Task splitting method, device, equipment and storage medium based on artificial intelligence

Abstract

The present application relates to the field of artificial intelligence technology, and discloses a task splitting method, device, equipment and storage medium based on artificial intelligence, wherein the method includes: case feature data includes: case information feature data and entrustment information feature data; using task extraction rules, generating tasks according to entrustment information feature data, and obtaining a first task list; inputting case information feature data into a preset task classification prediction model to perform task classification prediction, and obtaining a task classification prediction vector; determining tasks according to the task classification prediction vector, and obtaining a second task list; merging tasks according to the first task list and the second task list, and obtaining the target task splitting result. Accurate splitting of case tasks is achieved, which provides support for the subsequent determination of entrustment prices according to tasks, and provides support for determining the task underwriting end according to tasks. The present application can be applied to technical fields such as digital healthcare, smart government affairs, and science and technology finance.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based task splitting method, device, equipment and storage medium. Background Art

[0002] In the process of insurance companies' claims settlement, they need to entrust the cases to crowdsourcing companies, which assist insurance companies in claims settlement by investigating, collecting evidence, and drawing inferences and conclusions. Traditional cases are entrusted to the same crowdsourcing company, which has the disadvantage of using a fixed entrustment price for each case. Different cases require different numbers and types of specific tasks to be performed, which leads to technical problems such as the inability to adjust the entrustment price when the number of tasks in a case is small or the types are different. Summary of the invention

[0003] The main purpose of this application is to provide a task splitting method, device, equipment and storage medium based on artificial intelligence, aiming to solve the problem that in the prior art, when entrusting a case, the entire case is entrusted to the same crowdsourcing company, and a fixed entrustment price is used for each case. Because the number and type of specific tasks to be performed in different cases are different, the technical problem that the entrustment price cannot be adjusted when the number of tasks in the case is small or the type is different.

[0004] In order to achieve the above-mentioned invention object, the present application proposes a task splitting method based on artificial intelligence, the method comprising:

[0005] Acquiring case characteristic data, the case characteristic data including: case information characteristic data and entrustment information characteristic data;

[0006] Acquire a task extraction rule, adopt the task extraction rule, generate tasks according to the commission information feature data, and obtain a first task list;

[0007] Inputting the case information feature data into a preset task classification prediction model to perform task classification prediction, and obtaining a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on an XGboost model;

[0008] Determine tasks according to the task classification prediction vector to obtain a second task list;

[0009] Task merging processing is performed according to the first task list and the second task list to obtain a target task splitting result.

[0010] Furthermore, the step of obtaining case characteristic data includes:

[0011] Obtaining entrusted case data, wherein the entrusted case data includes: basic case information and entrusted information text;

[0012] Generate label features based on the basic information of the case to obtain the case information feature data;

[0013] According to the entrustment information text, task description statements and task features are generated respectively to obtain the entrustment information feature data.

[0014] Furthermore, the step of generating label features according to the basic information of the case to obtain the case information feature data includes:

[0015] Get the tag generation script;

[0016] Executing the label generation script to perform label prediction on the structured data in the basic information of the case to obtain a label prediction result;

[0017] A vector is generated according to the label prediction result to obtain the case information feature data.

[0018] Furthermore, the step of generating a task description statement and a task feature according to the entrustment information text to obtain the entrustment information feature data includes:

[0019] Get the statement segmentation regular expression;

[0020] Using the sentence segmentation regular expression, the entrustment information text is split into task description sentences to obtain a task description sentence set;

[0021] Obtaining a task feature extraction model, wherein the task feature extraction model is a model trained according to the Bert model;

[0022] The task feature extraction model is used to extract features from each of the task description statements in the task description statement set to obtain the entrustment information feature data.

[0023] Furthermore, the step of performing task merging processing according to the first task list and the second task list to obtain a target task splitting result includes:

[0024] Combine the tasks in the first task list and the second task list to obtain a task list to be deduplicated;

[0025] Performing task deduplication processing on the task list to be deduplicated to obtain a task list to be reviewed;

[0026] Sending the task list to be reviewed to the task splitting and reviewing end;

[0027] Obtaining the task review result corresponding to the task list to be reviewed sent by the task splitting review end;

[0028] When the audit result of the task audit result is passed, the task list to be audited is used as the target task splitting result;

[0029] When the audit result of the task audit result is not passed, the task list to be audited is corrected according to the task correction data of the task audit result to obtain the target task splitting result.

[0030] Furthermore, after the step of obtaining the task review result corresponding to the task list to be reviewed sent by the task splitting review terminal, it also includes:

[0031] When the task review result is not passed, a case is generated according to the entrusted case data corresponding to the case feature data and the target task splitting result to obtain a difficult case;

[0032] Updating the problem case to the problem case library;

[0033] The preset task classification prediction model is updated and trained according to the difficult case library.

[0034] Furthermore, after the step of performing task merging processing according to the first task list and the second task list to obtain the target task splitting result, the step further includes:

[0035] Get task allocation configuration data;

[0036] According to the task allocation configuration data, task matching is performed on the target task splitting result to obtain a hit task identification set and a miss task identification set;

[0037] Allocate each task in the hit task identifier set to a task underwriting end according to the task allocation configuration data;

[0038] A task underwriting end label library is obtained, and each task in the missed task identification set is allocated to the task underwriting end according to the case label of the case feature data and the task underwriting end label library.

[0039] The present application also proposes a task splitting device based on artificial intelligence, the device comprising:

[0040] A data acquisition module, used to acquire case characteristic data, wherein the case characteristic data includes: case information characteristic data and commission information characteristic data;

[0041] A first task list determination module is used to obtain a task extraction rule, adopt the task extraction rule, generate tasks according to the commission information feature data, and obtain a first task list;

[0042] A task classification prediction vector determination module is used to input the case information feature data into a preset task classification prediction model to perform task classification prediction and obtain a task classification prediction vector, wherein the preset task classification prediction model is a model obtained by training based on an XGboost model;

[0043] A second task list determination module, used to determine tasks according to the task classification prediction vector to obtain a second task list;

[0044] The target task splitting result determination module is used to perform task merging processing according to the first task list and the second task list to obtain the target task splitting result.

[0045] The present application also proposes a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0046] The present application also proposes a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0047] The present application discloses an artificial intelligence-based task splitting method, device, equipment and storage medium, wherein the method first generates tasks according to the commission information feature data by adopting the task extraction rule to obtain a first task list, and then inputs the case information feature data into a preset task classification prediction model to perform task classification prediction to obtain a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on the XGboost model; tasks are determined according to the task classification prediction vector to obtain a second task list, and finally tasks are merged according to the first task list and the second task list to obtain a target task splitting result, thereby achieving accurate splitting of case tasks, providing support for subsequent determination of commission prices according to tasks, which is beneficial to reducing case commission costs, and also providing support for determining the task underwriting end according to tasks, which is beneficial to improving the efficiency and quality of case execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a task splitting method based on artificial intelligence according to an embodiment of the present application;

[0049] Figure 2 This is a schematic block diagram of the structure of a task splitting device based on artificial intelligence according to an embodiment of the present application;

[0050] Figure 3 A schematic block diagram of the structure of a computer device according to an embodiment of the present application.

[0051] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] Reference Figure 1 In an embodiment of the present application, a task splitting method based on artificial intelligence is provided, and the method comprises:

[0054] S1: Acquire case characteristic data, wherein the case characteristic data includes: case information characteristic data and entrustment information characteristic data;

[0055] S2: Acquire a task extraction rule, adopt the task extraction rule, generate tasks according to the commission information feature data, and obtain a first task list;

[0056] S3: Inputting the case information feature data into a preset task classification prediction model to perform task classification prediction, and obtaining a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on an XGboost model;

[0057] S4: Determine tasks according to the task classification prediction vector to obtain a second task list;

[0058] S5: Perform task merging processing according to the first task list and the second task list to obtain a target task splitting result.

[0059] This embodiment adopts the task extraction rule to generate tasks according to the entrustment information feature data to obtain a first task list, and then inputs the case information feature data into a preset task classification prediction model to perform task classification prediction to obtain a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on the XGboost model; tasks are determined according to the task classification prediction vector to obtain a second task list, and finally tasks are merged according to the first task list and the second task list to obtain a target task splitting result, thereby achieving accurate splitting of case tasks, providing support for subsequent determination of entrustment prices according to tasks, which is beneficial to reducing case entrustment costs, and also providing support for determining the task underwriting end according to tasks, which is beneficial to improving the efficiency and quality of case execution.

[0060] For S1, the case feature data input by the user can be obtained, the case feature data can be obtained from the database, and the case feature data can also be obtained from a third-party application system.

[0061] The case corresponding to the case information characteristic data may be an insurance claim case or other cases, for example, cases in technical fields such as digital healthcare, smart government affairs, science and technology finance, etc., which are not limited here.

[0062] The case information characteristic data is characteristic data extracted based on the basic information of the case. The basic information of the case includes but is not limited to: name, gender, age, product identification, and text data corresponding to the picture. The product identification can be data that uniquely identifies a product from a product name, product ID, etc. It is understandable that the product identification can also be modified into a service identification, and the method steps of the present application can also be used to split the service tasks. The service identification can be data that uniquely identifies a service, such as a service name, service ID, etc.

[0063] The text data corresponding to the image is extracted from the image using OCR (Optical Character Recognition) technology. It is understandable that when the present application is applied to the underwriting business of the insurance industry, the images in the text data corresponding to the image include but are not limited to: medical record homepage, hospitalization summary, discharge summary, medical record, examination report, and fee receipt.

[0064] The commission information feature data is feature data extracted from the commission information text. The commission information text is a text used to describe the tasks and matters that need to be processed.

[0065] For S2, the task extraction rules may be obtained from a database, or may be written into a program for implementing the present application.

[0066] The task extraction rule is adopted to generate tasks from the commission information feature data, the generated tasks are placed in a list, and the obtained list is used as the first task list.

[0067] It is understandable that when the present application is applied to the underwriting business of the insurance industry, the task extraction rules include: rules for determining case types, rules for determining severity, and a task list set. The rules for determining case types are used to determine illness or accidents. The rules for determining severity are used to determine the severity of a case. The task list set includes: case type, product identification, severity, and task list.

[0068] For example, the case type is disease, the product identifier is health insurance, and the severity is an in-depth case. The task list includes the following tasks: designated medical institution investigation, case review of the hospital where the accident occurred, telephone interviews with the insured, investigation by peers, and investigation of medical insurance. No specific limitations are given in this example.

[0069] For example, the case type is accident, the product identification is accident insurance, and the severity is an in-depth case. The task list includes the following tasks: interview the accident victim or his family, review the medical records of medical institutions, visit the functional departments or doctors of the medical institutions, check designated hospitals (inpatient and outpatient), visit the place of residence, check the same industry (including insurance and mutual aid), and check social security. No specific limitations are given in this example.

[0070] Among them, the rule for judging the case type of the task extraction rule is adopted, and the case type is judged according to the commission information characteristic data to obtain the target case type; the rule for judging the severity of the task extraction rule is adopted, and the severity of the case is judged according to the commission information characteristic data to obtain the target severity; the product identification, target case type and target severity of the case characteristic data are used as associated data, and the associated data is searched in the task list set of the task extraction rule, and the task list corresponding to the associated data found in the task list set (that is, the associated data composed of case type, product identification and severity) is used as the first task list.

[0071] For S3, each vector element in the task classification prediction vector corresponds to a task label, and the number of vector elements in the task classification prediction vector is the same as the number of task labels.

[0072] Among them, multiple first training samples are obtained, and the first training samples include: case information feature sample data, task classification calibration values, and multiple first training samples are used to train the first initial model, and the first initial model after training is used as the preset task classification prediction model, wherein the first initial model adopts the XGboost model.

[0073] The method steps for training the first initial model using multiple first training samples are not described in detail here.

[0074] The XGboost model, or Extreme Gradient Boosting, is a tool for large-scale parallel boosted trees. It is one of the fastest and best open source boosted tree toolkits, more than 10 times faster than common toolkits. It is a fully enhanced version of the GB method and an algorithm based on residual optimization. It hopes to establish K regression trees so that the predicted value of the tree group is as close to the true value (accuracy) as possible and has the greatest possible generalization ability.

[0075] For S4, the task label corresponding to each vector element of 1 in the task classification prediction vector is used as a target task label, the tasks corresponding to each target task label are placed in a list, and the obtained list is used as the second task list.

[0076] For example, there are 15 task labels in total, the task classification prediction vector is [011111000000000], and the task labels corresponding to Task2, Task3, Task4, Task5, and Task6 are selected. Then the second task list includes: Task2, Task3, Task4, Task5, and Task6, and no specific limitation is made in this example.

[0077] For S5, all tasks in the first task list and the second task list are placed in the same list to obtain a task list to be deduplicated; then, the task list to be deduplicated is deduplicated, and the task list to be deduplicated after the task deduplication is used as the target task splitting result.

[0078] The target task splitting result includes: one or more task identifiers. The task identifier can be data that uniquely identifies a task, such as a task name and a task ID.

[0079] That is to say, the case corresponding to each case feature data can be split into one or more tasks.

[0080] In one embodiment, the above-mentioned obtaining of case characteristic data includes:

[0081] S11: Obtaining entrusted case data, wherein the entrusted case data includes: basic case information and entrusted information text;

[0082] S12: Generate label features according to the basic information of the case to obtain the case information feature data;

[0083] S13: Generate task description statements and task features according to the entrustment information text to obtain entrustment information feature data.

[0084] This embodiment obtains the case information characteristic data by generating label features according to the basic information of the case, and obtains the entrustment information characteristic data by generating task description statements and task features according to the entrustment information text, thereby providing support for subsequent task splitting based on characteristic data.

[0085] For S11, the entrusted case data input by the user may be obtained, the entrusted case data may be obtained from a database, or the entrusted case data may be obtained from a third-party application system.

[0086] For S12, label features are generated for the structured data in the case information text, and the case information feature data is obtained according to the generated label features.

[0087] The structured data in the case information text includes but is not limited to: name, gender, age, and product identification.

[0088] The case information feature data is a feature vector, and each vector element in the case information feature data corresponds to a label.

[0089] For example, the case information feature data is [2 3], where the label corresponding to the vector element value of 2 is gender, and the vector element value of 2 means the gender is female. The label corresponding to the vector element 3 is age, and the vector element value of 3 means the age is greater than or equal to 4 years old. No specific limitation is given in this example.

[0090] For S13, the delegation information text is split into task description sentences, and then task features are generated from each of the split task description sentences, and all the generated task features are used as the delegation information feature data.

[0091] The task description statement is a statement that describes the task. For example, the task description statement is to designate a medical institution for investigation, and no specific limitation is given in this example. Another example is that the task description statement is to interview the accident victim, and no specific limitation is given in this example.

[0092] In one embodiment, the above-mentioned generating label features according to the basic information of the case to obtain the case information feature data includes:

[0093] S121: Obtain a tag generation script;

[0094] S122: executing the label generation script to perform label prediction on the structured data in the basic information of the case to obtain a label prediction result;

[0095] S123: Generate a vector according to the label prediction result to obtain the case information feature data.

[0096] This embodiment implements label prediction for structured data in basic case information by executing a label generation script, and then generates case information feature data based on the label prediction results, providing support for subsequent task splitting based on feature data.

[0097] For S121, the label generation script may be obtained from a database or from a third-party application system.

[0098] The tag generation script is a script used to convert structured data into tag values. The tag generation script can be developed using Lua (a small scripting language written in standard C that can be compiled and run on almost all operating systems and platforms).

[0099] For S122, the label generation script is executed to perform label prediction on the structured data in the basic information of the case, that is, the structured data in the case information text is converted into label values.

[0100] For example, the label value for male is 1, the label value for female is 2, the label value for age less than 1 is 1, the label value for age greater than or equal to 1 and less than 4 is 2, and the label value for age greater than or equal to 4 is 3. No specific limitation is given in this example.

[0101] For example, the gender of the structured data in the case information text is female and the age is 3 years old, then the label prediction results are: gender 2 and age 2, and no specific limitation is given in this example.

[0102] For S123, a preset label arrangement order is used to generate a vector according to the label prediction result, and the generated vector is used as the case information feature data.

[0103] For example, the label prediction results are: gender 2 and age 3, and the preset label arrangement order is: gender, age, then the case information feature data is [2 3], and no specific limitation is given in this example.

[0104] In one embodiment, the generation of task description statements and task features according to the delegation information text is performed to obtain the delegation information feature data, including:

[0105] S131: Obtaining a sentence segmentation regular expression;

[0106] S132: using the sentence segmentation regular expression to split the task description sentence of the entrustment information text to obtain a task description sentence set;

[0107] S133: Acquire a task feature extraction model, wherein the task feature extraction model is a model trained according to the Bert model;

[0108] S134: Using the task feature extraction model, extract features from each of the task description statements in the task description statement set to obtain the entrustment information feature data.

[0109] This embodiment uses sentence segmentation regular expressions to split task description sentences and uses a task feature extraction model to generate task features, thereby providing support for subsequent task splitting based on feature data.

[0110] For S131, the sentence segmentation regular expression may be obtained from a database or from a third-party application system.

[0111] The statement segmentation regular expression is a regular expression obtained based on the separator, which includes but is not limited to numbers, commas, semicolons, and periods.

[0112] For S132, the sentence segmentation regular expression is used to segment the entrustment information text into short sentences, each short sentence obtained by segmentation is used as a task description sentence, and each task description sentence is used as the task description sentence set.

[0113] For S133, the task feature extraction model may be obtained from a database or from a third-party application system.

[0114] The task feature extraction model is a model used to extract features from text.

[0115] A plurality of second training samples are obtained, the second training samples include: a task description sample text, a task feature calibration value, and the second initial model is trained using the plurality of second training samples. The trained second initial model is used as a task feature extraction model, wherein the second initial model adopts a Bert (Bidirectional Encoder Representations from Transformers) model.

[0116] The method steps of using multiple second training samples to train the second initial model are not described in detail here.

[0117] For S134, each of the task description statements in the task description statement set is input into the task feature extraction model, the feature extraction data output by the task feature extraction model for each of the task description statements is spliced, and the spliced ​​data is used as the entrustment information feature data.

[0118] In one embodiment, the above-mentioned task merging process is performed according to the first task list and the second task list to obtain the target task splitting result, including:

[0119] S51: combining the tasks in the first task list and the second task list to obtain a task list to be deduplicated;

[0120] S52: performing task deduplication processing on the task list to be deduplicated to obtain a task list to be reviewed;

[0121] S53: Sending the task list to be reviewed to the task splitting and reviewing terminal;

[0122] S54: Obtaining the task review result corresponding to the task list to be reviewed and sent by the task splitting review end;

[0123] S55: When the task review result is passed, the task list to be reviewed is used as the target task splitting result;

[0124] S56: When the audit result of the task audit result is not passed, the task list to be audited is corrected according to the task correction data of the task audit result to obtain the target task splitting result.

[0125] This embodiment first performs task merging and deduplication processing on the first task list and the second task list, and then sends them to the task splitting review end for review. The target task splitting result is determined based on the review result of the task splitting review end, thereby achieving accurate splitting of the case tasks, providing support for the subsequent determination of the entrustment price based on the tasks, and helping to reduce the case entrustment cost.

[0126] For S51, the tasks in the first task list and the second task list are combined, and the data obtained after the combination is used as the task list to be deduplicated.

[0127] For S52, the tasks in the task list to be deduplicated are deduplicated, and the task list to be deduplicated that has completed the deduplication process is used as the task list to be reviewed.

[0128] That is to say, each task identifier in the task list to be reviewed is unique.

[0129] For S53, the task list to be reviewed is sent to the task splitting and reviewing end through a communication connection with the task splitting and reviewing end.

[0130] The auditor reviews the task list to be reviewed through the task splitting and reviewing terminal, and generates a task review result based on the task list to be reviewed.

[0131] The task review result includes: review result and task correction data. When the review result is passed, the task correction data is empty. When the review result is failed, the task correction data is not empty.

[0132] For S54, through the communication connection with the task splitting and reviewing end, the task review result corresponding to the task list to be reviewed and sent by the task splitting and reviewing end is obtained.

[0133] For S55, when the audit result of the task audit result is passed, it means that the result of the automatic splitting meets the audit requirements, so the task list to be reviewed is directly used as the target task splitting result.

[0134] For S56, when the audit result of the task audit result is failed, it means that the result of the automatic splitting does not meet the audit requirements. Therefore, according to the task correction data of the task audit result, the task list to be reviewed is replaced and updated, and the replaced and updated task list to be reviewed is used as the target task splitting result.

[0135] In one embodiment, after obtaining the task review result corresponding to the task list to be reviewed sent by the task splitting review terminal, the method further includes:

[0136] S541: When the task review result is not passed, a case is generated according to the entrusted case data corresponding to the case feature data and the target task splitting result to obtain a difficult case;

[0137] S542: updating the problem case to the problem case library;

[0138] S543: Update and train the preset task classification prediction model according to the difficult case library.

[0139] This embodiment generates a difficult case when the audit result of the task audit result is failed, thereby providing a data basis for the subsequent continuous optimization of the preset task classification prediction model.

[0140] For S541, when the audit result of the task audit result is failed, it means that the result of the automatic splitting does not meet the audit requirements. At this time, the accuracy of the automatically split tasks is not enough. Therefore, the preset case generation rules are used to generate cases based on the entrusted case data corresponding to the case feature data and the target task splitting results, and the generated cases are used as difficult cases.

[0141] For S542, the difficult problem case is added to the difficult problem case library, thereby completing the automatic update of the difficult problem case library.

[0142] For S543, a third training sample is generated according to each difficult case in the difficult case library; each third training sample is used to train the preset task classification prediction model, and the preset task classification prediction model after training is used for task classification prediction in an actual production environment.

[0143] In one embodiment, after performing task merging processing according to the first task list and the second task list to obtain the target task splitting result, the method further includes:

[0144] S61: Obtaining task allocation configuration data;

[0145] S62: performing task matching on the target task splitting result according to the task allocation configuration data to obtain a hit task identifier set and a miss task identifier set;

[0146] S63: Allocate each task in the hit task identifier set to a task underwriting end according to the task allocation configuration data;

[0147] S64: Obtain a task underwriting end label library, and assign each task in the missed task identification set to the task underwriting end according to the case label of the case feature data and the task underwriting end label library.

[0148] This embodiment implements the priority use of task allocation configuration data for task allocation. For tasks that cannot be allocated according to the task allocation configuration data, the task underwriting end label library is used for task allocation, thereby satisfying personalized task allocation and being beneficial to improving the efficiency and quality of case execution.

[0149] For S61, the task allocation configuration data may be obtained from a database, or may be obtained from a third-party application system.

[0150] The task allocation configuration data includes: task identification and task underwriting terminal identification, each task identification corresponds to a task underwriting terminal identification. The task underwriting terminal identification can be data that uniquely identifies a task underwriting terminal, such as the task underwriting terminal name and the task underwriting terminal ID. The task underwriting terminal is a software module used by crowdsourcing companies to receive tasks and feedback task execution results.

[0151] For S62, each task identifier in the task allocation configuration data is matched in the target task splitting result, each matched task identifier is used as a hit task identifier, and all hit task identifiers are used as a hit task identifier set.

[0152] Each hit task identifier is deleted from the target task splitting result, and the target task splitting result that has completed the deletion process is used as a set of missed task identifiers.

[0153] For S63, according to the task underwriting terminal identifier of the task allocation configuration data, the tasks corresponding to the respective task identifiers in the hit task identifier set are allocated to the task underwriting terminal.

[0154] For example, if the task identifier A1 is in the hit task identifier set, and the task underwriting end identifier corresponding to the task identifier A1 in the task allocation configuration data is CB1, then the task corresponding to the task identifier A1 in the hit task identifier set is allocated to the task underwriting end corresponding to CB1, and no specific limitation is given in this example.

[0155] For S64, the task underwriting end tag library can be obtained from a database, or from a third-party application system.

[0156] The task underwriting end label library includes: task underwriting end identification and task underwriting end label set. Each task underwriting end identification corresponds to a task underwriting end label set. The task underwriting end label set includes one or more task underwriting end labels. Task underwriting end labels include but are not limited to: province, city, administrative region, case type.

[0157] The case characteristic data also includes: case labels. Case labels include but are not limited to: case province, case city, case administrative region, and case type.

[0158] According to the case labels of the case feature data, a task underwriting end label set is matched from the task underwriting end label library, and the task underwriting end identifier corresponding to the matched task underwriting end label set is used as a candidate task underwriting end identifier.

[0159] The remaining executable data corresponding to each candidate task underwriting terminal identifier is obtained. The remaining executable data includes: task identifier and remaining executable quantity, wherein the remaining executable quantity is the number of tasks corresponding to the task identifier that can still be accepted.

[0160] According to the remaining executable data corresponding to each candidate task underwriting end identifier, each task in the missed task identifier set is allocated. For example, if the task identifier in the missed task identifier set is B1, B1 is searched for in the task identifiers in the remaining executable data corresponding to each candidate task underwriting end identifier, and the maximum value is found from the remaining executable quantities corresponding to each matched task identifier as the target remaining executable quantity, and the candidate task underwriting end identifier corresponding to the target remaining executable quantity is used as the target task underwriting end identifier, and the task identifier B1 in the missed task identifier set is allocated to the task underwriting end corresponding to the target task underwriting end identifier.

[0161] Reference Figure 2 , the present application also proposes a task splitting device based on artificial intelligence, the device comprising:

[0162] The data acquisition module 100 is used to acquire case characteristic data, wherein the case characteristic data includes: case information characteristic data and commission information characteristic data;

[0163] A first task list determination module 200 is used to obtain a task extraction rule, and use the task extraction rule to generate tasks according to the commission information feature data to obtain a first task list;

[0164] The task classification prediction vector determination module 300 is used to input the case information feature data into a preset task classification prediction model to perform task classification prediction and obtain a task classification prediction vector, wherein the preset task classification prediction model is a model obtained by training based on the XGboost model;

[0165] A second task list determination module 400, configured to determine tasks according to the task classification prediction vector to obtain a second task list;

[0166] The target task splitting result determination module 500 is used to perform task merging processing according to the first task list and the second task list to obtain a target task splitting result.

[0167] This embodiment adopts the task extraction rule to generate tasks according to the entrustment information feature data to obtain a first task list, and then inputs the case information feature data into a preset task classification prediction model to perform task classification prediction to obtain a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on the XGboost model; tasks are determined according to the task classification prediction vector to obtain a second task list, and finally tasks are merged according to the first task list and the second task list to obtain a target task splitting result, thereby achieving accurate splitting of case tasks, providing support for subsequent determination of entrustment prices according to tasks, which is beneficial to reducing case entrustment costs, and also providing support for determining the task underwriting end according to tasks, which is beneficial to improving the efficiency and quality of case execution.

[0168] In one embodiment, the data acquisition module 100 includes: a commissioned case data acquisition submodule, a case information characteristic data determination submodule, and a commissioned information characteristic data determination submodule;

[0169] The entrusted case data acquisition submodule is used to acquire entrusted case data, and the entrusted case data includes: basic case information and entrusted information text;

[0170] The case information feature data determination submodule is used to generate label features according to the basic information of the case to obtain the case information feature data;

[0171] The delegation information characteristic data determination submodule is used to generate a task description statement and a task characteristic according to the delegation information text to obtain the delegation information characteristic data.

[0172] In one embodiment, the case information feature data determination submodule includes: a script acquisition unit, a label prediction unit, and a vector generation unit;

[0173] The script acquisition unit is used to acquire a tag generation script;

[0174] The label prediction unit is used to execute the label generation script, perform label prediction on the structured data in the basic information of the case, and obtain a label prediction result;

[0175] The vector generation unit is used to generate a vector according to the label prediction result to obtain the case information feature data.

[0176] In one embodiment, the above-mentioned entrustment information characteristic data determination submodule includes: a task description statement splitting unit and a characteristic extraction unit;

[0177] The task description statement splitting unit is used to obtain a statement splitting regular expression, and use the statement splitting regular expression to split the task description statement of the entrustment information text to obtain a task description statement set;

[0178] The feature extraction unit is used to obtain a task feature extraction model, wherein the task feature extraction model is a model trained according to the Bert model. The task feature extraction model is used to perform feature extraction on each task description statement in the task description statement set to obtain the delegation information feature data.

[0179] In one embodiment, the target task splitting result determination module 500 comprises: a task list determination submodule to be reviewed, a task review submodule and a target task splitting result determination submodule;

[0180] The task list determination submodule to be reviewed is used to combine the tasks in the first task list and the second task list to obtain a task list to be deduplicated, and to perform task deduplication processing on the task list to be deduplicated to obtain a task list to be reviewed;

[0181] The task review submodule is used to send the task list to be reviewed to the task splitting review end, and obtain the task review result corresponding to the task list to be reviewed sent by the task splitting review end;

[0182] The target task splitting result determination submodule is used to, when the audit result of the task audit result is passed, use the task list to be reviewed as the target task splitting result; when the audit result of the task audit result is failed, correct the task list to be reviewed according to the task correction data of the task audit result to obtain the target task splitting result.

[0183] In one embodiment, the above-mentioned device further comprises: a difficult case library updating module and a preset task classification prediction model updating module;

[0184] The problem case library update module is used to generate a case based on the entrusted case data corresponding to the case feature data and the target task splitting result when the audit result of the task audit result is not passed, obtain a problem case, and update the problem case to the problem case library;

[0185] The preset task classification prediction model updating module is used to update and train the preset task classification prediction model according to the difficult case library.

[0186] In one embodiment, the above-mentioned device further includes: a first task allocation module and a second task allocation module;

[0187] The first task allocation module is used to obtain task allocation configuration data, perform task matching on the target task splitting result according to the task allocation configuration data, obtain a hit task identification set and a missed task identification set, and allocate each task in the hit task identification set to a task underwriting end according to the task allocation configuration data;

[0188] The second task assignment module is used to obtain a task underwriting end label library, and assign each task in the missed task identification set to the task underwriting end according to the case label of the case feature data and the task underwriting end label library.

[0189] Reference Figure 3 In an embodiment of the present application, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as a task splitting method based on artificial intelligence. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a task splitting method based on artificial intelligence is implemented. The artificial intelligence-based task splitting method includes: obtaining case feature data, the case feature data including: case information feature data and entrustment information feature data; obtaining task extraction rules, using the task extraction rules, generating tasks according to the entrustment information feature data, and obtaining a first task list; inputting the case information feature data into a preset task classification prediction model to perform task classification prediction, and obtaining a task classification prediction vector, wherein the preset task classification prediction model is a model obtained by training based on an XGboost model; determining tasks according to the task classification prediction vector, and obtaining a second task list; performing task merging processing according to the first task list and the second task list, and obtaining a target task splitting result.

[0190] This embodiment adopts the task extraction rule to generate tasks according to the entrustment information feature data to obtain a first task list, and then inputs the case information feature data into a preset task classification prediction model to perform task classification prediction to obtain a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on the XGboost model; tasks are determined according to the task classification prediction vector to obtain a second task list, and finally tasks are merged according to the first task list and the second task list to obtain a target task splitting result, thereby achieving accurate splitting of case tasks, providing support for subsequent determination of entrustment prices according to tasks, which is beneficial to reducing case entrustment costs, and also providing support for determining the task underwriting end according to tasks, which is beneficial to improving the efficiency and quality of case execution.

[0191] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, an artificial intelligence-based task splitting method is implemented, comprising the steps of: obtaining case feature data, wherein the case feature data comprises: case information feature data and entrustment information feature data; obtaining task extraction rules, adopting the task extraction rules, generating tasks according to the entrustment information feature data, and obtaining a first task list; inputting the case information feature data into a preset task classification prediction model for task classification prediction, and obtaining a task classification prediction vector, wherein the preset task classification prediction model is a model obtained by training based on an XGboost model; determining tasks according to the task classification prediction vector, and obtaining a second task list; and merging tasks according to the first task list and the second task list, and obtaining a target task splitting result.

[0192] The above-mentioned artificial intelligence-based task splitting method is performed, in this embodiment, by adopting the task extraction rule, generating tasks according to the entrustment information feature data, obtaining a first task list, and then inputting the case information feature data into a preset task classification prediction model for task classification prediction, to obtain a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on the XGboost model; tasks are determined according to the task classification prediction vector to obtain a second task list, and finally tasks are merged according to the first task list and the second task list to obtain the target task splitting result, thereby realizing accurate splitting of case tasks, providing support for subsequent determination of the entrustment price according to the tasks, which is beneficial to reducing the case entrustment cost, and also providing support for determining the task underwriting end according to the tasks, which is beneficial to improving the efficiency and quality of case execution.

[0193] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0194] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, device, article or method including the element.

[0195] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A task splitting method based on artificial intelligence, characterized in that: The method comprises: Acquiring case characteristic data, the case characteristic data including: case information characteristic data and entrustment information characteristic data; Acquire a task extraction rule, adopt the task extraction rule, generate tasks according to the commission information feature data, and obtain a first task list; Inputting the case information feature data into a preset task classification prediction model to perform task classification prediction, and obtaining a task classification prediction vector, wherein the preset task classification prediction model is a model trained based on an XGboost model; Determine tasks according to the task classification prediction vector to obtain a second task list; Perform task merging processing according to the first task list and the second task list to obtain a target task splitting result; The step of obtaining a task extraction rule, using the task extraction rule, generating tasks according to the entrustment information feature data, and obtaining a first task list includes: Adopting the rules for judging case type and judging severity of the task extraction rules, judging case type and severity according to the entrusted information feature data, and obtaining target case type and target severity; The product identification, target case type and target severity of the case feature data are used as associated data, and the associated data is searched in the task list set of the task extraction rule, and the task list corresponding to the associated data found in the task list set is used as the first task list; The acquisition of case characteristic data includes: Obtaining entrusted case data, wherein the entrusted case data includes: basic case information and entrusted information text; Generate label features based on the basic information of the case to obtain the case information feature data; According to the entrustment information text, task description sentences and task features are generated respectively to obtain the entrustment information feature data; The task description statement generation and the task feature generation are respectively performed according to the entrustment information text to obtain the entrustment information feature data, including: Get the statement segmentation regular expression; Using the sentence segmentation regular expression, the entrustment information text is split into task description sentences to obtain a task description sentence set; Obtaining a task feature extraction model, wherein the task feature extraction model is a model trained according to the Bert model; Using the task feature extraction model, extracting features from each of the task description statements in the task description statement set to obtain the entrustment information feature data; The performing task merging processing according to the first task list and the second task list to obtain a target task splitting result includes: Combine the tasks in the first task list and the second task list to obtain a task list to be deduplicated; Performing task deduplication processing on the task list to be deduplicated to obtain a task list to be reviewed; Sending the task list to be reviewed to the task splitting and reviewing end; Obtaining the task review result corresponding to the task list to be reviewed sent by the task splitting review end; When the audit result of the task audit result is passed, the task list to be audited is used as the target task splitting result; When the audit result of the task audit result is not passed, the task list to be audited is corrected according to the task correction data of the task audit result to obtain the target task splitting result.

2. The task splitting method based on artificial intelligence according to claim 1 is characterized in that: The label feature generation according to the basic information of the case to obtain the case information feature data includes: Get the tag generation script; Executing the label generation script to perform label prediction on the structured data in the basic information of the case to obtain a label prediction result; A vector is generated according to the label prediction result to obtain the case information feature data.

3. The task splitting method based on artificial intelligence according to claim 1 is characterized in that: After obtaining the task review result corresponding to the task list to be reviewed sent by the task splitting review end, the method further includes: When the task review result is not passed, a case is generated according to the entrusted case data corresponding to the case feature data and the target task splitting result to obtain a difficult case; Updating the problem case to the problem case library; The preset task classification prediction model is updated and trained according to the difficult case library.

4. The task splitting method based on artificial intelligence according to claim 1 is characterized in that: After performing task merging processing according to the first task list and the second task list to obtain a target task splitting result, the method further includes: Get task allocation configuration data; According to the task allocation configuration data, task matching is performed on the target task splitting result to obtain a hit task identification set and a miss task identification set; Allocate each task in the hit task identifier set to a task underwriting end according to the task allocation configuration data; A task underwriting end label library is obtained, and each task in the missed task identification set is allocated to the task underwriting end according to the case label of the case feature data and the task underwriting end label library.

5. A task splitting device based on artificial intelligence, characterized in that: The device comprises: A data acquisition module, used to acquire case characteristic data, wherein the case characteristic data includes: case information characteristic data and commission information characteristic data; A first task list determination module is used to obtain a task extraction rule, adopt the task extraction rule, generate tasks according to the commission information feature data, and obtain a first task list; A task classification prediction vector determination module is used to input the case information feature data into a preset task classification prediction model to perform task classification prediction and obtain a task classification prediction vector, wherein the preset task classification prediction model is a model obtained by training based on an XGboost model; A second task list determination module, used to determine tasks according to the task classification prediction vector to obtain a second task list; a target task splitting result determination module, configured to perform task merging processing according to the first task list and the second task list to obtain a target task splitting result; The acquiring task extraction rule, using the task extraction rule, generating tasks according to the commission information feature data, and obtaining a first task list includes: Adopting the rules for judging case type and judging severity of the task extraction rules, judging case type and severity according to the entrusted information feature data, and obtaining target case type and target severity; The product identification, target case type and target severity of the case feature data are used as associated data, and the associated data is searched in the task list set of the task extraction rule, and the task list corresponding to the associated data found in the task list set is used as the first task list; The task description statement generation and the task feature generation are respectively performed according to the entrustment information text to obtain the entrustment information feature data, including: Get the statement segmentation regular expression; Using the sentence segmentation regular expression, the entrustment information text is split into task description sentences to obtain a task description sentence set; Obtaining a task feature extraction model, wherein the task feature extraction model is a model trained according to the Bert model; Using the task feature extraction model, extracting features from each of the task description statements in the task description statement set to obtain the entrustment information feature data; The performing task merging processing according to the first task list and the second task list to obtain a target task splitting result includes: Combine the tasks in the first task list and the second task list to obtain a task list to be deduplicated; Performing task deduplication processing on the task list to be deduplicated to obtain a task list to be reviewed; Sending the task list to be reviewed to the task splitting and reviewing end; Obtaining the task review result corresponding to the task list to be reviewed sent by the task splitting review end; When the audit result of the task audit result is passed, the task list to be audited is used as the target task splitting result; When the audit result of the task audit result is not passed, the task list to be audited is corrected according to the task correction data of the task audit result to obtain the target task splitting result.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images