning model based on the matching probability matrix and the resource state heat map, the method further includes:

[0185] Step 701, receiving a case closing report.

[0186] Specifically, the case closing report is structured data generated after the case is handled, which can include actual processing time, successful result, change of lawyer load, customer feedback, etc.

[0187] Step 702, comparing the closing report with the matching scheme set to generate a deviation set.

[0188] Specifically, the deviation set is a data set recording the difference between the matching scheme and the actual result, and each deviation can contain case ID, deviation type and deviation level. The deviation can include time deviation, result deviation and load deviation.

[0189] Step 703, based on the deviation set, comparing the legal knowledge graph to generate an association rule set.

[0190] Specifically, the association rule is a logical rule describing the causal relationship between the deviation and the legal knowledge entity. The legal entity associated with the deviation case is extracted as a subgraph; the frequent item set is analyzed; it is verified whether the expertise mismatch leads to low efficiency; the rule is formalized to obtain the association rule.

[0191] Step 704, adjusting the weight based on the association rule set to generate a model correction instruction set; the model correction instruction set is used to update the reinforcement learning model.

[0192] Specifically, the model correction instruction set is an operation instruction for adjusting the parameters or structure of the reinforcement learning model, and the coefficients in the reward function are modified to balance multiple objectives.

[0193] The embodiment realizes precise parameter adjustment, continuous learning and reduces manual intervention through the data-driven feedback mechanism and the rule mining guided by the domain knowledge (legal graph) to convert the actual case result into model optimization instructions.

[0194] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or stages.

[0195] Based on the same inventive concept, the embodiments of the present application also provide a legal aid resource intelligent allocation and scheduling device for implementing the legal aid resource intelligent allocation and scheduling method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more legal aid resource intelligent allocation and scheduling device embodiments provided below can refer to the limitations of the legal aid resource intelligent allocation and scheduling method described above, which will not be repeated here.

[0196] In an exemplary embodiment, as shown in Figure 2 a legal aid resource intelligent allocation and scheduling device 800 is provided, comprising:

[0197] The consultation module 801 is configured to obtain consultation data, wherein the consultation data includes original consultation text, voice recording, and video files.

[0198] The encoding module 802 is configured to extract legal entities based on the consultation data through a multi-task learning model to obtain three-dimensional classification labels, wherein the three-dimensional classification labels include geographic location encoding, case type encoding, and urgency level encoding.

[0199] The matching module 803 is configured to match the three-dimensional classification labels and lawyer features through a deep forest model to obtain a matching probability matrix.

[0200] The state module 804 is configured to monitor lawyer load based on the current state of the lawyer to obtain a resource state heat map.

[0201] The decision module 805 is configured to obtain a matching scheme set through a reinforcement learning model based on the matching probability matrix and the resource state heat map, wherein the matching scheme set includes a legal aid scheme.

[0202] Further, the encoding module 802 is further configured to:

[0203] generate context embedding vectors through a BERT-base model based on the consultation data.

[0204] Based on the legal knowledge graph, the context embedding vector is calculated by the following formula to obtain a legal entity list:

[0205] e ij =LeakyReLU(W g [h i ||h j ])

[0206]

[0207] P i =Softmax(W c h′ i +b c )

[0208] LegalEntityList=Node i |arg max(P i )∈ValidClasses,max(P i )≥τ

[0209] where LegalEntityList is a legal entity list, ValidClasses is a set of predefined valid legal entity categories, τ is a confidence threshold, P i is the class probability distribution of node i, W c is the classification weight matrix, b c is the bias term of the classification layer, h i , h j is the initial feature vector of the i,jth legal entity, carrying semantic information of the legal entity, W g is the weight matrix, LeakyReLU is the leaky rectified linear unit activation function, e ij is the attention coefficient of edge (i,j), β ij is the normalized attention weight, W v is the linear transformation matrix, N i is the neighbor set of node i, h′ i is the updated node feature vector.

[0210] Based on the context embedding vector, a case type probability distribution is obtained.

[0211] Based on the case type probability distribution and the legal entity list, a case type code is obtained.

[0212] The context embedding vector is input into the BiLSTM-CRF model to obtain a geographic location code.

[0213] The consultation data is calculated by a mixed model to generate an emergency level code; the mixed model includes a Wav2Vec2.0 unit and an NLP unit;

[0214] The case type code, the geographic location code and the emergency level code are integrated to obtain a three-dimensional classification label.

[0215] Further, the lawyer features are obtained by the following method:

[0216] Based on the lawyer practice database, the professional field set, the practice time and the historical win rate of the lawyer are extracted;

[0217] Based on the professional field set, the professional field vector is obtained through the legal knowledge graph;

[0218] Based on the practice time and the historical win rate, the seniority vector is calculated by the following formula:

[0219]

[0220] q=[Y norm ,W′]

[0221] Wherein, q is the seniority vector, indicating the quantitative representation of the lawyer's seniority, Y is the practice time of the lawyer, Y norm is the normalized practice time, W is the historical win rate, N is the total number of historical cases, and W' is the adjusted win rate;

[0222] Based on the current state of the lawyer, the current latitude and longitude coordinates and the service radius of the lawyer are extracted to obtain a spatial identifier;

[0223] The professional field vector, the seniority vector and the spatial identifier feature are spliced to obtain the lawyer feature.

[0224] Further, the matching module 803 is also used for:

[0225] Based on the geographic location code in the three-dimensional classification label and the spatial identifier in the lawyer feature, the spatial matching degree of the consultation case and the lawyer position is obtained;

[0226] Based on the lawyer feature and the case type code, the professional similarity between the consultation case and the lawyer field is calculated by the following formula:

[0227]

[0228] Wherein, S pro is the professional similarity; P l is the professional field vector of the lawyer, t c is the case type code, and γ is the professional weight adjustment coefficient, and SpecialtyWeight is the professional field weight;

[0229] The time limit baseline of the lawyer is calculated based on the seniority vector in the lawyer feature and the emergency level coding by the following formula:

[0230]

[0231] wherein T base is the time limit baseline, the smaller the time limit baseline, the higher the efficiency of the lawyer in handling cases, Y norm is the normalized years of practice, W′ is the adjusted win rate, and EmergencyLevel is the emergency level coding;

[0232] Based on the spatial matching degree, professional similarity and time limit baseline, the matching probability matrix is obtained by cascade forest training through the following formula:

[0233]

[0234] wherein, is an element in the matching probability matrix, w space is the spatial matching degree, S pro is the professional similarity, T base is the time limit baseline, and α, β, γ are weight coefficients, P init is the initial matching probability matrix, F (i,j) is the feature vector, CascadeForest is the cascade forest model, C is the cascade forest configuration parameter, 1e-5 is the scientific notation, indicating 1×10 -5 , the minimum value, to avoid the denominator being zero.

[0235] Further, the state module 804 is also used for:

[0236] Based on the current state of the lawyer, the real-time load is calculated by the following formula:

[0237]

[0238] wherein LoadRate i is the real-time load, is the number of cases handled by the lawyer i, is the maximum number of cases handled by the lawyer i;

[0239] Based on the current state of the lawyer, the remaining in-transit time of the lawyer is determined;

[0240] The remaining in-transit time is equivalent to the case load to obtain the in-transit load;

[0241] Based on the historical case handling time sequence, the future case quantity is predicted by the Prophet prediction model to obtain the future load; the historical case handling time sequence is obtained from the lawyer practice database;

[0242] Based on real-time load, on-the-way load and future load, the resource health degree is calculated by the following formula:

[0243]

[0244] wherein Health i is the resource health degree, the closer to 1 indicates that the lawyer is more idle, and the negative value indicates overload; is the total predicted case volume in the future time period a, is the real-time equivalent case volume, TravelLoad i is the on-the-way equivalent case volume, is the total processing capacity of the lawyer in the time period a;

[0245] Based on the resource health degree and the current position of the lawyer, a resource state heat map is generated.

[0246] Further, the decision module 805 is also used for:

[0247] The matching probability matrix and the resource state heat map features are fused to obtain a state matrix;

[0248] Based on the matching probability matrix and the resource state heat map, the reward matrix is calculated by the following formula:

[0249]

[0250] wherein R ij is an element in the reward matrix, P ij is an element in the matching probability matrix, is the predicted processing time of the lawyer i for the case j, and a, b, and q are weight coefficients, HealthLevel i is the resource health degree of the lawyer i, HealthLevel i -5 indicates that the monitoring degree exceeds the threshold 5 to trigger the penalty to prevent the overloaded lawyer from being excessively allocated;

[0251] Based on the state matrix and the reward matrix, the Q value is updated by the following formula to obtain an optimized Q value matrix:

[0252]

[0253] wherein Q new (s, a) is the updated Q value, s is the current state, a is the current action, represents the matching of the lawyer i and the case j, s' is the next state, a is the learning rate, g is the discount factor, and Q(s, a) is the current Q value;

[0254] Based on the optimized Q value matrix, the matching probability matrix is optimized to obtain an optimized matching probability matrix;

[0255] Based on the optimized matching probability matrix, a matching scheme set is determined.

[0256] Further, the apparatus further comprises an updating module configured to:

[0257] receiving a case closing report;

[0258] comparing the closing report with the matching scheme set to generate a deviation set;

[0259] based on the deviation set, comparing the legal knowledge graph to generate a set of association rules;

[0260] based on the set of association rules, adjusting the weights to generate a model correction instruction set; the model correction instruction set is used to update the reinforcement learning model.

[0261] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the legal aid resource intelligent allocation and scheduling method as described above when executing the computer program.

[0262] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above method embodiments.

[0263] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part can be referred to the part of the method embodiment. The device embodiments described above are only schematic, and the components described as separate components can or can not be physically separate, and the components displayed as a unit can or can not be a physical unit, that is, they can be located in one place, or also distributed on multiple network units. According to the actual needs, some or all of the modules can be selected to achieve the purpose of the present disclosure. Those skilled in the art can understand and implement it without creative labor.

[0264] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A legal aid resource intelligent allocation scheduling method, characterized in that, The method comprises: obtaining consultation data; the consultation data comprises original consultation text, voice recording and video files; extracting legal entities based on the consultation data through a multi-task learning model to obtain a three-dimensional classification label; the three-dimensional classification label comprises a geographical location code, a case type code and an emergency level code; matching the three-dimensional classification label and lawyer features through a deep forest model to obtain a matching probability matrix; monitoring lawyer load based on the current state of the lawyer to obtain a resource state heat map; based on the matching probability matrix and the resource state heat map, a matching scheme set is obtained through a reinforcement learning model; the matching scheme set contains legal aid schemes; wherein the matching of the three-dimensional classification label and the lawyer features through the deep forest model to obtain the matching probability matrix comprises: based on the geographical location code in the three-dimensional classification label and the spatial identifier in the lawyer feature, the spatial matching degree of the consultation case and the lawyer location is obtained; based on the lawyer feature and the case type code, the professional similarity of the consultation case and the lawyer field is calculated through the following formula: Wherein, S pro is the professional similarity; P l is the professional field vector of the lawyer, t c is the case type code, γ is the professional weight adjustment coefficient, and SpecialtyWeight is the professional field weight. based on the seniority vector in the lawyer feature and the emergency level code, the time efficiency baseline of the lawyer is calculated through the following formula: Wherein, T base is the time baseline, the smaller the time baseline, the higher the efficiency of the lawyer handling the case, Y norm is the normalized years of practice, W' is the adjusted win rate, and EmergencyLevel is the emergency level code; based on the spatial matching degree, the professional similarity and the time efficiency baseline, cascade forest training is carried out through the following formula to obtain the matching probability matrix: wherein, w is the weight of the element in the probability matrix, space S is the spatial matching degree, pro T is the professional similarity, base α, β, γ are weight coefficients, and P is the time effectiveness baseline, init F is the initial matching probability matrix, (i,j) X is the feature vector, CascadeForest is the cascade forest model, C is the cascade forest configuration parameter, 1e-5 is the scientific notation, indicating 1 × 10 -5 , minimum value, to avoid the denominator being zero.

2. The method of claim 1, wherein, the extraction of legal entities based on the consultation data through the multi-task learning model to obtain the three-dimensional classification label comprises: generating context embedding vectors through the BERT-base model based on the consultation data; based on the legal knowledge graph, the context embedding vectors are calculated through the following formula to obtain a list of legal entities: e ij = LeakyReLU(W g [h i ‖h j ]) P i = Softmax(W c h' i + b c ) LegalEntityList = Node i | argmax(P i ) ∈ ValidClasses, max(P i ) ≥ τ where LegalEntityList is a list of legal entities, ValidClasses is a set of predefined valid legal entity classes, τ is a confidence threshold, P i is the class probability distribution of node i, W c is the classification weight matrix, b c is the bias term of the classification layer, h i , h j is the initial feature vector of the i,jth legal entity, carrying semantic information of the legal entity, W g is the weight matrix, LeakyReLU is a leaky rectified linear unit activation function, e ij is the attention coefficient of edge (i,j), β ij is the normalized attention weight, W v is the linear transformation matrix, N i is the neighbor set of node i, h' i is the updated node feature vector; based on the context embedding vectors, a case type probability distribution is obtained; based on the case type probability distribution and the list of legal entities, a case type code is obtained; inputting the context embedding vectors into the BiLSTM-CRF model to obtain a geographical location code; generating an emergency level code by calculating the consultation data through a hybrid model; the hybrid model comprises a Wav2Vec2.0 unit and an NLP unit; integrating the case type code, the geographical location code and the emergency level code to obtain a three-dimensional classification label.

3. The method of claim 1, wherein, The lawyer feature is obtained by the following method: based on the lawyer practice database, the professional field set, the practice period and the historical success rate of the lawyer are extracted; based on the professional field set, the professional field vector is obtained through the legal knowledge graph; based on the practice period and the historical success rate, the seniority vector is calculated through the following formula: q = [Y norm ,W'] wherein q is a vector of qualifications, representing a quantitative representation of the qualifications of the lawyer, Y is the number of years of practice of the lawyer, Y norm is the normalized number of years of practice, W is the historical win rate, N is the total number of historical cases, and W' is the adjusted win rate. based on the current state of the lawyer, the current latitude and longitude coordinates and the service radius of the lawyer are extracted to obtain a spatial identifier; the professional field vector, the seniority vector and the spatial identifier feature are spliced to obtain the lawyer feature.

4. The method of claim 1, wherein, The monitoring of lawyer load based on the current state of the lawyer to obtain a resource state heat map comprises: based on the current state of the lawyer, the real-time load is calculated through the following formula: LoadRate = LoadRate + (LoadRate - LoadRate) / 100 i LoadRate = LoadRate + (LoadRate - LoadRate) / 100 LoadRate = LoadRate + (LoadRate - LoadRate) / 100 LoadRate = LoadRate + (LoadRate - LoadRate) / 100 based on the current state of the lawyer, the remaining time of the lawyer is determined; equivalent to the case load, to obtain the in-transit load; based on the historical case processing time sequence, the future case quantity is predicted through the Prophet prediction model to obtain the future load; the historical case processing time sequence is obtained from the lawyer practice database; based on the real-time load, the in-transit load and the future load, the resource health degree is calculated through the following formula: where Health i is the resource health, with closer to 1 indicating the lawyer is more idle and negative values indicating overload; is the total predicted caseload for future time period a, is the real-time equivalent caseload, TravelLoad i is the en-route equivalent caseload, is the total processing capacity of the lawyer for time period a; based on the resource health degree and the current position of the lawyer, a resource state heat map is generated.

5. The method of claim 1, wherein, based on the matching probability matrix and the resource state heat map, a matching scheme set is obtained through a reinforcement learning model, including: the matching probability matrix and the resource state heat map characteristics are fused to obtain a state matrix; based on the matching probability matrix and the resource state heat map, the reward matrix is calculated through the following formula: where R ij is an element in the reward matrix, P ij is an element in the matching probability matrix, is the predicted processing time of case j by lawyer i, and α, β, θ are weight coefficients, HealthLevel i is the resource health level of lawyer i, HealthLevel i -5 means that the punishment is triggered when the monitoring degree exceeds the threshold value 5, preventing the overloading lawyer from being excessively allocated; based on the state matrix and the reward matrix, the Q value is updated through the following formula to obtain an optimized Q value matrix: wherein Q new (s, a) is the updated Q value, s is the current state, a is the current action, represents the matching of the lawyer i and the case j, s' is the next state, a is the learning rate, g is the discount factor, and Q(s, a) is the current Q value. based on the optimized Q value matrix, the matching probability matrix is optimized to obtain an optimized matching probability matrix; based on the optimized matching probability matrix, a matching scheme set is determined.

6. The method of claim 1 to 5, wherein, after the matching scheme set is obtained based on the matching probability matrix and the resource state heat map through the reinforcement learning model, it further includes: receiving a case closing report; comparing the closing report with the matching scheme set to generate a deviation set; based on the deviation set, comparing the legal knowledge graph to generate a set of association rules; based on the set of association rules, adjusting the weight to generate a model correction instruction set; the model correction instruction set is used to update the reinforcement learning model.

7. A legal aid resource intelligent allocation and scheduling device characterized in that, The device comprises: a source module for obtaining consultation data; the consultation data includes original consultation text, voice recording and video file; a classification module for extracting legal entities based on the consultation data through a multi-task learning model to obtain a three-dimensional classification label; the three-dimensional classification label includes geographic location code, case type code and urgency level code; a matching module for matching the three-dimensional classification label and lawyer features through a deep forest model to obtain a matching probability matrix; a load module for monitoring lawyer load based on the current state of the lawyer to obtain a resource state heat map; a decision module for obtaining a matching scheme set through a reinforcement learning model based on the matching probability matrix and the resource state heat map; the matching scheme set includes legal aid scheme; wherein the matching module is further configured to obtain the spatial matching degree of the consultation case and the lawyer position based on the geographic location code in the three-dimensional classification label and the spatial identifier in the lawyer features; the professional similarity of the consultation case and the lawyer field is calculated based on the lawyer features and the case type code through the following formula: wherein S pro is the professional similarity; P l is the professional field vector of the lawyer, t c is the case type code, γ is the professional weight adjustment coefficient, and SpecialtyWeight is the professional field weight; the timeliness baseline of the lawyer is calculated based on the seniority vector in the lawyer features and the urgency level code through the following formula: Wherein, T base is the time baseline, the smaller the time baseline, the higher the efficiency of the lawyer handling the case, Y norm is the normalized years of practice, W' is the adjusted win rate, and EmergencyLevel is the emergency level code. based on the spatial matching degree, the professional similarity and the timeliness baseline, the matching probability matrix is obtained through the following formula for cascade forest training: wherein, w is the weight of the element in the probability matrix, space S is the spatial matching degree, pro T is the professional similarity, base P is the time effectiveness baseline, and a, b, g are weight coefficients, init F is the initial matching probability matrix, (i,j) X is the feature vector, CascadeForest is the cascade forest model, C is the cascade forest configuration parameter, 1e-5 is the scientific notation, indicating 1 x 10 -5 , minimum value, to avoid the denominator being zero.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, the processor executes the computer program to realize the steps of the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which is executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images