Large model-based quality inspection method, device, equipment, storage medium and program

Abstract

The disclosure provides a large model-based quality inspection method, device, equipment, storage medium and program, relating to the technical field of data processing, specifically relating to artificial intelligence, deep learning and large language model technology, which can be applied in the field of AI medical technology, wherein the method comprises: obtaining to-be-inspected data of a target platform and a target quality inspection prompt template; wherein the target quality inspection prompt template comprises quality inspection task instructions and quality inspection question classification content; splicing the to-be-inspected data and the target quality inspection prompt template to obtain a target quality inspection prompt; inputting the target quality inspection prompt into a target large model to automatically inspect the to-be-inspected data according to the target quality inspection prompt through the target large model, and obtaining first quality inspection data. The embodiment of the disclosure can reduce the artificial quality inspection cost of platform quality inspection and improve the efficiency and accuracy of platform quality inspection.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, specifically artificial intelligence, deep learning, and large language model technology, and can be applied to the field of AI medical technology. Background Technology

[0002] In the operation of a platform, in order to improve product quality and user satisfaction, it is often necessary to conduct quality checks on the platform's operational data. For example, in online medical platforms, the quality of doctors' consultation services needs to be checked; in consultation platforms, the data on the consultation services provided needs to be checked; and in customer service platforms, the quality of customer service needs to be checked, and so on. Summary of the Invention

[0003] This disclosure provides a quality inspection method, apparatus, device, storage medium, and program based on a large model, which can reduce the manual quality inspection cost of platform quality inspection and improve the efficiency and accuracy of platform quality inspection.

[0004] In a first aspect, embodiments of this disclosure provide a quality inspection method based on a large model, including:

[0005] Obtain the data to be inspected and the target quality inspection prompt template from the target platform; wherein, the target quality inspection prompt template includes a description of the quality inspection task and a classification of quality inspection issues;

[0006] The data to be inspected and the target quality inspection prompt template are combined to obtain the target quality inspection prompt.

[0007] The target quality inspection prompt is input into the target large model, so that the target large model can automatically perform quality inspection on the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

[0008] Secondly, embodiments of this disclosure provide a quality inspection device based on a large model, comprising:

[0009] The data template acquisition module is used to acquire the data to be inspected and the target quality inspection prompt template of the target platform; wherein, the target quality inspection prompt template includes the quality inspection task description and the quality inspection problem classification content;

[0010] The target quality inspection prompt acquisition module is used to concatenate the data to be inspected and the target quality inspection prompt template to obtain the target quality inspection prompt.

[0011] The first quality inspection data acquisition module is used to input the target quality inspection prompt into the target large model, so that the target large model can automatically perform quality inspection on the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

[0012] Thirdly, embodiments of this disclosure provide an electronic device, including:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the large-model-based quality inspection method provided in the first aspect embodiment.

[0016] Fourthly, embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the large-model-based quality inspection method provided in the first aspect embodiment.

[0017] Fifthly, this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the large-model-based quality inspection method provided in the first aspect embodiment.

[0018] This embodiment of the disclosure obtains the data to be inspected from the target platform, and a target quality inspection prompt template including quality inspection task descriptions and quality inspection problem classifications. The data to be inspected and the target quality inspection prompt template are then concatenated to obtain target quality inspection prompts. These prompts are then input into a target large-scale model, which performs automated quality inspection on the data to be inspected based on the target quality inspection prompts, resulting in first quality inspection data. Since the quality inspection task descriptions and quality inspection problem classifications in the target quality inspection prompt template clearly define the quality inspection rules, the target large-scale model can quickly and accurately output the platform's quality inspection results based on the concatenated target quality inspection prompts. This technical solution solves the problems of low efficiency and low accuracy in existing platform quality inspection methods, thereby reducing the manual quality inspection cost of the platform and improving the efficiency and accuracy of platform quality inspection.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0020] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0021] Figure 1 This is a flowchart of a quality inspection method based on a large model provided in an embodiment of this disclosure;

[0022] Figure 2 This is a flowchart of another quality inspection method based on a large model provided in this disclosure embodiment;

[0023] Figure 3 This is a schematic diagram of a process for quality inspection of online consultation data on an online medical platform, provided by an embodiment of this disclosure;

[0024] Figure 4 This is a structural diagram of a quality inspection device based on a large model provided in an embodiment of this disclosure;

[0025] Figure 5 This is a schematic diagram of the structure of an electronic device used to implement the large-model-based quality inspection method of the present disclosure embodiments. Detailed Implementation

[0026] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0027] In one example Figure 1 This is a flowchart of a large-model-based quality inspection method provided in this embodiment. This embodiment is applicable to situations where a target quality inspection prompt, obtained by concatenating a target quality inspection prompt template including quality inspection task descriptions and quality inspection problem classifications, is input into a target large model to perform quality inspection on the data to be inspected on the target platform. This method can be executed by a large-model-based quality inspection device, which can be implemented in software and / or hardware and is generally integrated into an electronic device. This electronic device can be a terminal device or a server device, as long as it can execute the large-model-based quality inspection method. This embodiment does not limit the specific type of electronic device. Correspondingly, as... Figure 1 As shown, the method includes the following operations:

[0028] S110. Obtain the data to be inspected from the target platform and the target quality inspection prompt template; wherein, the target quality inspection prompt template includes a description of the quality inspection task and a classification of quality inspection issues.

[0029] The target platform can be a platform type that requires quality inspection of the data generated on the platform to determine the quality of services provided based on the inspection results. The data to be inspected can be data generated by the target platform during its operation that requires quality inspection. The target quality inspection prompt template can be a pre-configured prompt template based on the target platform type. This template can be combined with the data to be inspected to generate a complete prompt (prompt). The quality inspection task description can be a task description of inspecting the data to be inspected on the target platform. The quality inspection problem classification content can be used to list and configure rules for classifying and judging potential problems in the target platform.

[0030] When quality inspection of data to be inspected on a target platform is required, the data can be obtained first. For example, the target platform may include, but is not limited to, at least one of an online medical platform, a consultation platform, and a customer service platform. When the target platform is an online medical platform, the data to be inspected may be online consultation data generated by the online medical platform. When the target platform is a consultation platform, the data to be inspected may be online consultation data generated by the consultation platform. When the target platform is a customer service platform, the data to be inspected may be user customer service dialogue data generated by the customer service platform. Furthermore, a target quality inspection prompt template suitable for the target platform can be obtained. It is understood that the structure and content of the target quality inspection prompt template may differ depending on the type of target platform, and this disclosure does not impose limitations on this. The target quality inspection prompt template must include at least a quality inspection task description and a quality inspection problem classification. The quality inspection task description in the target quality inspection prompt template can instruct the large model on an automated quality inspection process for the data to be inspected. The quality inspection problem categories in the target quality inspection prompt template can instruct the large model to detect problems in the data to be inspected according to the set quality inspection rules.

[0031] S120. The data to be inspected and the target quality inspection prompt template are combined to obtain the target quality inspection prompt.

[0032] Among them, the target quality inspection prompt can be a prompt obtained by splicing the data to be inspected and the target quality inspection prompt template.

[0033] Accordingly, after obtaining the data to be inspected from the target platform and the target quality inspection prompt template, the data to be inspected from the target platform can be added to the relevant positions in the target quality inspection prompt template to obtain the final target quality inspection prompt that can be input into the large model. For example, it can be added to the quality inspection task description section of the target quality inspection prompt template to provide specific task data to the large model.

[0034] S130. Input the target quality inspection prompt into the target large model, so that the target large model can automatically inspect the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

[0035] The target large model can be any large model that can automatically perform quality inspection functions based on prompts. The first quality inspection data can be the quality inspection result output by the target large model on the data to be inspected on the target platform.

[0036] Large Language Models (LLMs), also known as large-scale language models, are deep learning models trained on massive amounts of relevant data (such as text, speech, or combined text and image data) to process text sequences. They can generate natural language text or understand the meaning of spoken language text. These models typically have billions of parameters. Large LLMs can handle various natural language tasks, such as text classification, question answering, and dialogue, and have wide applications. The input to a large LLM is data, such as text, speech, or combined text and image data. The large LLM encodes the input data to obtain corresponding word vectors, and then decodes these word vectors to automatically process the input data and obtain the corresponding output data. For example, text can be input into a large LLM, which processes and predicts the input text, outputting the corresponding response text.

[0037] After concatenating the target quality inspection prompt, it can be input into the target large model. The target large model then uses the target quality inspection prompt as input data, encodes it to obtain the corresponding word vector representation, and further decodes the encoded word vectors, thereby automatically completing the processing of the input data and obtaining the output data corresponding to the input data as the first quality inspection data, realizing automated quality inspection of the data to be inspected.

[0038] This embodiment of the disclosure obtains the data to be inspected from the target platform, and a target quality inspection prompt template including quality inspection task descriptions and quality inspection problem classifications. The data to be inspected and the target quality inspection prompt template are then concatenated to obtain target quality inspection prompts. These prompts are then input into a target large-scale model, which performs automated quality inspection on the data to be inspected based on the target quality inspection prompts, resulting in first quality inspection data. Since the quality inspection task descriptions and quality inspection problem classifications in the target quality inspection prompt template clearly define the quality inspection rules, the target large-scale model can quickly and accurately output the platform's quality inspection results based on the concatenated target quality inspection prompts. This technical solution solves the problems of low efficiency and low accuracy in existing platform quality inspection methods, thereby reducing the manual quality inspection cost of the platform and improving the efficiency and accuracy of platform quality inspection.

[0039] In one example Figure 2 This is a flowchart of another quality inspection method based on a large model provided in this disclosure. Based on the technical solutions of the above embodiments, this disclosure has made optimizations and improvements, and provides a specific structure for obtaining the quality inspection data to be inspected from the target platform and the classification content of quality inspection issues, as well as a variety of specific optional implementation methods for generating the final target quality inspection data based on the first quality inspection data.

[0040] like Figure 2 The quality inspection method based on a large model shown includes:

[0041] S210. In response to the quality inspection process triggering command initiated by the target platform, obtain the original quality inspection data.

[0042] The quality inspection process trigger instruction can be an instruction triggered by a relevant event on the target platform, which can be used to instruct the start of the quality inspection process. The raw quality inspection data can be unprocessed data generated by the target platform that requires quality inspection.

[0043] During operation, the target platform may be triggered by various events that generate quality inspection process trigger commands. For example, taking an online medical platform, a quality inspection process trigger command can be generated when a doctor issues a prescription or when an online consultation ends. Similarly, on a customer service platform, a quality inspection process trigger command can be generated when the service provider sends an evaluation request to a user or when the current customer service process ends. Upon detecting a quality inspection process trigger command initiated by the target platform, the system can respond by obtaining raw quality inspection data from the target platform.

[0044] S220. The original quality inspection data is filtered to obtain the data to be inspected.

[0045] The data to be inspected may include at least one of text data, image data, card data, and audio / video data.

[0046] Understandably, some of the raw quality inspection data generated by the target platform is of low quality and does not require quality inspection. For example, in an online medical platform, if a user simply sends "Hello" and ends the consultation process on the current consultation interface, this type of data does not require quality inspection. Therefore, after obtaining the raw quality inspection data from the target platform, the raw quality inspection data can be filtered to select data with quality inspection requirements and value as the data to be inspected.

[0047] The above technical solution filters out data with quality inspection needs and value from the original quality inspection data of the target platform as the data to be inspected. This can improve the quality of the data to be inspected, prevent low-quality data from entering the quality inspection process, thereby reducing the resource consumption of low-quality data and improving quality inspection efficiency.

[0048] In an optional embodiment of this disclosure, the method may further include: when it is determined that there are multiple quality inspection process triggering instructions initiated by the target platform, selecting a target quality inspection process triggering instruction from the multiple quality inspection process triggering instructions by means of Redis lock acquisition; and obtaining the original quality inspection data in response to the target quality inspection process triggering instruction.

[0049] The target quality inspection process trigger instruction can be one of the multiple quality inspection process trigger instructions selected from among them.

[0050] Understandably, in some situations, multiple events may simultaneously trigger quality inspection process trigger commands on the target platform. For example, if a user leaves a negative review while ending a conversation on a customer service platform, both the conversation ending event and the negative review event can trigger a quality inspection process trigger command. Since each quality inspection process trigger command can initiate a quality inspection process, when multiple quality inspection process trigger commands are detected from the target platform at the same time, a Redis lock-grabbing mechanism can be used to select one of the multiple quality inspection process trigger commands as the target quality inspection process trigger command. Furthermore, the original quality inspection data can be retrieved in response to the target quality inspection process trigger command. Redis lock-grabbing refers to a mechanism implemented by Redis when multiple processes access shared resources, ensuring that only one process can acquire the lock and execute operations at a time, thereby avoiding data competition and conflicts.

[0051] The above technical solution uses Redis lock-grabbing to select the target quality inspection process trigger instruction from multiple quality inspection process trigger instructions. This avoids multiple quality inspection process trigger instructions triggering the same quality inspection process multiple times, and prevents the target large model from performing multiple quality inspections on the same data to be inspected, thereby improving resource utilization and the quality inspection efficiency of the large model.

[0052] S230. Obtain the target quality inspection prompt template, and concatenate the data to be inspected and the target quality inspection prompt template to obtain the target quality inspection prompt.

[0053] S240. Input the target quality inspection prompt into the target large model, so that the target large model can automatically inspect the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

[0054] In one optional embodiment of this disclosure, the quality inspection issue classification content may include issue ID (Identification), issue type, issue definition description, and exemption conditions: wherein, the issue ID is configured with a predefined quality inspection score; the first quality inspection data may include a list of issue IDs, severity, and cause of occurrence.

[0055] The issue ID uniquely identifies the issue type. The predefined quality inspection score is the score configured for each issue ID; a higher score indicates a more severe issue. The issue definition description is the data defining each issue. Exemption conditions are the conditions configured to prevent issues of a specific type from being classified as that type.

[0056] In this embodiment of the disclosure, optionally, the quality inspection issue classification content may include an issue ID to identify the issue type. Each issue ID can be pre-configured with a predefined quality inspection score to identify the severity of the corresponding issue type. Furthermore, for any type of issue that may occur in the target platform, a corresponding issue type can be configured. Additionally, the quality inspection issue classification content can define the concepts of different issue types by configuring issue definition descriptions. To avoid large-scale model illusions, exemption conditions corresponding to each issue type can also be configured in the quality inspection issue classification content. Accordingly, the target large-scale model can determine which aspects of the problem exist in the data to be inspected based on the issue type and issue definition description, and review the detected problems according to the exemption conditions, deleting misjudged issue types to obtain the final quality inspection result.

[0057] Optionally, the first quality inspection data output by the target large model may include, but is not limited to, a list of issue IDs, severity levels, and reasons for detection. The list of issue IDs may include all issue IDs detected by the target large model in the quality inspection data. The severity level reflects the target large model's preliminary assessment of the severity of the issues in the quality inspection data. The reasons for detection explain why the target large model made this quality inspection result.

[0058] Therefore, by constructing quality inspection issue classification content with elements such as issue ID, issue type, issue definition description, and exemption conditions, the target large model can detect all issue types that meet the set issue classification rules from the data to be inspected based on the issue type, issue definition description, and exemption conditions, and generate a final issue ID list. This allows the target large model to output the final first quality inspection data based on the issue ID list, thereby improving the accuracy of the target large model's quality inspection of the target platform.

[0059] S250. Input the data to be inspected into the target association model to perform automated quality inspection on the data to be inspected through the target association model, and obtain the second quality inspection data.

[0060] The target association model can be any other model used to perform automated quality inspection on the data to be inspected individually. The second quality inspection data can be the result data obtained by performing automated quality inspection on the data to be inspected using the target association model.

[0061] In this embodiment of the disclosure, the data to be inspected can also be input into other target association models as input data, so that the data to be inspected can be automatically inspected independently by the target association models to obtain the second quality inspection data.

[0062] S260. Generate target quality inspection data based on the first quality inspection data and the second quality inspection data.

[0063] Correspondingly, after obtaining the two quality inspection results generated by the target large model and the target association model for the data to be inspected, the final target quality inspection data can be generated by referring to the first quality inspection data output by the target large model for the data to be inspected and the second quality inspection data generated by the target association model for the same data to be inspected.

[0064] The above technical solution can further improve the accuracy of quality inspection of target platform data by referring to the quality inspection results of different models on the same data to be inspected.

[0065] In an optional embodiment of this disclosure, generating target quality inspection data based on the first quality inspection data and the second quality inspection data may include: generating a summary problem ID list based on the problem ID list of the first quality inspection data and the problem ID list of the second quality inspection data; calculating a summary quality inspection score based on the predefined quality inspection score of each problem ID included in the summary problem ID list; and using the summary quality inspection score, the severity, and the cause of the hit as the target quality inspection data.

[0066] The summary issue ID list can be a list obtained by summarizing the issue ID lists of the first and second quality inspection data, and can include all issue IDs output by the target large model and the target related model. The summary quality inspection score can be the final score calculated for the data to be inspected.

[0067] Optionally, the second quality inspection data obtained by the target association model through automated quality inspection of the data to be inspected can also include a list of issue IDs. Correspondingly, all issue IDs included in the issue ID lists of the first and second quality inspection data can be aggregated to obtain a aggregated issue ID list. Further, based on the aggregated issue ID list, predefined quality inspection scores for all issue IDs in the aggregated issue ID list are obtained and accumulated to obtain a aggregated quality inspection score. Finally, the aggregated quality inspection score, severity, and cause of occurrence can be used as the target quality inspection data.

[0068] The above technical solution generates target quality inspection data that includes the scores of different models, which can further improve the accuracy of quality inspection of target platform data and reduce the cost of manual quality inspection.

[0069] In an optional embodiment of this disclosure, the method may further include: filtering the target quality inspection data whose aggregated quality inspection score is greater than or equal to the quality inspection score threshold as the first quality inspection data to be reviewed; sending the first quality inspection data to be reviewed to the first reviewer for manual review; obtaining the second quality inspection data to be reviewed from the first reviewer; and sending the second quality inspection data to be reviewed to the second reviewer for manual review.

[0070] The quality inspection score threshold can be configured according to actual needs, and this embodiment does not limit the specific value of the quality inspection score threshold. The first quality inspection data to be reviewed can be data that requires manual review by a first reviewer. The first reviewer can be the person who performs the first manual review of the quality inspection results output by the model. The second quality inspection data to be reviewed can be quality inspection data that requires further manual review after being reviewed by the first reviewer. The second reviewer can be the person who performs a second manual review of the quality inspection data reviewed by the first reviewer.

[0071] To further ensure the accuracy of quality inspection of the data to be inspected on the target platform, after obtaining the target quality inspection data based on the target large model and the target association model, the target quality inspection data can be organized. Specifically, it can be sorted according to the aggregate quality inspection score in descending order. Further, the sorted target quality inspection data can be filtered, with those having an aggregate quality inspection score greater than or equal to the quality inspection score threshold selected as the first batch of data to be reviewed. Finally, the first batch of data to be reviewed can be sent to the first reviewer for manual review. The first reviewer can be a professional on the target platform, such as a quality inspection doctor for an online medical platform, a back-end consultation service staff for a consultation platform, or a back-end customer service staff for a customer service platform. The first reviewer selecting target quality inspection data with higher aggregate quality inspection scores for review can improve the efficiency and accuracy of the quality inspection review. Considering the heavy workload of the first reviewer and the potential for subjective bias, which could lead to unreasonable review results, a second review of the quality inspection data can be obtained from the first reviewer. This second review data can then be sent to a more specialized second reviewer for further manual review, ensuring the accuracy of the target quality inspection data. Optionally, the second reviewer could be an expert. Furthermore, target quality inspection data with scores below the threshold can be collected as a batch of data to be reviewed. This batch of data can then be sent to relevant personnel for mass manual review, further improving the efficiency of the target quality inspection data review.

[0072] In one optional embodiment of this disclosure, obtaining the second quality inspection data to be reviewed by the first reviewer may include: obtaining the original quality inspection data of the first reviewer; and filtering the original quality inspection data, including quality inspection complaint feedback data, as the second quality inspection data to be reviewed.

[0073] The original review and quality inspection data can be the original review result data obtained by the first reviewer through manual review of the target quality inspection data. The quality inspection appeal feedback data can be the data on appeal feedback opinions raised regarding the original review and quality inspection data.

[0074] In some cases, the original quality inspection data obtained through manual review by the first reviewer may receive quality inspection complaint feedback data. For example, in online medical platforms, doctors providing consultation services may appeal the original quality inspection data generated by the quality inspection doctor. In customer service platforms, customer service personnel may appeal the original quality inspection data generated by the customer service representative. Furthermore, the original quality inspection data, including quality inspection complaint feedback data, can be filtered into second-stage quality inspection data to be reviewed, and then sent to a second reviewer for manual review. The advantage of this setup is that it can protect the legitimate rights and interests of those being inspected and improve the professionalism and accuracy of the quality inspection results.

[0075] The above-mentioned technical solution, by integrating large models, intelligent strategies and precise intervention of human rules, can overcome the problem of quality inspection misjudgment caused by the illusion of large models on a point-to-point basis. It can not only improve the accuracy and reliability of quality inspection, but also promote the widespread application of artificial intelligence technology of large models in the field of quality inspection, bringing intelligent transformation to the automated quality inspection industry.

[0076] Specific application scenarios

[0077] In AI (Artificial Intelligence) healthcare scenarios involving online consultations, online medical platforms provide doctors and patients with online communication capabilities, enabling them to consult with doctors online. These platforms have a supervisory responsibility for the quality of doctors' consultation services and are obligated to help doctors improve their skills to better serve patients. Currently, the quality of online consultation data from online medical platforms is often assessed through sampling or full-scale testing by quality control experts and operational staff. This existing method of quality control for online consultation data has the following problems: sampling tests may result in incomplete samples and omissions, leading to low accuracy; full-scale testing requires a large number of doctors and operational resources, consuming significant manpower and incurring high costs; manual testing is subject to subjective bias, resulting in low accuracy, long processing times, poor timeliness, and significant difficulty in quality control.

[0078] Figure 3 This is a schematic diagram illustrating a process for quality inspection of online consultation data on an online medical platform, provided by an embodiment of this disclosure. In one example, such as... Figure 3 As shown, taking an online medical platform as an example, the process of quality inspection of online consultation data on an online medical platform may include the following operations:

[0079] In online medical platforms, a quality inspection process can be triggered after each consultation process is completed. If multiple events trigger a quality inspection process simultaneously within a unified consultation process, a Redis lock-grabbing mechanism can be used to limit the quality inspection process to one event at a time. Once a consultation process triggers a quality inspection, it's first necessary to determine if the consultation data requires inspection. Some consultation data doesn't inherently require inspection; for example, some patients might simply say "hello" before ending the consultation. If it's determined that the consultation data requires inspection, then a collection of doctor-patient communication, prescription information, and transcripts of phone conversations, along with key information from key cards, can be retrieved and entered into the AI ​​quality inspection process.

[0080] In the AI ​​quality inspection process, the first step is to determine if the operating environment is an online environment. This avoids introducing large-scale model billing processes into offline testing environments, which could lead to unnecessary resource waste. Additionally, if the large model is configured with business rate limiting, the traffic requests for quality inspection data also need to be rate-limited through a business rate limiter to prevent excessive quality inspection requests from impacting the large model's processing of other business operations. Further, the target quality inspection prompt template (prompt template) for the large model is obtained, and the consultation data to be inspected is concatenated with the prompt template to obtain the final prompt. The concatenated prompt is then input into the large model, which uses the prompt to perform automated initial quality inspection screening of the consultation data. The large model can combine the problem classification and exemption conditions defined in the prompt to assess the risk of the consultation dialogue. Simultaneously, the request metadata of the large model, such as input and output parameters, is asynchronously saved for later optimization of the large model's quality inspection process.

[0081] After the large model returns the quality inspection results, the data processing stage begins. During data processing, other target association models can be used to apply strategic risk control to key information from doctor-patient communication, yielding another type of quality inspection result. By combining various intelligent quality inspection analysis strategies from the large model and target association models, a quality inspection work order is generated, identifying problem risk items and quality inspection scores, and summarizing whether the consultation service was satisfactory.

[0082] For unqualified consultation data, a quality control work order, including the model's quality control results and the consultation data, can be sent to the quality control physician for review. Data that is disputed during manual review can also be sent to a quality control expert for further review.

[0083] In an optional embodiment of this disclosure, the target quality inspection prompt template may include various structural elements such as role definition, quality inspection task description, quality inspection task requirements, quality inspection problem classification, and output format description. The role definition instructs the large model to determine the current quality inspection role, such as a quality inspection doctor or quality inspection customer service representative. The quality inspection task description defines the quality inspection task. The quality inspection task requirements clearly specify a series of requirements related to the quality inspection task. The output format description defines the output format of the first quality inspection data.

[0084] The quality control issues in the prompt template applicable to online medical platforms can include, but are not limited to: failure to disclose routine medical orders, failure to highlight critical illnesses, incomplete medication instructions, potential critical disease risks, and unanswered patient questions.

[0085] In a specific example, a prompt template suitable for an online healthcare platform may include the following:

[0086] ##Role

[0087] You are a highly qualified quality control physician with extensive medical expertise and experience, and who is fair and objective.

[0088] ##Task Definition

[0089] Identify the categories and causes of problems exhibited by doctors in the following doctor-patient dialogues; for specific issues requiring assessment, please refer to the "Quality Inspection Problem Classification Content".

[0090] [[dialog]]

[0091] ##Require

[0092] - When one or more problems exist, output the results according to the `output format sample`.

[0093] - The professional content output needs to be based on high-level evidence-based medicine information sources such as medical professional guidelines and instruction manuals.

[0094] - The subject of the test is "doctor-patient dialogue"; fabricated dialogue information is not allowed; hypothetical content is not allowed.

[0095] - If the "Doctor-Patient Dialogue" is empty, or only contains a doctor's greeting, terminate the process and do not perform problem detection.

[0096] - Concise and clear language style

[0097] ##Categorization of Quality Inspection Issues

[0098] |id|Problem Type|Problem Definition Description|This item will not be evaluated in the following situations and scenarios|

[0099] |----|----|----|----|

[0100] |8050|AI Routine Medical Orders Not Informed|The doctor did not provide recommendations for examinations / treatments appropriate for the patient's current condition; for some chronic diseases, the doctor did not provide lifestyle recommendations. |1. This item will not be awarded if the patient's condition does not require examinations / treatments / medications, etc.; 2. This item will not be awarded if a diagnosis cannot be reached based on the doctor-patient dialogue; 3. This item will not be awarded if the patient requests or the doctor proactively cancels the consultation and refunds the fees. In the above situations, this item will not be awarded.|

[0101] |8060|AI-generated Critical Illness Not Notified| Based on a complete doctor-patient dialogue, the doctor newly identified a patient who needed to be reminded to seek medical attention, but did not remind the patient to go to the hospital. |1. If the doctor expressed the need for the patient to seek medical attention at any point in the complete dialogue, this is considered a notification and will not be penalized; 2. The doctor has already reported the issue to the platform; 3. The patient is currently in a hospital / hospitalized; 4. The patient's condition was critical only in the past / historical period, and is currently stable / mild. In the above situations, this will not be penalized.|

[0102] |8070|Incomplete Medication Instructions|When the doctor recommends a prescription drug, the usage / dosage is not mentioned.|1. The doctor's instructions are for a non-prescription drug; 2. The patient explicitly refuses the medication; 3. The patient did not fill in / provide the basic medical history information required for prescription. In the above situations, this item will not be judged.|

[0103] |8801|AI Potential Risk of Critical Diseases|Accurately identifies potential risks of exposure to critical diseases in doctor-patient conversations. Risks include, but are not limited to, bites, scratches, licking of mucous membranes or broken skin by host animals, or open wounds. |This item is not considered if the doctor has already reminded the patient to receive the relevant vaccine.|

[0104] |9021|Unanswered Patient Questions by AI|Questions raised by patients but not answered by doctors|1. Questions that do not require a doctor's answer; 2. Questions outside the doctor's current specialty; 3. Brief answers given by the doctor without detailed explanations are not considered unanswered; 4. When the patient's expression affects the doctor's understanding and judgment, the doctor can follow up to clarify the patient's true question / intention, which is not considered unanswered; 5. In the context of a complete doctor-patient dialogue, the doctor has answered all the patient's main questions. In the above situations, this item will not be considered.|

[0105] ## Output Format Description

[0106] -id(question category ID)(string): See the id column of the question category table for details.

[0107] -level(severity) (string): Value range: high, medium, low (for quality control physicians' reference).

[0108] -reason(hit reason)(string): The reason why you made this judgment (for quality control doctors' reference)

[0109] - Just return a single JSON file.

[0110] Example 1: When there is a problem with the task definition, return...

[0111] {"items":[{"id":"1","level":"high","reason":"xxxxx"},{"id":"2","level":"low","reason":"xxxxxx"}]}

[0112] Example 2: When there is no problem defining the task, return...

[0113] {"items":[]}

[0114] Referring to the specific example above, in the target quality inspection prompt template applicable to online medical platforms, the structure "##role(...)" defines the role in the template, indicating that the current quality inspection role used by the large model is that of a quality inspection doctor. The structure "##task definition(...)" describes the quality inspection task, indicating the specific definition of the large model's quality inspection task and the rules to be followed. In the example above, the structure "##task definition(...)" explicitly requires performing quality inspection operations on the given data according to the quality inspection problem classification. The "dialog" content can be replaced with actual data to be inspected, allowing for the seamless integration of the data with the target quality inspection prompt template. The structure "##requirements(...)" specifies the quality inspection task requirements, clearly defining a series of requirements related to the quality inspection task of the online consultation data for the large model. For example, the structure "##requirements(...)" specifies that the large model outputs results according to the output format example, and also specifies other requirements to be followed in the quality inspection process.

[0115] In the aforementioned target quality inspection prompt template applicable to online medical platforms, the structure "##Quality Inspection Problem Classification Content(...)" clearly defines the potential problems that may arise on the online medical platform in the quality inspection scenario of online consultation data, the specific evaluation rules for each type of problem, and the exemption conditions set for each type of problem. Here, "id" represents the problem ID, such as "8050, 8060, 8070, 8801, and 9021," which are specific problem IDs. Each problem ID can be bound to a predefined quality inspection score. For example, the predefined quality inspection score bound to the problem ID "8050" is 40 points, the predefined quality inspection score bound to the problem ID "8060" is 20 points, and the predefined quality inspection score bound to the problem ID "8070" is 30 points, and so on. The problem types in the structure "##Quality Inspection Problem Classification Content(...)" can list the problem categories corresponding to the problem IDs. It is understood that different problem IDs require different problem types to be configured. For example, the issue ID "8050" corresponds to the specific issue type "AI routine medical orders not disclosed". Besides the issue IDs and types listed in the examples above, they can be expanded according to the platform's actual quality inspection needs. For example, an issue ID "8071" can be added, with the corresponding issue type being "AI medication orders pose risks," etc. The issue definition description in the "##Quality Inspection Issue Classification Content(...)" structure can provide specific definitions for the corresponding issue types, with each issue type corresponding to one issue definition description. For example, the issue definition description for issue ID "8050" is specifically "The doctor did not provide recommendations for examinations / treatments that meet the patient's current condition; for chronic diseases where clinical guidelines clearly indicate the need for lifestyle intervention, the doctor did not provide lifestyle recommendations." The "This item will not be judged under the following situations and scenarios" section in the "##Quality Inspection Issue Classification Content(...)" structure indicates exemption conditions; different issue types can be configured with one or more exemption conditions. For example, the "8050" issue ID corresponds to three exemption conditions: "1. If the patient's condition does not require examination / treatment / medication, this item will not be judged; 2. If a diagnosis cannot be reached based on the doctor-patient dialogue information, this item will not be judged; 3. If the patient requests or the doctor voluntarily cancels the consultation and requests a refund, this item will not be judged." Similarly, the quality inspection issue classification content of other issue IDs is similar to the structure of the "8050" issue ID above, and will not be repeated.

[0116] It should be noted that the quality inspection issue classification content in the above-mentioned target quality inspection prompt template only illustrates a portion of the issue types and corresponding exemption conditions. In practical applications, the quality inspection issue classification content can be enriched and improved according to specific quality inspection needs, listing all possible issue types on the target platform and the corresponding configurable exemption conditions. Regarding the output format specification, in addition to requiring the large model to output data in JSON (JavaScript Object Notation) format, other formats can also be required; this embodiment does not impose any limitations on this.

[0117] In a specific example, suppose the output of the large model is “{"items":[{"id":"8050","level":"High","reason":"The doctor did not provide suggestions for examination / treatment that meet the patient's current condition"},{"id":"8060","level":"Low","reason":"Based on the complete doctor-patient dialogue, the doctor newly discovered that the patient has or is suspected of having an acute, critical, or severe illness, but did not remind the patient to seek medical attention at the hospital"}]}”. Based on this output, it can be confirmed that the problem ID list includes two problem IDs, "8050" and "8060". Assuming that the predefined quality control score bound to "8050" is 40 points and the predefined quality control score bound to the problem ID "8060" is 20 points, then the first quality control data of the data to be inspected includes a quality control score of 60 points. If the second quality control data output by other target association models includes the problem ID "8060", then the second quality control data of the data to be inspected includes a quality control score of 20 points. Accordingly, the total quality inspection score for the data to be inspected is 80 points. If the quality inspection score threshold is 20 points, then the data to be inspected can be selected as the first data to be reviewed, and this first data to be reviewed will be sent to the quality inspection doctor for manual review.

[0118] In the above technical solution, a large-scale model is used to conduct preliminary quality checks on all doctor services. Continuous fine-tuning of the large-scale model prompt improves the accuracy of the quality checks, thereby increasing the accuracy of the model's returns and reducing the risk of missed detections. Simultaneously, the high efficiency of the large-scale model enhances the timeliness of the quality checks service. Supplementing this with doctor-patient dialogues, prescription information, and telephone recordings transcribed into text provides a comprehensive approach to risk mitigation. Risk control strategies are used to identify scenarios with potential medical risks, such as the prescription drug red line strategy: in consultation scenarios where doctors provide incorrect medication guidance, this strategy identifies prescription drugs and dosage instructions in doctor-patient dialogues and pushes them into the quality check process. Quality inspectors determine if there are any issues, and by guiding and educating doctors and following up with patients, the risk of mistreatment due to incorrect medication advice is reduced. After combining the large-scale model with risk control strategies, quality inspection experts conduct a second review based on the initial check, avoiding the risk of illusions from the large-scale model and further improving the accuracy of the quality checks. This initial screening filters out most services without problems, significantly reducing the manpower costs of quality inspection experts. Compared to traditional operational spot checks / full-volume inspections, the introduction of AI big data models has significantly reduced labor costs. The big data model, combined with other model quality inspection strategies, ensures 100% initial screening and quality inspection of online consultation orders, greatly improving the quality of medical services while protecting the experience of patients on the platform.

[0119] The aforementioned technical solution enables AI-driven quality inspection of doctor consultation data. By leveraging cutting-edge large-scale model technology, integrating powerful large-scale models, intelligent strategies, and precise intervention through human rules, it overcomes the problem of misjudgments caused by the illusion of large-scale models. This allows for preliminary quality inspection of a large volume of online consultation services, identifying substandard services, thereby reducing the workload of quality inspection experts, lowering human costs, improving the timeliness of quality inspection, and effectively ensuring the quality of online consultations. Through continuous feedback from the large-scale model, the prompt structure can be continuously fine-tuned, making the model's quality inspection results more accurate, forming a data flywheel for rapid and precise processing of massive amounts of doctor-patient service quality inspection work. Further integration with manual review allows for analysis and quality verification of doctor services, generating and outputting service quality inspection conclusions. This provides reference and support for various judgment behaviors of doctors and platform operations, not only improving the accuracy and reliability of quality inspection but also promoting the widespread application of large-scale model technology in the field of medical quality inspection, bringing intelligent transformation to the medical industry.

[0120] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information (such as online consultation data, online consultation data, and user customer service dialogue data) in this technical solution comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0121] It should be noted that any arrangement or combination of the technical features in the above embodiments also falls within the protection scope of this disclosure.

[0122] In one example Figure 4 This is a structural diagram of a large-model-based quality inspection device provided in this embodiment. This embodiment is applicable to situations where a target quality inspection prompt, obtained by concatenating a target quality inspection prompt template including quality inspection task descriptions and quality inspection problem classifications, is input into a target large model to perform quality inspection on the data to be inspected on the target platform. This device is implemented through software and / or hardware and is specifically configured in an electronic device. This electronic device can be a terminal device or a server device, as long as it can execute the large-model-based quality inspection method. This embodiment does not limit the specific type of electronic device.

[0123] like Figure 4 The quality inspection device 400 based on a large model shown includes: a data template acquisition module 410, a target quality inspection prompt acquisition module 420, and a first quality inspection data acquisition module 430.

[0124] in,

[0125] The data template acquisition module 410 is used to acquire the data to be inspected and the target quality inspection prompt template of the target platform; wherein, the target quality inspection prompt template includes a description of the quality inspection task and a classification of quality inspection issues;

[0126] The target quality inspection prompt acquisition module 420 is used to splice the data to be inspected and the target quality inspection prompt template to obtain the target quality inspection prompt.

[0127] The first quality inspection data acquisition module 430 is used to input the target quality inspection prompt into the target large model, so that the target large model can automatically perform quality inspection on the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

[0128] This embodiment of the disclosure obtains the data to be inspected from the target platform, and a target quality inspection prompt template including quality inspection task descriptions and quality inspection problem classifications. The data to be inspected and the target quality inspection prompt template are then concatenated to obtain target quality inspection prompts. These prompts are then input into a target large-scale model, which performs automated quality inspection on the data to be inspected based on the target quality inspection prompts, resulting in first quality inspection data. Since the quality inspection task descriptions and quality inspection problem classifications in the target quality inspection prompt template clearly define the quality inspection rules, the target large-scale model can quickly and accurately output the platform's quality inspection results based on the concatenated target quality inspection prompts. This technical solution solves the problems of low efficiency and low accuracy in existing platform quality inspection methods, thereby reducing the manual quality inspection cost of the platform and improving the efficiency and accuracy of platform quality inspection.

[0129] Optionally, the data template acquisition module 410 is further configured to: in response to a quality inspection process triggering instruction initiated by the target platform, acquire raw quality inspection data; filter the raw quality inspection data to obtain the data to be inspected; wherein the data to be inspected includes at least one of text data, image data, card data, and audio / video data.

[0130] Optionally, the data template acquisition module 410 is further configured to: when it is determined that there are multiple quality inspection process triggering instructions initiated by the target platform, select a target quality inspection process triggering instruction from the multiple quality inspection process triggering instructions by means of Redis lock acquisition; and acquire the original quality inspection data in response to the target quality inspection process triggering instruction.

[0131] Optionally, the quality inspection issue classification includes issue ID, issue type, issue definition description, and exemption conditions: wherein, the issue ID is configured with a predefined quality inspection score; the first quality inspection data includes a list of issue IDs, severity, and cause of occurrence.

[0132] Optionally, the above-mentioned device further includes a target quality inspection data acquisition module, used for: inputting the data to be inspected into a target association model, so as to perform automated quality inspection on the data to be inspected through the target association model to obtain second quality inspection data; and generating target quality inspection data based on the first quality inspection data and the second quality inspection data.

[0133] Optionally, the target quality inspection data acquisition module is further configured to: generate a summary problem ID list based on the problem ID list of the first quality inspection data and the problem ID list of the second quality inspection data; calculate a summary quality inspection score based on the predefined quality inspection score of each problem ID included in the summary problem ID list; and use the summary quality inspection score, the severity, and the cause of the hit as the target quality inspection data.

[0134] Optionally, the above-mentioned device further includes a quality inspection data sending module, used for: filtering target quality inspection data whose summarized quality inspection score is greater than or equal to the quality inspection score threshold as first quality inspection data to be reviewed; sending the first quality inspection data to be reviewed to a first reviewer for manual review; and obtaining the second quality inspection data to be reviewed from the first reviewer.

[0135] The second quality inspection data to be reviewed is sent to the second reviewer for manual review.

[0136] Optionally, the review and quality inspection data sending module is also used to: obtain the original review and quality inspection data of the first reviewer; and filter the original review and quality inspection data, including quality inspection complaint feedback data, as the second review and quality inspection data to be reviewed.

[0137] Optionally, the target platform includes at least one of an online medical platform, a consultation platform, and a customer service platform.

[0138] The aforementioned large-model-based quality inspection device can execute the large-model-based quality inspection method provided in any embodiment of this disclosure, and possesses the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the large-model-based quality inspection method provided in any embodiment of this disclosure.

[0139] Since the large-model-based quality inspection device described above is capable of executing the large-model-based quality inspection method in the embodiments of this disclosure, those skilled in the art can understand the specific implementation and various variations of the large-model-based quality inspection device based on the large-model-based quality inspection method described in the embodiments of this disclosure. Therefore, how the large-model-based quality inspection device implements the large-model-based quality inspection method in the embodiments of this disclosure will not be described in detail here. Any device used by those skilled in the art to implement the large-model-based quality inspection method in the embodiments of this disclosure falls within the scope of protection of this disclosure.

[0140] In one example, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0141] Figure 5 A schematic block diagram of an example electronic device 500 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0142] like Figure 5 As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 502 or a computer program loaded from storage unit 508 into random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. Input / output (I / O) interface 505 is also connected to bus 504.

[0143] Multiple components in device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0144] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as quality inspection methods based on large models.

[0145] For example, in some embodiments, the large-model-based quality inspection method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by computing unit 501, one or more steps of the large-model-based quality inspection method described above may be performed. Alternatively, in other embodiments, computing unit 501 may be configured to perform the large-model-based quality inspection method by any other suitable means (e.g., by means of firmware).

[0146] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0147] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0148] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0149] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0150] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0151] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is established by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are hosting products within the cloud computing service ecosystem to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers integrated with blockchain technology.

[0152] This embodiment of the disclosure obtains the data to be inspected from the target platform and a target quality inspection prompt template including quality inspection task descriptions and quality inspection problem classifications. The data to be inspected and the target quality inspection prompt template are then concatenated to obtain the target quality inspection prompt. This target quality inspection prompt is then input into a target large-scale model, which performs automated quality inspection on the data to be inspected based on the target quality inspection prompt, resulting in the first quality inspection data. Since the quality inspection task descriptions and quality inspection problem classifications in the target quality inspection prompt template clearly define the quality inspection rules, the target large-scale model can quickly and accurately output the platform's quality inspection results based on the concatenated target quality inspection prompt. The above technical solution solves the problems of low efficiency and low accuracy in existing platform quality inspection methods, thereby reducing the manual quality inspection cost of the platform and improving the efficiency and accuracy of platform quality inspection.

[0153] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0154] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A quality inspection method based on a large model, comprising: Obtain the data to be inspected and the target quality inspection prompt template from the target platform; wherein, the target quality inspection prompt template includes a description of the quality inspection task and a classification of quality inspection issues; The data to be inspected and the target quality inspection prompt template are combined to obtain the target quality inspection prompt. The target quality inspection prompt is input into the target large model, so that the target large model can automatically perform quality inspection on the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

2. The method according to claim 1, wherein, The acquisition of the quality inspection data of the target platform includes: In response to the quality inspection process trigger command initiated by the target platform, the original quality inspection data is obtained; The original quality inspection data is filtered to obtain the data to be inspected; The data to be inspected includes at least one of text data, image data, card data, and audio / video data.

3. The method according to claim 1, further comprising: If it is determined that there are multiple quality inspection process triggering instructions initiated by the target platform, the target quality inspection process triggering instruction is selected from the multiple quality inspection process triggering instructions by using Redis lock acquisition. The raw quality inspection data is obtained in response to the target quality inspection process trigger instruction.

4. The method according to claim 1, wherein, The quality inspection problem classification includes problem ID, problem type, problem definition description, and exemption conditions: wherein, the problem ID is configured with a predefined quality inspection score; The first quality inspection data includes a list of problem IDs, severity, and cause of occurrence.

5. The method according to claim 4, further comprising: The data to be inspected is input into the target association model so that the data to be inspected is automatically inspected through the target association model to obtain the second quality inspection data; Target quality inspection data is generated based on the first quality inspection data and the second quality inspection data.

6. The method according to claim 5, wherein, The step of generating target quality inspection data based on the first quality inspection data and the second quality inspection data includes: A summary problem ID list is generated based on the problem ID list of the first quality inspection data and the problem ID list of the second quality inspection data; The overall quality inspection score is calculated based on the predefined quality inspection scores of each issue ID included in the overall issue ID list; The aggregated quality inspection score, the severity, and the cause of the hit are used as the target quality inspection data.

7. The method according to claim 6, further comprising: The target quality inspection data whose aggregated quality inspection score is greater than or equal to the quality inspection score threshold are selected as the first quality inspection data to be reviewed; The first quality inspection data to be reviewed is sent to the first reviewer for manual review. Obtain the second quality inspection data to be reviewed from the first reviewer; The second quality inspection data to be reviewed is sent to the second reviewer for manual review.

8. The method according to claim 7, wherein, The step of obtaining the second quality inspection data to be reviewed by the first reviewer includes: Obtain the original review and quality inspection data from the first reviewer; The original quality inspection data, including the quality inspection complaint feedback data, is selected as the second quality inspection data to be reviewed.

9. The method according to any one of claims 4-8, wherein, The target platform includes at least one of an online medical platform, a consultation platform, and a customer service platform.

10. A quality inspection device based on a large model, comprising: The data template acquisition module is used to acquire the data to be inspected and the target quality inspection prompt template of the target platform; wherein, the target quality inspection prompt template includes the quality inspection task description and the quality inspection problem classification content; The target quality inspection prompt acquisition module is used to concatenate the data to be inspected and the target quality inspection prompt template to obtain the target quality inspection prompt. The first quality inspection data acquisition module is used to input the target quality inspection prompt into the target large model, so that the target large model can automatically perform quality inspection on the data to be inspected according to the target quality inspection prompt to obtain the first quality inspection data.

11. The apparatus according to claim 10, wherein, The data template acquisition module is also used for: In response to the quality inspection process trigger command initiated by the target platform, the original quality inspection data is obtained; The original quality inspection data is filtered to obtain the data to be inspected; The data to be inspected includes at least one of text data, image data, card data, and audio / video data.

12. The apparatus according to claim 11, wherein, The data template acquisition module is also used for: If it is determined that there are multiple quality inspection process triggering instructions initiated by the target platform, the target quality inspection process triggering instruction is selected from the multiple quality inspection process triggering instructions by using Redis lock acquisition. The raw quality inspection data is obtained in response to the target quality inspection process trigger instruction.

13. The apparatus according to claim 11, wherein, The quality inspection problem classification includes problem ID, problem type, problem definition description, and exemption conditions: wherein, the problem ID is configured with a predefined quality inspection score; The first quality inspection data includes a list of problem IDs, severity, and cause of occurrence.

14. The apparatus according to claim 13, further comprising a target quality inspection data acquisition module, used for: The data to be inspected is input into the target association model so that the data to be inspected is automatically inspected through the target association model to obtain the second quality inspection data; Target quality inspection data is generated based on the first quality inspection data and the second quality inspection data.

15. The apparatus according to claim 14, wherein, The target quality inspection data acquisition module is also used for: A summary problem ID list is generated based on the problem ID list of the first quality inspection data and the problem ID list of the second quality inspection data; The overall quality inspection score is calculated based on the predefined quality inspection scores of each issue ID included in the overall issue ID list; The aggregated quality inspection score, the severity, and the cause of the hit are used as the target quality inspection data.

16. The apparatus according to claim 15, further comprising a verification quality inspection data sending module, used for: The target quality inspection data whose aggregated quality inspection score is greater than or equal to the quality inspection score threshold are selected as the first quality inspection data to be reviewed; The first quality inspection data to be reviewed is sent to the first reviewer for manual review. Obtain the second quality inspection data to be reviewed from the first reviewer; The second quality inspection data to be reviewed is sent to the second reviewer for manual review.

17. The apparatus according to claim 16, wherein, The review and quality inspection data sending module is also used for: Obtain the original review and quality inspection data from the first reviewer; The original quality inspection data, including the quality inspection complaint feedback data, is selected as the second quality inspection data to be reviewed.

18. The apparatus according to any one of claims 13-17, wherein, The target platform includes at least one of an online medical platform, a consultation platform, and a customer service platform.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the large-model-based quality inspection method according to any one of claims 1-9.

20. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the quality inspection method based on a large model as described in any one of claims 1-9.

21. A computer program product comprising a computer program / instructions, wherein, When the computer program / instruction is executed by the processor, it implements the quality inspection method based on a large model as described in any one of claims 1-9.

Images