Quality inspection question answering method based on human-computer interaction and electronic device

Abstract

This application provides a quality inspection Q&A method and electronic device based on human-computer interaction. The method includes: during the quality inspection of a target product, receiving initial questions from a user via a chatbot; wherein the chatbot includes multiple quality inspection Q&A modules, each corresponding to a quality inspection item; based on the initial questions, determining at least one target quality inspection Q&A module that matches the initial questions; and based on preset judgment rules in the target quality inspection Q&A module, determining the judgment result of the corresponding quality inspection item through multiple rounds of dialogue. This method aims to improve the efficiency and accuracy of quality inspection.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a quality inspection Q&A method and electronic device based on human-computer interaction. Background Technology

[0002] In the second-hand electronics recycling business, on-site engineers are required to independently complete the entire process of product quality inspection at the user's location. The inspection results will directly serve as the basis for determining the recycling price, and their accuracy and efficiency directly affect the user experience and the company's operating efficiency.

[0003] However, the standards involved in the quality inspection process are complex, covering multiple dimensions such as hardware status, software function, and appearance integrity. More importantly, the judgment logic of most quality inspection items is complex and usually requires verification of multiple conditions. Engineers find it difficult to accurately grasp all the quality inspection rules in a short period of time, resulting in low quality inspection efficiency and accuracy. Summary of the Invention

[0004] The quality inspection Q&A method and electronic device based on human-computer interaction provided in this application are used to improve the efficiency and accuracy of quality inspection.

[0005] In a first aspect, embodiments of this application provide a quality inspection Q&A method based on human-computer interaction, including:

[0006] During the quality inspection of the target product, the system receives the first round of questions from the user via a chatbot; the chatbot includes multiple quality inspection question-and-answer modules, each corresponding to a quality inspection item.

[0007] Based on the information from the first round of questions, at least one target quality inspection question-and-answer module that matches the information from the first round of questions is identified;

[0008] Based on the preset judgment rules in the target quality inspection question and answer module, the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module is determined through multi-round dialogue.

[0009] In one possible implementation, determining at least one target quality inspection question-answering module that matches the first-round question information based on the first-round question information includes:

[0010] The intent recognition process is performed on the first round of questioning information to obtain the target quality inspection intent;

[0011] Based on the target quality inspection intent and the preset mapping relationship between the quality inspection intent and the quality inspection question and answer module, at least one target quality inspection question and answer module that matches the first round of question information is determined.

[0012] In one possible implementation, the step of performing intent recognition processing on the first round of questioning information to obtain the target quality inspection intent includes:

[0013] The first round of question information is matched using at least one of regular expression matching, keyword matching, and semantic similarity matching to determine the target quality inspection intent.

[0014] In one possible implementation, the judgment rule is a tree-structured decision rule; the step of determining the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module through multi-round dialogue based on the preset judgment rule in the target quality inspection question and answer module includes:

[0015] For any of the target quality inspection question and answer modules, the user is guided to supplement and submit the multimodal condition information required for the judgment through a multi-turn dialogue, according to the node logical order of the tree-shaped decision rules in the target quality inspection question and answer module.

[0016] Based on the multimodal conditional information, the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module is determined.

[0017] In one possible implementation, the step of determining the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module through multi-round dialogue based on the judgment rules preset in the target quality inspection question and answer module further includes:

[0018] In response to the presence of multiple target quality inspection question and answer modules, the judgment order of the corresponding quality inspection items for each target quality inspection question and answer module is determined;

[0019] Based on the aforementioned judgment order, the judgment results of the corresponding quality inspection items of each target quality inspection question and answer module are determined sequentially based on the preset judgment rules in each target quality inspection question and answer module through a multi-round dialogue.

[0020] In one possible implementation, the quality inspection question-and-answer module is generated by configuring exclusive system prompts for the multimodal visual language model. The system prompts include the judgment rules for the corresponding quality inspection items of the quality inspection question-and-answer module, as well as the image recognition area and feature requirements for the quality inspection items.

[0021] In one possible implementation, the quality inspection item is the determination of missing International Mobile Equipment Identity (IMEI) serial numbers, and the preset judgment rules in the target quality inspection question-and-answer module corresponding to the quality inspection item include:

[0022] Guide users to provide their local page and determine if the local page is working correctly;

[0023] If the local page is normal, it is determined that there is no IMEI serial number missing anomaly;

[0024] If the local page is abnormal, the user is guided to provide the dial page and it is determined whether the dial page is abnormal.

[0025] If the dialing page is normal, it is determined that there is no IMEI serial number missing anomaly.

[0026] If the dialing page is abnormal, it is determined that there is an IMEI serial number missing error, and the user is guided to provide the user identification SIM card reading status information;

[0027] If the card reading status information indicates that the card reading is normal, then it is determined that there is an IMEI serial number missing anomaly; if the card reading status information indicates that the card reading is abnormal, then it is determined that there is both an IMEI serial number missing anomaly and a SIM card anomaly.

[0028] In one possible implementation, the quality inspection item is a determination of discrepancies between the official website information and the target quality inspection Q&A module corresponding to the quality inspection item has preset judgment rules including:

[0029] Query the preset database to see if the configuration information for the target product exists;

[0030] If the configuration information of the target product exists in the database, and the configuration information is consistent with the local machine information provided by the user, it is determined that there is no official website information discrepancy; if the configuration information is inconsistent with the local machine information, it is determined that there is an official website information discrepancy.

[0031] If the database does not contain configuration information for the target product, determine whether the target product supports online querying.

[0032] If the target product does not support official website search, it is determined that there is no official website information discrepancy; if the target product supports official website search, the user is guided to search the official website.

[0033] If the official website does not contain the configuration information for the target product, it is determined that there is an error due to mismatch between the official website information and the target product information.

[0034] If the configuration information of the target product exists on the official website and the model is consistent with the local machine, then if the configuration information does not display memory and storage information, it is determined that there is no official website information mismatch anomaly; if the configuration information displays memory and storage information and the memory and storage information is consistent with the local machine, it is determined that there is no official website information mismatch anomaly; if the configuration information displays memory and storage information but the memory and storage information is inconsistent with the local machine, it is determined that there is an official website information mismatch anomaly.

[0035] In one possible implementation, the quality inspection item is a new machine assessment, and the preset assessment rules in the target quality inspection question and answer module corresponding to the quality inspection item include:

[0036] Guide users to replenish the packaging of the target product;

[0037] If the packaging is opened, the device is considered not brand new.

[0038] If the packaging is unopened and the purchase channel is a preset channel, then guide the user to provide the packaging box, product label, and device page of the target product, and verify whether the IMEI serial numbers in the packaging box, product label, and device page are consistent.

[0039] If the IMEI serial number is inconsistent, the device is determined to be not brand new;

[0040] If the IMEI serial numbers match, for preset brand models that support official website queries, verify whether the retention period of the target product displayed on the official website is greater than the preset threshold.

[0041] If the retention period of the target product is greater than a preset threshold, it is determined to be a non-brand new product; if the retention period of the target product is less than or equal to the preset threshold, it is determined to be a brand new product.

[0042] In one possible implementation, the method further includes:

[0043] Based on the judgment results of the target quality inspection items, the recycling price of the target product is determined.

[0044] Secondly, embodiments of this application provide a quality inspection Q&A device based on human-computer interaction, comprising:

[0045] The receiving module is used to receive the first round of questions input by the user based on the chatbot during the quality inspection process of the target product; wherein, the chatbot includes multiple quality inspection question and answer modules, and the quality inspection question and answer modules correspond one-to-one with the quality inspection items;

[0046] The determination module is used to determine at least one target quality inspection question-and-answer module that matches the first round of question information based on the first round of question information;

[0047] The judgment module is used to determine the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module through a multi-round dialogue based on the judgment rules preset in the target quality inspection question and answer module.

[0048] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0049] The memory stores computer-executed instructions;

[0050] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0051] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0052] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0053] The human-computer interaction-based quality inspection Q&A method and electronic device provided in this application receive the first round of questions input by the user during the quality inspection of the target product, matches at least one dedicated quality inspection Q&A module corresponding to the first round of questions, and then the module automatically generates standardized quality inspection judgment results through multi-round dialogue according to its built-in standardized judgment rules. This method eliminates the need for engineers to memorize thousands of complex and constantly updated quality inspection standards, completely freeing them from reliance on human experience and memory, significantly shortening the quality inspection time for a single product, and improving overall quality inspection efficiency. At the same time, the system strictly follows official standards for guidance and judgment, effectively avoiding omissions and misjudgments that are prone to occur in manual judgment, significantly improving the accuracy and consistency of quality inspection results, simplifying on-site operation procedures, and comprehensively improving the user experience and service quality in on-site quality inspection scenarios. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0055] Figure 1 A flowchart illustrating the human-computer interaction-based quality inspection Q&A method provided in this application. Figure 1 ;

[0056] Figure 2 A schematic diagram illustrating a determination rule for missing International Mobile Equipment Identity (IMEI) serial numbers provided in this application;

[0057] Figure 3 A schematic diagram illustrating a rule for determining discrepancies between official website information and the information provided in this application;

[0058] Figure 4 A flowchart illustrating the human-computer interaction-based quality inspection Q&A method provided in this application. Figure 2 ;

[0059] Figure 5 A schematic diagram illustrating one application scenario provided in this application;

[0060] Figure 6A code diagram illustrating regular expression matching provided in this application;

[0061] Figure 7 A schematic diagram of a system prompt word provided in this application;

[0062] Figure 8 A schematic diagram illustrating another application scenario provided by this application;

[0063] Figure 9 A schematic diagram illustrating one application scenario provided in this application;

[0064] Figure 10 A schematic diagram of the structure of the quality inspection Q&A device based on human-computer interaction provided in this application;

[0065] Figure 11 A schematic diagram of the structure of the electronic device provided in this application.

[0066] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0067] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0068] In business scenarios such as door-to-door recycling of used electronic products, in-store inspection, and centralized quality inspection centers, it is usually necessary to determine the condition and evaluate the price based on detailed rules. After receiving the target product, engineers often need to perform operations such as powering it on, checking system information, testing functional status, taking pictures of the device's appearance or capturing screenshots of the interface, and quickly determine whether the target product has missing serial numbers, storage discrepancies, functional abnormalities, or cosmetic damage based on the judgment criteria for the corresponding quality inspection items. However, the judgment criteria involved in the quality inspection process are complex, covering multiple dimensions such as hardware status, software functionality, and cosmetic integrity. More importantly, the judgment logic for most quality inspection items is complex, usually requiring verification based on multiple conditions. Engineers find it difficult to accurately grasp all the quality inspection rules in a short time, resulting in low quality inspection efficiency and accuracy.

[0069] Therefore, the inventors conceived of introducing a conversational robot capable of real-time interaction with humans into the quality inspection process. Instead of using a uniform, universal model to handle all quality inspection issues, they devised a dedicated quality inspection question-and-answer module for each quality inspection item. During on-site product quality inspection, the robot interacts with engineers in multiple rounds based on the dedicated question-and-answer module that matches the information from the first round of questions. It automatically outputs standardized quality inspection judgment results according to the preset judgment rules in the dedicated quality inspection question-and-answer module, eliminating the need for engineers to memorize complex quality inspection rules. This avoids misjudgments and omissions caused by insufficient human experience and simplifies the on-site quality inspection judgment process, thereby effectively improving the execution efficiency and judgment accuracy of quality inspection work in the context of used product recycling.

[0070] The execution subject of this application embodiment can be an electronic device with processing capabilities, such as a mobile terminal, computer, server, cloud computing node, etc.

[0071] In some implementations, the execution entity of this application embodiment can be a quality inspection and Q&A system, which includes a server device and a client device. The client device is used to receive the first round of questions input by the user, display the multi-round dialogue interaction content, and output the quality inspection item judgment result. The server device is used to execute the core processing steps involved in this application embodiment, such as intent matching, quality inspection and Q&A module scheduling, multi-round interaction logic control, multi-modal information processing, and quality inspection item judgment result generation.

[0072] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0073] Figure 1 A flowchart illustrating the human-computer interaction-based quality inspection Q&A method provided in this application. Figure 1 ,like Figure 1 As shown, the method includes:

[0074] S101. During the quality inspection of the target product, the first round of questions input by the user is received based on the chatbot.

[0075] The chatbot includes multiple quality inspection question-and-answer modules, with each module corresponding to a quality inspection item.

[0076] In this embodiment of the application, this step corresponds to the entry processing stage of the entire Q&A process. The executing entity can be a quality inspection Q&A system deployed on a mobile terminal, store operation terminal, quality inspection workstation or cloud server, or a distributed processing system composed of the aforementioned devices.

[0077] For example, the target product can be a mobile phone, tablet computer, smart wearable device, or other electronic product with electronic functions and recyclable value.

[0078] The quality inspection process refers to the process of checking and verifying the various states of a target product according to preset standards. This quality inspection process can occur in scenarios such as door-to-door collection, store inspection, warehouse re-inspection, or centralized quality inspection center.

[0079] A chatbot is an intelligent system that can interact with users in natural language and provide professional answers. Its interaction medium can be a graphical interface, a chat window, a voice interaction interface, or an embedded quality inspection page.

[0080] Quality inspection items refer to the specific inspection objects that need to be judged separately in the recycling quality inspection process, covering multiple dimensions such as hardware configuration, software function, appearance condition, and identity legality. Examples include missing International Mobile Equipment Identity (IMEI) serial number, whether the storage capacity matches, small device model lookup, screen display status, camera function status, battery health status, appearance wear and tear, and account lock status.

[0081] The quality inspection question and answer module refers to a rule-based question and answer processing unit built around a single quality inspection item. Each module stores the judgment conditions, follow-up question logic, input requirements, evidence types, and conclusion generation rules for the corresponding quality inspection item, thus enabling continuous interaction to be organized for specific quality inspection items.

[0082] The initial question information refers to the first inquiry submitted by a user regarding a quality inspection issue. It can be natural language text, such as "How do I determine if the capacity displayed on this machine is inconsistent with the outer box?", or it can be images, screenshots, speech-to-text conversion, or structured device information collected by the client device.

[0083] Specifically, in one possible embodiment, when inspecting a target product, the engineer opens the quality inspection page through a work terminal (client device) and enters the first round of questions. The terminal sends the user account, target product identifier, current work order number, equipment model, and question content to the server device equipped with a chatbot. After receiving the information, the chatbot first performs session initialization processing, generates a session identifier for the current question-and-answer task, and associates and stores this session identifier with the unique number of the target product, the engineer's identity identifier, and the current quality inspection process node, so that subsequent rounds of dialogue can use the same context.

[0084] Optionally, if the user input is voice, the voice recognition module first converts it into text; if the user uploads a screenshot or photo, the image receiving module records the image file path, shooting time, and source page, and can further call optical character recognition capabilities to extract text information from the image; if the user input contains system parameters automatically captured by the terminal, such as device serial number, IMEI, system version, and storage specifications, these parameters can be directly written into the session context as structured fields.

[0085] It should be noted that the first round of questions actually received by the chatbot is not an isolated question, but a composite input object that includes the user's question content, product background information, and operating environment information.

[0086] Optionally, to ensure the accuracy of subsequent judgments, standardization processing can be performed after receiving the first round of questions. This includes text segmentation, synonym unification, typo correction, and industry dictionary matching. For example, expressions such as "incorrect memory," "inconsistent capacity," and "different storage and labels" can be uniformly mapped to candidate quality inspection items related to storage specifications. Images can undergo clarity detection, orientation correction, and key area cropping. Structured parameters can be validated for field legality. If the first round of questions is too vague, such as only "how to judge this," the system can infer the range of quality inspection items most likely associated with it by combining the current workflow node, the most recent operation page, and equipment inspection logs, and retain multiple candidate directions in the current session. Based on the above processing methods, the process of receiving the first round of questions not only completes data collection but also establishes the contextual foundation required for subsequent rule-driven question answering. This allows subsequent module routing to no longer rely solely on keyword matching but to make judgments based on a comprehensive consideration of text, images, and business status, thereby reducing misunderstandings caused by incomplete expressions, arbitrary input, or scattered evidence during on-site quality inspections.

[0087] This step involves directly receiving the initial questions submitted by users at the quality inspection site and performing multi-source aggregation and contextual processing on the input content. This enables the system to enter the judgment chain from the very beginning by combining the target product, operating environment, and initial evidence. This avoids the limitations of traditional text retrieval-based question answering, which only returns static rule descriptions based on a single question. As a result, a data foundation is established for accurately identifying the attribution of the problem, continuously inquiring about missing conditions, and outputting actionable judgment results.

[0088] S102. Based on the information from the first round of questions, identify at least one target quality inspection question-and-answer module that matches the information from the first round of questions.

[0089] In this embodiment of the application, this step can locate the target quality inspection question and answer module that the current session should enter from multiple independent quality inspection question and answer modules, so that subsequent interactions revolve around a single target quality inspection item.

[0090] Each quality inspection question and answer module corresponds one-to-one with a quality inspection item, meaning that each quality inspection item has a specific rule expression, condition set, and question and answer path. Different quality inspection question and answer modules are distinguished from each other in terms of rule definition. For example, quality inspection question and answer module 1 focuses on the consistency judgment between equipment identification code, packaging information, and system information; quality inspection question and answer module 2 focuses on conditions such as bright spots, leakage, color difference, and touch abnormality; and quality inspection question and answer module 3 focuses on the correspondence between system identification capacity, body label, order information, and historical disassembly and repair information.

[0091] The target quality inspection question and answer module refers to the quality inspection question and answer module selected for the current session after identification. It is responsible for subsequent follow-up questions, receiving supplementary evidence, and generating judgment results.

[0092] In practical implementation, multi-dimensional feature extraction can be performed on the initial round of questions. For text content, keywords, phrase entities, question types, negation words, and conditional words are extracted, such as identifying key expressions like "missing," "inconsistent," "display," "blurry photo," and "cannot power on." For image content, image categories, interface tags, text content, and feature regions are extracted, such as identifying the capacity field in the system settings page, the serial number information in the "About Me" page, and crack areas in screen appearance photos. For structured data, device model, brand, system version, detection tool return value, and current operation steps are extracted. Then, based on these features, rule matching, vector similarity calculation, classification model inference, or a combination thereof can be used to determine the target quality inspection question-and-answer module. For example, a module trigger feature set can be pre-configured for each quality inspection question-and-answer module. When the initial round of questions matches a module's trigger feature set with a preset matching threshold, that module is determined as a candidate module. If multiple candidate modules exist, disambiguation sorting is performed based on the current quality inspection process node, the user's recently viewed pages, historical session content, and evidence type, selecting the module with the highest score as the target quality inspection question-and-answer module.

[0093] In one possible embodiment, the system uses a comprehensive evaluation method: "Module Matching Score = Text Semantic Score + Image Evidence Score + Structured Field Matching Score + Process Context Score." The text semantic score measures the semantic similarity between the initial question information and the knowledge descriptions of each quality inspection item. The image evidence score characterizes the degree of conformity between the uploaded image and the evidence template required by the module. The structured field matching score reflects whether the currently collected parameters cover the core judgment fields of a module. The process context score reflects the prior probability of a certain type of quality inspection item when the current session occurs at a specific job node. Through this comprehensive scoring mechanism, the system can reliably determine the target quality inspection question-and-answer module even when user questions are not standardized, the initial input information is incomplete, or multiple quality inspection items overlap. If the difference between the highest and second-highest scores is below a preset range, it indicates ambiguity in the current question's attribution. In this case, the system can generate clarifying follow-up questions, such as "Do we need to determine whether the storage capacity matches the label, or whether the serial number is complete?" The target quality inspection question-and-answer module is then locked after the user makes their selection. If the initial input is sufficient to directly locate the module, no further clarification is needed, and the system can immediately proceed to the subsequent rule-driven question-and-answer stage.

[0094] After routing is complete, the system writes the target quality inspection question-and-answer module identifier into the current session context and simultaneously loads the corresponding rule template, condition list, evidence requirements, and conclusion output format for that module. At the same time, the processing logic of non-target modules can be frozen to avoid cross-module follow-up questions in subsequent rounds of interaction.

[0095] Optionally, when new information added by the user in the current session clearly points to another quality inspection item, the system can also trigger the module switching judgment mechanism to re-evaluate the current module. However, the switching process will retain the collected information to avoid the user from re-entering it.

[0096] Based on the above processing method, the target quality inspection items no longer rely on engineers to accurately remember the rule names or menu paths. Instead, the system automatically identifies and routes them to the corresponding modules based on on-site input, making the quality inspection Q&A more relevant to the work process.

[0097] Based on the above analysis, this step solves the problems of ambiguous first-round questions and difficulty in accurately corresponding to specific rules when multiple quality inspection items overlap in traditional solutions. It converts the information from the first round of questions into multi-dimensional features that can be used for module identification, and determines the target quality inspection question and answer module from multiple quality inspection question and answer modules that correspond one-to-one with the quality inspection items. This enables subsequent dialogues to converge around a single target quality inspection item, reduces the workload of engineers in repeated descriptions and retrievals, and improves the accuracy of the starting point of the question and answer chain in complex rule scenarios.

[0098] S103. Based on the preset judgment rules in the target quality inspection question and answer module, determine the judgment result of the corresponding quality inspection item in the target quality inspection question and answer module through multi-round dialogue.

[0099] In this embodiment, this step collects sufficient multimodal conditional information to support the judgment through multiple rounds of dialogue by the target quality inspection question-and-answer module. It then uses preset judgment rules to comprehensively analyze each condition and outputs the final judgment result for the target quality inspection item. The judgment result can be a binary conclusion, a multi-state conclusion, or a structured result with explanations of reasons. For example, it can output result categories such as "meets standards," "does not meet standards," "requires manual review," and "insufficient evidence to be supplemented," and can simultaneously provide the combination of conditions that triggered the result, a summary of evidence, and processing suggestions.

[0100] For example, a judgment rule refers to a pre-built set of structured rules for a specific quality inspection item. This set of rules not only includes a list of conditions required for the judgment conclusion, but also includes the dependencies, order, exclusivity, priority, evidence type requirements, and abnormal branch handling logic between conditions. For example, for the quality inspection item "storage specifications do not match," the judgment rule may require confirmation of conditions such as order registration capacity, system read capacity, device label capacity, whether there are signs of disassembly or repair, and whether there are modification records, and stipulates that when the system read capacity and label capacity conflict, a screenshot of the device or the results of a third-party testing tool must be provided. For the quality inspection item "screen function abnormality," the judgment rule may require confirmation of whether the device can be turned on first, and then confirmation of whether there are display abnormalities, touch abnormalities, local dead pixels, bright lines, etc., with different confirmation results triggering different follow-up paths.

[0101] Multi-turn dialogue refers to a system that, in a quality inspection question-and-answer task, does not output a fixed answer all at once, but continuously interacts with the user based on the currently known conditions, gradually supplementing the information needed for judgment. The organization of multi-turn dialogue can include text questions, button options, drop-down selections, photo guidance, screenshot prompts, automatic reading, and mixed input. For example, after a user asks "How do I judge if the capacity of this machine is incorrect?", the target quality inspection question-and-answer module can first output "Please confirm the storage capacity displayed in the system settings and upload a screenshot of the About This Machine page." After receiving the screenshot, it extracts the capacity value through image recognition; if the recognition result is unclear, it continues to prompt "Please retake a complete photo of the page containing the capacity field"; after obtaining the system capacity, it asks "Are the capacities marked on the outer box or the machine label consistent?" and allows the user to supplement by text input or taking a photo of the label; if the two are inconsistent, it further asks "Are there any signs of disassembly, repair, or motherboard replacement on the device?" and requires uploading photos of the appearance or repair records. This process continues round by round until the conditions required for the judgment rule are met.

[0102] Specifically, after invoking the target quality inspection question-and-answer module, the system can first extract known condition information from the current session context and compare it with the module's condition list to generate a list of missing conditions. Subsequently, the question-and-answer engine determines the conditions that should be prioritized for collection based on the dependencies in the decision rules. For example, if a subsequent condition is only meaningful if the preceding condition is met, the system will first ask about the preceding condition; if a piece of evidence has a diversionary effect on multiple branches, the system will prioritize guiding the user to upload that evidence.

[0103] In its implementation, the system first performs a condition completeness check to confirm whether all core conditions required by the preset judgment rules of the target quality inspection question-and-answer module have been effectively assigned. If key conditions are still missing, the final judgment result cannot be output at this time. The system can mark the result as "to be supplemented" and continue to ask follow-up questions based on the missing items. If the conditions are met, the rule engine performs calculations according to the judgment rules corresponding to the target quality inspection question-and-answer module. This calculation can be expressed as Boolean logic judgment, condition tree traversal, rule table lookup, scoring calculation, or a combination of the above methods. For example, in the storage specification judgment scenario, when the system capacity field is inconsistent with the tag capacity field and the disassembly and repair record exists, "Storage mismatch" can be output; when the system capacity is consistent with the tag capacity but the order registration value is inconsistent, "Order information abnormal and needs to be reviewed" can be output; when the image recognition credibility is insufficient, "Insufficient evidence and needs to be retaken" can be output.

[0104] Based on the above analysis, this step, by invoking a dedicated question-and-answer module corresponding to the initial question information and using pre-defined judgment rules to drive multiple rounds of dialogue, gradually guides the user to supplement conditional information. This transforms the judgment process for complex quality inspection items from "engineers manually reviewing rules and organizing conditions" to an interactive method where "the system actively asks follow-up questions based on the rules and dynamically converges the path." This approach effectively solves the problems of traditional text-based question-and-answer systems, such as the inability to incorporate image evidence, the difficulty in continuously asking follow-up questions about missing conditions, and the inability to adjust subsequent judgment paths based on confirmed information. Therefore, it improves the completeness of condition collection, the coherence of the interaction process, and the efficiency of on-site quality inspection in complex rule scenarios.

[0105] Optionally, the rule system can also set evidence credibility parameters. This allows the system to comprehensively evaluate results based on confidence level, input source level, and manual verification status. For example, automatically identified screenshots with confidence levels above a preset threshold can be directly considered valid evidence, while results below the threshold are marked as requiring manual verification. These thresholds can be set according to the risk level of different quality inspection items; higher-risk items will have stricter confidence requirements to reduce the probability of misjudgment.

[0106] After determining the judgment result, the system can simultaneously generate a result description, which includes at least the current conclusion, the main conditions that triggered the conclusion, a list of collected evidence, and subsequent processing actions. For example, it might output "The judgment result is a serial number mismatch, based on the system reading an IMEI that does not match the IMEI on the device label, and the packaging label is missing; it is recommended to proceed with the abnormal receipt process"; or "The judgment result is that the screen display is normal, based on successful power-on, no bright lines, no liquid leakage, and the touch test passed." This result can be displayed directly in the dialog window or written into the quality inspection work order, pricing engine, or risk control system for subsequent pricing, review, and archiving.

[0107] In one possible implementation, the system also stores the condition variables, evidence file references, and result labels used in this determination in a structured manner for subsequent audit traceability, model training, or rule optimization. If engineers subsequently provide new evidence, the system can call the target module again to update the determination result and retain both the previous and current versions.

[0108] In this step, multimodal conditional information is collected through multiple rounds of dialogue and uniformly converted into conditional variables required for executable rules. Then, inference is performed according to the preset judgment rules of the target quality inspection question and answer module. The system output is no longer a fuzzy explanation or generalized suggestion, but a clear judgment conclusion that can directly serve the quality inspection decision, realizing a closed-loop transformation from unstructured on-site input to structured quality inspection conclusion.

[0109] This application provides a human-computer interaction-based quality inspection Q&A method. During the quality inspection of a target product, the method receives initial user questions and matches them with at least one dedicated quality inspection Q&A module. This module then automatically generates standardized quality inspection results through multi-round dialogue based on its built-in standardized judgment rules. This approach eliminates the need for engineers to memorize thousands of complex and constantly updated quality inspection standards, completely freeing them from reliance on human experience and memory. This significantly shortens the inspection time for a single product, improves overall quality inspection efficiency, and ensures that the system strictly adheres to official standards for guidance and judgment. This effectively avoids omissions and misjudgments that are prone to occur during manual judgment, significantly improving the accuracy and consistency of inspection results, simplifying on-site operation procedures, and comprehensively enhancing the user experience and service quality in on-site quality inspection scenarios.

[0110] Taking mobile phones as an example, the judgment rules for quality inspection items are explained exemplarily.

[0111] (1), reference Figure 2 As shown, Figure 2This diagram illustrates a judgment rule for determining missing International Mobile Equipment Identity (IMEI) serial numbers (also known as missing serial numbers) provided in this application. The target quality inspection item is the determination of missing IMEI serial numbers, and the preset judgment rule for the target quality inspection item includes:

[0112] Guide users to provide their local page and determine if the local page is working correctly;

[0113] If the local page is normal, it is determined that there is no missing IMEI serial number.

[0114] If the local page is abnormal, guide the user to provide the dial page and determine if the dial page is abnormal;

[0115] If the dialing page is normal, it is determined that there is no IMEI serial number missing anomaly.

[0116] If the dialing page is abnormal, it is determined that there is an IMEI serial number missing error, and the user is guided to provide the user identification SIM card reading status information;

[0117] If the card reading status information indicates that the card reading is normal, it is determined that there is an IMEI serial number missing abnormality; if the card reading status information indicates that the card reading is abnormal, it is determined that there is both an IMEI serial number missing abnormality and a SIM card abnormality.

[0118] The International Mobile Equipment Identity (IMEI) is a 15-digit code that uniquely identifies each mobile phone and is a crucial basis for verifying the phone's legality. The "About This Phone" page, often called the "About This Phone" or "About Phone" page, displays basic device information, including the phone's model, system version, and IMEI serial number. A "normal" page means the page loads and displays all standard information, including the IMEI serial number. An "abnormal" IMEI serial number means the phone can provide a valid IMEI serial number through legitimate channels, meeting the quality inspection standards for normal phone recalls. An "abnormal" page means the "About This Phone" page fails to load, loads normally but does not display the IMEI serial number, or displays an incorrect or incomplete IMEI serial number.

[0119] The dial pad refers to the phone's dial pad interface, where you enter specific codes (such as "..."). This will directly bring up the phone's IMEI serial number display screen. The dialer page should be normal when you enter "..." Afterwards, the IMEI display interface should pop up normally, and the interface should fully display a valid IMEI serial number. The dialing page error refers to entering " "After that, there is no response, or the pop-up interface does not display the IMEI serial number, or the displayed IMEI serial number is in an incorrect or incomplete format. An IMEI serial number missing anomaly means that the phone cannot provide a valid IMEI serial number through the two core channels of the device's internal page and the dialer page, which does not meet the quality inspection standards for normal recycling."

[0120] A Subscriber Identity Module (SIM) card is a smart card used to access mobile communication networks. SIM card reader status information refers to the phone's recognition status of the inserted SIM card, including two situations: normal reader status and abnormal reader status. Normal reader status means the phone can correctly recognize the inserted SIM card, and the operator name and signal strength are displayed at the top of the screen. Abnormal reader status means the phone cannot recognize the inserted SIM card, and the screen displays "No SIM card" or "No service." SIM card abnormality indicates a malfunction in the phone's SIM card reader function, preventing normal recognition of the SIM card. It should be noted that since a missing IMEI number may be related to baseband damage, which often causes the device to fail to recognize the SIM card, the presence of a missing IMEI serial number can be used as a secondary indicator to determine if a SIM card abnormality is present.

[0121] In actual implementation, after the system calls the target quality inspection Q&A module, it first generates the guiding message "Please provide a complete screenshot of the phone's 'About This Phone' page". The engineer operates the phone to enter the "About This Phone" page according to the guiding message, takes a clear photo of the page and uploads it to the system. The target quality inspection Q&A module performs image recognition processing on the uploaded photo, extracts the text information in the page, and determines whether the complete 15-digit IMEI serial number can be recognized, thereby determining whether the phone's page is normal.

[0122] If a complete and correctly formatted 15-digit IMEI serial number is successfully extracted, the device page is deemed normal, and a final judgment result of "No missing IMEI serial number" is generated, along with the corresponding quality inspection standard clauses, thus completing this quality inspection Q&A process.

[0123] If the uploaded "About This Phone" page photo does not display the IMEI serial number, the page is deemed abnormal. A prompt message is then generated: "Please enter '#06#' in the phone's dial pad and upload a screenshot of the IMEI display page that appears." After the engineer follows the prompt and uploads the corresponding screenshot, the module performs image recognition processing on the screenshot to determine whether a complete IMEI serial number can be recognized, thereby determining whether the dialing page is abnormal.

[0124] If the module successfully identifies the complete 15-digit IMEI serial number from the uploaded screenshot of the dialing page, it determines that the dialing page is normal, and then generates the final judgment result "No IMEI serial number missing anomaly", along with the corresponding quality inspection standard clauses, thus completing this quality inspection Q&A process.

[0125] If the module detects that the uploaded dialer page screenshot does not display any IMEI-related information, it determines that the dialer page is abnormal. First, it generates a preliminary judgment result: "Preliminary judgment indicates that there is an abnormality of missing IMEI serial number." Then, it generates a guiding message: "Please insert a working SIM card to confirm whether the phone can recognize the SIM card and display a signal." The engineer is then guided to provide additional SIM card reading status information.

[0126] If the engineer confirms via text input that the phone can display a signal normally after inserting the SIM card, the module determines that the card reading is normal and generates a final judgment result of "IMEI serial number missing anomaly," along with key judgment criteria and corresponding quality inspection standard clauses. If the engineer confirms via text input that the phone displays "No Service" after inserting the SIM card, the module determines that the card reading is abnormal and generates a final judgment result of "both IMEI serial number missing anomaly and SIM card anomaly exist," listing the key judgment criteria and corresponding quality inspection standard clauses for each anomaly.

[0127] This judgment rule guides users to gradually supplement information by judging the device's internal interface, the dialing interface, and the SIM card reader status in a progressive order. Based on the result of each step, corresponding branch logic is executed to arrive at an accurate judgment. This approach transforms the originally complex, multi-condition tree-like judgment process into a standardized, step-by-step guided process. The system strictly adheres to official quality inspection standards for each judgment node, completely avoiding logical confusion and omissions that easily occur during manual judgment. This significantly improves the accuracy and consistency of IMEI serial number missing determination. At the same time, the clear step-by-step guidance reduces the operational difficulty for engineers and improves the efficiency of on-site quality inspection.

[0128] (2), reference Figure 3 As shown, Figure 3 This is a schematic diagram illustrating a judgment rule for determining discrepancies between official website information and the target quality inspection item provided in this application. The judgment rule for the target quality inspection item includes:

[0129] Query the preset database to see if the target product's configuration information exists (whether the crawler has the information);

[0130] If the configuration information for the target product exists in the database, and the configuration information is consistent with the user-provided local information, then it is determined that there is no discrepancy between the official website information and the local information; if the configuration information is inconsistent with the local information, then it is determined that there is a discrepancy between the official website information and the local information.

[0131] If the configuration information for the target product does not exist in the database, determine whether the target product supports querying from the official website;

[0132] If the target product does not support official website search, then it is determined that there is no discrepancy between the official website information and the target product information; if the target product supports official website search, then the user is guided to search the official website.

[0133] If the configuration information for the target product is not available on the official website, it is determined that there is an information discrepancy issue on the official website.

[0134] If the configuration information for the target product exists on the official website and the device model matches the machine, then if the configuration information does not display memory and storage information, it is determined that there is no discrepancy between the official website information and the configuration information; if the configuration information displays memory and storage information and the memory and storage information matches the machine, it is determined that there is no discrepancy between the official website information and the configuration information; if the configuration information displays memory and storage information but the memory and storage information does not match the machine, it is determined that there is a discrepancy between the official website information and the configuration information.

[0135] The "Official Website Information Discrepancy Determination" refers to the quality inspection item that verifies whether the actual hardware configuration information of the target product is consistent with the standard configuration information officially released. The pre-set database refers to a massive database of electronic product configuration information collected and stored in advance from legitimate channels such as official websites and product databases of various brands using web crawling technology. This database is regularly updated with configuration data for newly released models from various brands. The configuration information of the target product refers to the standard hardware parameters of the product officially released, including core information such as model name, memory capacity, storage capacity, processor model, and color.

[0136] The user-provided device information refers to the actual hardware configuration information displayed by the target product, provided by the user through methods such as uploading system screenshots or text descriptions. "No discrepancies with official website information" means the actual configuration of the target product matches the official standard configuration and meets the quality inspection standards for normal product recall. "Discrepancies with official website information" means there are differences between the actual configuration of the target product and the official standard configuration, which may indicate modifications or expansions.

[0137] "Supports official website query" means that the brand's official website provides a public service to query product configuration information by serial number; "Does not support official website query" means that the brand does not provide a public serial number query service, or that the product is an old model that has been discontinued for many years and its relevant configuration information has been removed from the official website.

[0138] In actual implementation, after the system calls the target quality inspection Q&A module, it first extracts the serial number information of the target product provided by the user, sends a query request to the preset crawler database, queries the product standard configuration information corresponding to the serial number, and the database returns the query results, informing the system whether the configuration data corresponding to the serial number exists.

[0139] If the preset database returns configuration information corresponding to the serial number, showing that the standard configuration of the product is 128GB of storage capacity, while the screenshot of the user's own system shows an actual storage capacity of 256GB, the module compares the two and finds an inconsistency, thus determining that there is an anomaly of mismatch between official website information and the device's information. If the configuration information returned by the database is completely consistent with the device's information, then it is determined that there is no anomaly of mismatch between official website information and the device's information, and the quality inspection and Q&A process is completed. If the preset database returns that the configuration information corresponding to the serial number does not exist, the module then queries the pre-stored brand support list to determine whether the brand of the target product is in the list of brands supported by the official website query. This list includes mainstream mobile phone brands on the market.

[0140] If the target product is a niche brand mobile phone that is not on the list of brands that support official website queries, the module will immediately determine that there is no discrepancy between the official website information and the target product, and complete the quality inspection and Q&A process. If the target product is a brand A mobile phone that is on the list of brands that support official website queries, the module will generate a guiding message: "Please go to the official website of brand A, find the 'Serial Number Query' entry under the 'Service and Support' section, enter the mobile phone serial number and upload a screenshot of the query result." After the engineer follows the instructions and uploads the screenshot of the official website query result, the module will perform image recognition processing on the screenshot to determine whether the configuration information corresponding to the serial number exists on the official website.

[0141] If the module recognizes the uploaded screenshot of the official website query results and finds that the official website displays "No relevant information found for this serial number," it determines that the official website does not contain configuration information for the target product, and then generates a final judgment result of "Official website information mismatch anomaly," along with the corresponding quality inspection standard clauses. If the module recognizes that the official website query results show the product model as "Brand A60," which matches the user's provided device model information, it further checks whether the official configuration information includes relevant information on memory and storage capacity. If the module recognizes that the official website query results only display the product model name and warranty information, without displaying specific memory and storage capacity information, it determines that there is no official website information mismatch anomaly, and completes this quality inspection Q&A process. If the module recognizes that the official configuration information displayed on the official website is "8GB memory + 256GB storage," while the screenshot of the user's device system shows an actual configuration of "8GB memory + 128GB storage," and a comparison reveals a discrepancy in storage capacity, it determines that there is an official website information mismatch anomaly; if the memory and storage information displayed on the official website is completely consistent with the device information, it determines that there is no official website information mismatch anomaly, and completes this quality inspection Q&A process.

[0142] This judgment rule first performs a rapid preliminary verification by querying a preset database. Then, for cases where the database lacks information, it differentiates processing based on brand. For products that support official website queries, it guides users to obtain official configuration information and performs layer-by-layer comparison and verification in the order of model consistency, memory storage information integrity, and memory storage information consistency, ultimately arriving at an accurate judgment result. This method significantly improves the judgment efficiency for mainstream popular models by utilizing preliminary verification with a preset database, avoiding unnecessary official website query steps. The layered and progressive judgment logic fully covers all possible business scenarios. Strict comparison based on official authoritative configuration information ensures the accuracy and credibility of the judgment results. At the same time, clear step-by-step operation guidance reduces the operational difficulty for engineers, effectively solving the problems of cumbersome and error-prone manual official website query processes, and comprehensively improving the processing efficiency and standardization of quality inspection items with inconsistent official website information.

[0143] (3) The target quality inspection item is for the new machine. The preset judgment rules for the target quality inspection item include:

[0144] Guide users to replenish the packaging of the target product;

[0145] If the packaging is opened, the device is considered not brand new.

[0146] If the packaging is unopened and the purchase channel is the preset channel, guide the user to provide the packaging box, product label and device page of the target product, and verify whether the IMEI serial numbers in the packaging box, product label and device page are consistent;

[0147] If the IMEI serial number is inconsistent, the device is determined to be not brand new;

[0148] If the IMEI serial numbers match, for preset brand models that support official website queries, verify whether the retention period of the target product displayed on the official website is greater than the preset threshold.

[0149] If the retention period of the target product is greater than the preset threshold, it is determined to be a non-brand new machine; if the retention period of the target product is less than or equal to the preset threshold, it is determined to be a brand new machine.

[0150] Among these criteria, "brand new" refers to the quality inspection item that verifies whether the target product is an unused, unactivated, original, genuine new device. The result of this determination directly affects the product's maximum buyback price. Packaging refers to the external packaging carrier that reflects the target product's factory packaging condition, including the outer box seal, tamper-evident seal, and the closure status of the box lid. The system can identify this through user-uploaded images, video frames, or text descriptions. "Unopened packaging" means that the original factory seal on the product's outer packaging is intact and undamaged, not torn or re-attached, and the outer packaging shows no obvious signs of compression, damage, or tampering. "Not brand new" refers to products whose outer packaging has been opened or that have been activated and used, thus not meeting the standards for brand new device buyback. Packaging box labels refer to the labels printed on the outer packaging box of the product, which usually contain information such as product model, IMEI serial number, and production date. Product body labels refer to the labels affixed to the back of the product body or inside the battery compartment, which also contain the product's IMEI serial number information. The IMEI serial number is the unique identifier of a mobile terminal. The IMEI on the packaging box, product label, and device page is usually located on the outer packaging label, device nameplate, or system information page, respectively. The system can complete the comparison through a combination of character recognition and manual confirmation. The purchase channel is a preset channel, indicating that the target product comes from a pre-configured trusted sales channel. The channel can be an official direct-sale online store, authorized distribution channel, or cooperative recycling registration channel. The system makes this judgment based on order information, channel identification, or user declaration. The official website retention period refers to the remaining warranty period or the effective period after activation displayed in the product's official website search results. The preset threshold can be configured as a number of days according to business rules.

[0151] In actual implementation, after the system calls the target quality inspection Q&A module, it first generates a guiding message: "Please take photos of all the seals on the phone's outer packaging to confirm whether the original factory seals are intact." Engineers take photos of the top, bottom, and side seals of the phone's outer packaging and upload them to the system. The system performs integrity recognition on the packaging images. If it detects damaged sealing film, opened seals, or signs of secondary opening and closing of the box, it directly outputs a conclusion that the phone is not brand new. If the packaging is not opened, it further verifies the purchase channel information. If the channel meets the preset channel conditions, it obtains the IMEI serial number from the packaging box, product label, and the phone's internal page, and compares the consistency of the serial numbers from the three sources. If any identification result is inconsistent with the other two, the system outputs a conclusion that the phone is not brand new. If the three codes match, the system then initiates an official website query request for the preset brand model that supports official website queries, reads the retention period information returned by the official website, and compares the retention period with a preset threshold. When the retention period is greater than the threshold, the system determines that the target product has undergone circulation or activation beyond the permitted scope of business, and thus outputs a conclusion that it is not a brand new device. When the retention period is less than or equal to the threshold, the system outputs a conclusion that it is a brand new device. For different brand models, the system can preset corresponding official website query addresses, query fields, and parsing rules to improve recognition stability. In practical applications, this component can also select other models, which is not limited in this embodiment.

[0152] This judgment rule employs a progressive process: first verifying the sealing of the product's outer packaging; then verifying the consistency of the three codes (code, serial number, and QR code) for products with unopened packaging and purchased from legitimate channels; and finally verifying the official website retention period for mainstream brand products with matching codes. This step-by-step process gradually eliminates various situations where the product is not brand new, ultimately arriving at an accurate judgment result. This method constructs a multi-dimensional, multi-layered brand new product verification system, cross-checking the three core dimensions of packaging integrity, identity consistency, and manufacturing date. This effectively avoids the misjudgment problems that can easily occur with single verification methods. The strict progressive judgment logic comprehensively covers all key aspects of brand new product judgment. The system strictly follows official standards to execute each verification step, completely avoiding problems such as inconsistent standards and empirical errors that can easily occur with manual judgment. This significantly improves the accuracy and credibility of brand new product judgment. At the same time, clear step-by-step operation instructions reduce the operational difficulty for engineers, improve the efficiency of on-site quality inspection, and ensure the reasonableness and fairness of brand new product buyback prices.

[0153] Figure 4 A flowchart illustrating the human-computer interaction-based quality inspection Q&A method provided in this application. Figure 2 , Figure 5 A schematic diagram illustrating an application scenario provided in this application, such as... Figure 4 and Figure 5 As shown, in this embodiment... Figure 1Based on the examples, a detailed description of the quality inspection Q&A method based on human-computer interaction is provided. This method includes:

[0154] S201. During the quality inspection of the target product, the first round of questions input by the user is received based on the chatbot.

[0155] It should be noted that this step is similar to the aforementioned step S101, and will not be repeated here.

[0156] S202. Perform intent recognition processing on the first round of questioning information to obtain the target quality inspection intent.

[0157] For example, the target quality inspection intent is the intent label obtained after identifying the equipment type and semantically classifying the information in the first round of questions. It is used to characterize the quality inspection concern that the engineer currently wants to query, such as appearance damage, system information verification, or functional abnormality confirmation.

[0158] Understandably, the quality inspection Q&A system has various quality inspection intentions corresponding to different device types, as well as corresponding Q&A modules for each intention. In other words, the quality inspection intentions and Q&A modules differ for different device types such as mobile phones, cameras, and tablets.

[0159] This quality inspection intent can be generated based on clustering of massive historical quality inspection question data. Taking mobile phones as an example, this intent can cover various common quality inspection issue types, such as missing IMEI serial number determination, memory storage specification mismatch determination, and mobile phone model lookup. Specifically, a missing IMEI serial number means that the phone's IMEI number cannot be found in the phone information; a phone without an IMEI number may have a baseband failure (inability to read the SIM card). A memory storage specification mismatch means that the phone's internal memory / storage information does not match the information found on the official website, indicating that the user may have privately expanded the phone's storage. A mobile phone model lookup (specific device model code) means that the on-site engineer needs to fill in the model number of the phone to be recycled; this intent indicates that it is impossible to confirm the model of some less common mobile phones.

[0160] In some possible implementations, the system can first perform word segmentation, noise reduction, and synonym merging on the initial question information, and then combine it with a pre-set intent vocabulary, rule templates, or a trained classification model for matching to output the target quality inspection intent. The classification model can use a lightweight intent classification network, whose input is a vector representation of the question information and whose output is the probability distribution of each quality inspection intent. The system selects the category with the highest probability as the target quality inspection intent. In practical applications, other models can also be selected, and this application embodiment does not limit this.

[0161] Among some possible implementation methods, at least one of regular expression matching, keyword matching, and semantic similarity matching is used to match the first round of question information and determine the target quality inspection intent.

[0162] Among them, regular expression matching is used to identify expressions with fixed formats or stable patterns; keyword matching is used to extract core terms in the quality inspection field from the first round of questions; and semantic similarity matching is used to determine the semantic closeness between user expressions and pre-set intent descriptions, thereby adapting to colloquial, simplified, or synonymous questioning methods.

[0163] In its implementation, after receiving the initial query information, the system first performs word segmentation, noise reduction, and standardization on the text, and then compares the processed text with a pre-defined set of intents. When using regular expression matching, the system can pre-define expression rules corresponding to device serial numbers, model numbers, condition descriptions, and descriptions of functional abnormalities. For example, it can construct patterns for high-frequency quality inspection expressions such as "IMEI," "serial number," "official website query," and "whether it's a brand new device," so as to directly identify the corresponding intent in the question. When using keyword matching, the system can establish a keyword library corresponding to each quality inspection intent. When a term with a high degree of relevance to a certain intent appears in the initial query information, that intent is taken as the target quality inspection intent. When using semantic similarity matching, the system can input the initial query information and the standard descriptions of each quality inspection intent into a vectorized model, calculate the similarity between the two, and determine the target quality inspection intent based on a preset threshold.

[0164] When the system employs multiple matching methods simultaneously, it can first use regular expressions or keyword matching for initial screening, and then perform semantic similarity verification on the candidate intents to reduce misjudgments caused by synonymous or ambiguous expressions. If the matching results of multiple intents all meet the judgment criteria, the final target quality inspection intent can be determined by combining the matching score, rule priority, or historical interaction records, and the recognition result corresponding to that intent can be output.

[0165] For example, Figure 6This application provides a code illustration of regular expression matching, defining a function to determine whether user input text contains intent related to missing International Mobile Equipment Identity (IMEI) serial numbers. This function improves the recall rate of intent recognition by expanding the coverage of negative words. The function takes a string parameter `text`, i.e., the text content entered by the user, and returns a boolean value: `True` if the text matches the intent of missing IMEI, and `False` otherwise. Internally, the function defines a multi-branch regular expression pattern covering various common expression orders, including IMEI keywords followed by a problem description, a problem description followed by IMEI keywords, directly expressing "no IMEI," and IMEI keywords followed directly by a negative expression. The IMEI keyword part can match various spellings such as `imei`, `imei number`, and `imei code`. The problem description part covers all related expressions such as "unknown," "not displayed," "invalid," "unusable," "not found," "not visible," "missing," "absent," "unknown," "not appearing," "not recognized," "abnormal," and "incorrect." The function compiles the regular expression in a case-insensitive mode and supports detailed comments. Then, it converts the user-input text to lowercase before performing regular expression matching, and finally returns the corresponding boolean result based on whether the corresponding content is matched.

[0166] It should be noted that, in order to improve the adaptability, the regular expression rules, keyword library and semantic description templates used can be extended and adjusted according to different brands, models and quality inspection rules. In practical applications, other models or rule configurations can also be selected for this part, and this application embodiment does not limit this.

[0167] Through the above methods, the system can accurately identify user intent even when the initial questions are brief or inconsistent in expression, and map them to the corresponding quality inspection Q&A module. This provides a stable entry point for subsequent rounds of interaction and supplementary judgment information. Because the matching process considers both rule-based and semantic recognition, it improves adaptability to colloquial questions, abbreviated questions, and complex descriptive questions, reduces the misjudgment rate of intent, and enhances response speed and consistency in judgment during the quality inspection and Q&A process for used electronic products.

[0168] S203. Based on the target quality inspection intent and the mapping relationship between the preset quality inspection intent and the quality inspection question and answer module, determine at least one target quality inspection question and answer module that matches the information in the first round of questions.

[0169] For example, the mapping relationship between quality inspection intent and quality inspection Q&A modules can be stored in a backend database or rule configuration table. Each quality inspection intent corresponds to a quality inspection Q&A module identifier. The quality inspection Q&A module has pre-set judgment rules, follow-up question templates, and evidence collection logic that match the intent. After the system identifies the target quality inspection intent, it retrieves the corresponding module identifier based on the mapping relationship and calls the corresponding target quality inspection Q&A module. This allows subsequent rounds of interaction to directly revolve around the current quality inspection item, avoiding module selection bias caused by differences in the initial question wording.

[0170] By first identifying intent and then routing modules, the system can reliably map engineers' natural language questions to specific quality inspection scenarios, ensuring that the target quality inspection question-and-answer module responds and asks follow-up questions within the correct rule domain. This approach reduces the probability of generalized answers, improves the accuracy of locating complex quality inspection items, and ensures that subsequent supplementation of multimodal conditional information is based on the correct quality inspection task, thereby improving overall question-and-answer efficiency and judgment consistency.

[0171] S204. For any target quality inspection question and answer module, according to the node logical order of the tree-shaped decision rules in the target quality inspection question and answer module, guide the user to supplement and submit the multimodal condition information required for the judgment through multi-turn dialogue.

[0172] In this embodiment of the application, this step is not simply to show the rule text to the user, but the target quality inspection question and answer module actively organizes the conversation process according to the preset judgment rules, identifies the currently missing judgment conditions, and completes the collection of judgment basis through round-by-round questioning, evidence prompts, input verification and path convergence, so as to obtain all the condition information that can be used to judge the target quality inspection item.

[0173] For example, multimodal conditional information refers to the judgment criteria from different information formats, including text descriptions, image evidence, screenshots, device reading parameters, user selections, timestamps, operation locations, system logs, etc. For text-based conditions, users can input them directly, or fixed options can be used to reduce ambiguity; for image-based conditions, images can be captured in real time by the terminal camera or selected from the album, and the system will record the image source and perform clarity verification, illumination detection, and key area recognition; for screenshot-based conditions, the terminal can guide the user to jump to a specified system page and prompt them to capture the area containing the target field; for structured device information, it can be automatically read through local detection programs, system interfaces, or external detection tools; for temporal conditions, the timestamp of each round of input can be recorded to determine whether the evidence is generated by the same device and the same detection. After receiving each round of supplementary information, the target quality inspection Q&A module performs condition mapping, that is, converts the original input submitted by the user into condition variables that can be consumed by the rules. For example, it writes "system screenshot is identified as 256GB" into the variable "system capacity = 256GB" and "labeled photo manually confirmed to be 128GB" into the variable "label capacity = 128GB". Then it triggers the rule engine to re-evaluate the currently met and unmet conditions.

[0174] Optionally, to ensure the on-site usability of multi-round dialogues, the system can also be configured with fault tolerance and guidance mechanisms. If a user's answer is inconsistent with the expected rules, or if uploaded evidence cannot be parsed, the system will provide targeted supplementary explanations instead of terminating the session. For example, when a blurry image causes text recognition to fail, the system prompts the user to retake the image and indicates the page location to be photographed; when the user enters "cannot see," the system can switch to a more basic confirmation question or provide example images to assist in the judgment; when a certain condition can be automatically read by the system, the system directly asks the user whether to authorize the reading, reducing manual operation. As the dialogue rounds progress, the target quality inspection question-and-answer module dynamically converges the number of subsequent questions based on the confirmed conditions. For cases where an exclusive conclusion can be reached, irrelevant questions will not be asked, thereby shortening the on-site Q&A time.

[0175] For example, the decision rules in the target quality inspection question-and-answer module can be tree-structured decision rules. Tree-structured decision rules are hierarchical decision models that organize decision conditions according to dependencies. Their nodes can correspond to decision branches such as whether the device is powered on, whether access to the system is possible, or whether a specific identifier exists. In other words, tree-structured decision rules are used to organize the decision conditions of quality inspection items into a hierarchical structure with parent nodes, child nodes, and leaf nodes. After receiving a user's quality inspection inquiry, the system first locates the current decision node according to preset rules, and then generates follow-up questions based on the decision conditions corresponding to that node. The information required for the current node is decomposed into condition items that can be carried by text descriptions, on-site photos, or interface screenshots. The system determines whether the conditions for entering the next level node are met based on the information submitted by the user. If not, it continues to initiate supplementary inquiries around the missing items until the required information for the node is complete.

[0176] In some possible implementations, the system first matches the initial question information with the target quality inspection Q&A module. Then, based on the pre-defined tree-structured decision rules in that target quality inspection Q&A module, it prompts the user with the current node. After the user submits text, photos, or screenshots, the module parses and verifies the submitted content. When it confirms that the conditions for the current node are met, the system marks the node status as passed and moves to the next node; when it confirms that the conditions are not met, it directly outputs the judgment result or exception description corresponding to that node. Because the judgment logic is organized in a tree structure, the system can continuously supplement questions when information is missing and converge conclusions early when information is sufficient, thereby reducing irrelevant interactions.

[0177] By adopting this approach, the quality inspection Q&A process can be kept consistent with the actual machine inspection scenario. The system can organize follow-up questions according to the logical sequence of nodes, avoiding the operational burden of requiring users to submit all information at once. By incorporating text, photos, and screenshots into the same judgment chain, the completeness and accuracy of on-site evidence collection can be improved, and the probability of misjudgment due to description deviations can be reduced. At the same time, the confirmation time of complex quality inspection items can be shortened, and the consistency and efficiency of recycling quality inspection and price evaluation can be improved.

[0178] Optionally, the quality inspection question and answer module is generated by configuring exclusive system prompts for the multimodal visual language model. The system prompts include the judgment rules for the corresponding quality inspection item of the quality inspection question and answer module, as well as the image recognition area and feature requirements for the quality inspection item.

[0179] In this embodiment, the Vision-Language Model (VLM) is used to simultaneously process multimodal inputs such as text input, image input, and interface screenshots, and performs controlled question answering and judgment on the current quality inspection item based on system prompts. It should be noted that this application embodiment does not limit the specific type of the visual language model; for example, it could be doubao-1.6-vision.

[0180] The dedicated system prompts are control texts pre-written into the model context. Their content corresponds one-to-one with a single quality inspection item, limiting the model's response boundaries, evidence focus, and output format. This ensures that the model provides conclusions, explanations, or follow-up questions solely related to the target quality inspection item in the quality inspection scenario. The judgment rules can be written as a structured description of the conditions for the quality inspection item to be valid, excluded, or abnormal. The image recognition region is used to define the page location, appearance part, or screenshot area that the model should focus on. The feature requirements are used to define the keywords, icon states, numerical information, or defect manifestations that should be identified within this region.

[0181] In one specific implementation, system prompts can be dynamically generated and stored in a rule template library based on different quality inspection items. The image recognition area for each quality inspection item can be written into the prompt as a coordinate description, layout description, or object description. For example, it can specify a local area in the model number field, storage field, IMEI display field, or photo album interface on a system information page. For appearance-related quality inspection items, the camera edge, screen corners, back panel labels, and interface locations can also be set as recognition areas, and judgment requirements can be proposed for features such as cracks, deformation, missing parts, stains, or signs of tampering. After these prompts are loaded into the model, the model can prioritize searching for the corresponding area when receiving images and combine it with rules to determine whether the quality inspection conclusion is met.

[0182] In another specific implementation, the system prompts can also impose format constraints on the output content, enabling the model to return structured conclusions, evidence descriptions, and necessary follow-up suggestions corresponding to the quality inspection items after recognizing features in the image recognition region. Since the prompts already contain the judgment rules and feature requirements for the quality inspection items, the model can avoid generalizing inferences to irrelevant regions and reduce misjudgments caused by interference from image background information.

[0183] For example, Figure 7 A schematic diagram of a system prompt word provided in this application, such as Figure 7 As shown, the system's prompts include core judgment logic (judgment rules), triggered responses, answer specifications, image recognition requirements (image recognition area and feature requirements), and data return format. Specifically, the core judgment logic stipulates that determining "missing serial number" requires simultaneously meeting the following conditions: the IMEI displayed on the phone's local screen is unknown or empty, and the dialer input... The pop-up window displays either an unknown IMEI or an empty IMEI. It's also important to note other similar "unknown IMEI" displays on the dialer interface. To determine "SIM card not read," the "serial number missing" condition must first be met, followed by further verification that the phone cannot read the SIM card normally. Both conditions must be met for a judgment to be made. All judgments require screenshots or clear explanations from the user. If information is incomplete, the user will be guided to provide it; conclusions will not be drawn arbitrarily. If the "serial number missing" condition is not met, as long as any IMEI on the phone's settings page or dialer interface is normal, the user will be informed that it is normal. There is also a special judgment condition: when two IMEI numbers are the same, "serial number missing" is directly determined.

[0184] The system prompts also include an additional response module that triggers only when a user asks a specific question. For example, if a user asks "How do I fill in the IMEI number?", the response is "Just fill in the order number in the IMEI field." If a user asks "Can a phone with a damaged baseband / missing serial number be recycled?", the response is "Android phones with damaged basebands can be recycled." If a user expresses intent related to "missing serial number" in text, the response is "Dial". Is the page working properly? When a user expresses the intention of "how to check the official website without an IMEI number", the reply is "after checking the missing serial number, there is no need to judge the information on the official website (no need to check the official website)".

[0185] Regarding the answer requirements, you can output the conclusion and request supplementary information at the same time. The word count is limited to 30 words, and no analysis or logical deduction is output. Only the result and completion prompts are given.

[0186] The image recognition enhancement description states that the "About" page must be recognized first, referring to the system information page via "Settings - About" or a similar path. The "Dial Key" page refers to the page related to the phone's dial keys, displaying IMEI information or device identification code information.

[0187] Finally, it is stipulated that all results should be returned in JSON format, containing three fields: current_state, image_info, and answer. The current_state is used to summarize the results of multiple rounds of dialogue and to summarize the information already obtained; the image_info is used to determine the information in the image that is relevant to the question when the user provides an image; and the answer is used to represent the answer result.

[0188] It should be noted that in practical applications, the system prompts can also be switched according to different brands, models or quality inspection levels to adapt to the page layout and appearance differences of different devices. This application embodiment does not limit this.

[0189] Based on the above settings, the quality inspection question-and-answer module first defines the task boundaries of the model using system prompts, then the model jointly analyzes the input images, screenshots, or text information, and outputs results based on the preset judgment rules and recognition areas in the prompts. This enables the module to have dedicated question-and-answer capabilities for specific quality inspection items. Because the prompts simultaneously constrain the judgment logic and the visual attention range, the module can continuously collect evidence around the same quality inspection item and provide consistent judgments in multiple rounds of interaction.

[0190] With this implementation method, the quality inspection question-and-answer module no longer relies on generalized text retrieval, but instead uses exclusive system prompts to achieve precise constraints on a single quality inspection item. This enables the multimodal visual language model to more accurately identify key evidence areas and match judgment rules, thereby improving the accuracy of answering questions about complex quality inspection items, the consistency of conclusions, and the efficiency of on-site quality inspection, while reducing errors caused by manual retelling of image information.

[0191] S205. Based on multimodal conditional information, determine the judgment result of the corresponding quality inspection item in the target quality inspection question and answer module.

[0192] It should be noted that "conditional information based on multimodality" in this step not only means summarizing various types of information in parallel, but also means mapping text, images, screenshots, automatically read parameters and contextual information to rule variables and then making association judgments, so that evidence from different sources can participate in the same judgment logic.

[0193] Optionally, in response to the presence of multiple target quality inspection question and answer modules, the judgment order of the corresponding quality inspection items for each target quality inspection question and answer module is determined; based on the judgment order, the judgment result of the corresponding quality inspection item for each target quality inspection question and answer module is determined through multiple rounds of dialogue, based on the preset judgment rules in each target quality inspection question and answer module.

[0194] The judgment order refers to the order in which multiple quality inspection items are judged. This order can be preset based on factors such as the logical dependencies between quality inspection items, the degree of impact on the recycling price, the complexity of operation, and the information reuse rate. Generally, it follows the principle of prioritizing basic anomalies over derived anomalies, prioritizing high-impact anomalies over low-impact anomalies, and prioritizing anomalies with high information reuse rate over anomalies with low information reuse rate.

[0195] In one example, refer to Figure 2 As shown, a missing IMEI number may be related to baseband damage, which often causes the device to be unable to recognize the SIM card. Therefore, after calling the target quality inspection question and answer module corresponding to the missing IMEI serial number quality inspection item to determine that the IMEI serial number is missing, the target quality inspection question and answer module corresponding to the SIM card abnormality quality inspection item can then be called to determine whether there is a SIM card abnormality, and the judgment results of each target quality inspection question and answer module corresponding to the quality inspection item can be obtained.

[0196] In another example, the user simultaneously entered "This phone's IMEI number isn't displayed, and the memory usage is different from the official website" in the first round of questions, and uploaded a screenshot of the "About This Phone" page. The chatbot, through intent recognition, simultaneously matched two target quality inspection question-and-answer modules: IMEI serial number missing determination and memory storage mismatch determination. According to the preset determination order rules, since an IMEI serial number missing is a basic identity anomaly that directly affects the validity of all subsequent quality inspection items, and the information required for its determination can be reused by the subsequent memory storage mismatch determination, the system determined to execute the IMEI serial number missing determination first, followed by the memory storage mismatch determination. The system first initiated the IMEI serial number missing determination process, recognizing an empty IMEI display based on the existing "About This Phone" page screenshot, and guided the user to complete the dialing. After the user uploaded a screenshot of the page, the system detected that the dial-up page could not obtain the device identification code, resulting in the final conclusion of "missing serial number". Subsequently, the system automatically started the memory storage mismatch judgment process, automatically reusing the memory and storage information collected from the "About This Phone" page screenshot, and guiding the user to supplement with a screenshot of the official website query results. After the user uploaded it, the system compared and found that the device information was inconsistent with the configuration information on the official website, resulting in the final conclusion of "memory storage mismatch". The system summarized the judgment results of the two quality inspection items and returned them to the user.

[0197] S206. Based on the judgment results, determine the recycling price of the target product.

[0198] For example, the recycling price refers to the recycling quotation formed after the target product has completed quality inspection. The quotation is usually composed of a base recycling price and a quality inspection adjustment price. The judgment result of the target quality inspection item is used to characterize whether the target product has any abnormalities in the corresponding quality inspection dimension, as well as the type and severity of the abnormalities. Based on this, the system reads the price adjustment coefficient, deduction amount, or surcharge amount corresponding to the judgment result from the price rule table, and associates it with the target product's model, capacity, condition grade, and current market conditions to obtain the recycling price of the target product. The price rule table here can be pre-stored in the recycling pricing module, or it can share the same backend rule base with the quality inspection Q&A module, which facilitates direct pricing linkage after the quality inspection conclusion is output.

[0199] In its implementation, after receiving the judgment result, the system converts it into a pricing label, such as "No Anomaly," "Minor Anomaly," or "Severe Anomaly," and then calls the corresponding price calculation logic based on this label. When the judgment result indicates no anomaly, the system outputs the recycling price according to the base recycling price of the target product model. When the judgment result indicates an anomaly, the system deducts the corresponding repair costs, distribution loss costs, or residual value reduction amount according to the anomaly type, and adjusts the price based on factors such as whether the anomaly affects the main function, whether it affects resale, and whether disassembly and repair are required. For cases where multiple quality inspection items are involved in pricing simultaneously, the system can also summarize the judgment results of each quality inspection item and output the final recycling price according to preset superposition rules or priority rules, thereby avoiding duplicate pricing or omission of a single anomaly.

[0200] The process of determining the recycling price can be automated by a pricing engine. This engine can serve software programs, rule engines, or models. After receiving the quality inspection results, it outputs the price and feeds it back to the recycling operation interface for engineer confirmation. This way, after completing the assessment of the target quality inspection items, engineers can directly obtain the recycling price corresponding to the inspection conclusions without manually consulting price lists or performing calculations, thus improving the consistency between on-site inspection and quotation. Because a mapping relationship is established between the assessment results and price calculations, the system can transform complex quality inspection information into price results that can be directly used for business transactions, reducing errors from manual pricing.

[0201] By adopting the above method, the target product can automatically generate a recycling price that matches its actual condition after quality inspection. This ensures the timely reflection of the impact of abnormal situations on prices and improves the standardization and efficiency of recycling quotations. For recycling operations, this method helps reduce the uncertainty of manual valuation, improves the accuracy of engineers' on-site quotations, and ensures consistency between quality inspection results and pricing results, thereby improving the automation level and closed-loop capability of the overall recycling process.

[0202] The human-computer interaction-based quality inspection Q&A method provided in this application receives the first round of questions input by the user during the quality inspection of the target product. The first round of questions is processed by intent recognition to obtain the target quality inspection intent. The corresponding target quality inspection Q&A module is determined according to the preset mapping relationship. Then, the module guides the user to submit supplementary multimodal condition information through multiple rounds of dialogue according to the node logic order of its built-in tree-shaped decision rules. Based on all the collected information, the judgment result of the target quality inspection Q&A module is determined. Finally, the recycling price of the target product is determined according to the judgment result. This approach eliminates the need for engineers to memorize thousands of complex and dynamically updated quality inspection standards, completely freeing them from reliance on human experience and memory. It significantly reduces the inspection time for a single product, improving overall efficiency. Furthermore, the system strictly follows official tree-structured decision-making rules for guidance and judgment, effectively avoiding omissions and misjudgments that are prone to occur during manual assessment. This significantly improves the accuracy and consistency of inspection results. The support for multimodal information input addresses the industry pain point of engineers struggling to accurately describe visual quality inspection issues verbally, simplifying on-site operation procedures. Based on standardized judgment results, it automatically calculates recycling prices, ensuring fairness and reasonableness, and comprehensively enhancing service quality and user experience in door-to-door recycling scenarios.

[0203] In some embodiments, the aforementioned quality inspection Q&A system includes an intent recognition module and multiple quality inspection Q&A modules. The intent recognition module is responsible for analyzing and processing the initial Q&A information input by engineers. This module can extract key semantic features from the initial dialogue using regular expression technology to identify the corresponding target quality inspection intent. The system will then intelligently route based on the target quality inspection intent output by the intent recognition module, calling functions that match the intent. Figure 1 A corresponding quality inspection Q&A module is provided. This module can use a multi-round text and image interaction method to guide engineers to supplement the multimodal condition information required for the judgment step by step according to the built-in standardized judgment rules, and finally provide on-site engineers with professional, accurate quality inspection Q&A and standardized judgment support services.

[0204] In some embodiments, reference Figure 8 As shown, Figure 8 A schematic diagram illustrating another application scenario provided by this application.

[0205] In the first round, the user sends a screenshot of the phone's "Settings - About" page. The system uses image recognition to conclude that "IMEI1 and IMEI2 on the settings page are both empty," fulfilling the first necessary condition for determining "missing serial number." Since determining "missing serial number" requires both the phone's settings page and the dialer page to be met simultaneously, and the current information is incomplete, the system guides the user to complete the dialer information. The subsequent page information. In the standardized data returned by the system, the current_state field accurately records the currently available information, the image_info field records the image recognition results, and the answer field outputs the guiding dialogue that meets the word limit.

[0206] Second round: User sends dialing number The subsequent screenshot showed the page displaying "Unable to obtain device identification code," fulfilling the second necessary condition for "missing serial number." At this point, the system had collected both core conditions and officially determined it to be a "missing serial number." Simultaneously, according to the rules, the "no card reading" determination requires further verification of the SIM card status on top of the "missing serial number" condition. Therefore, the system proactively guided the user to provide additional information on SIM card reading, preparing for subsequent composite anomaly determinations.

[0207] In the third round, the user provided text feedback that "the SIM card is not being read." The system combined the previously confirmed "missing serial number" judgment result with the text information provided by the user this time, and found that all the judgment conditions for "not reading the card" were met (first, the missing serial number was met, and then the inability to read the SIM card was verified). Therefore, the final judgment was "not reading the card (missing serial number and inability to read the SIM card)".

[0208] This interaction can integrate multimodal information for comprehensive judgment. All responses are output in a unified JSON format and have complete context memory capabilities. Users do not need to repeatedly provide information that has already been submitted, which meets the needs of efficient and accurate quality inspection and Q&A in the on-site recycling scenario.

[0209] In some embodiments, reference Figure 9 As shown, Figure 9 This is a schematic diagram illustrating one application scenario provided in this application.

[0210] Round 1: Users send "Can I still recycle my phone if I can't find the IMEI code?" along with a screenshot of their phone's "About phone" page.

[0211] The system detected the user's intent regarding a missing IMEI and, through image recognition, concluded that "the IMEI on the device settings page is unknown." Since determining "missing IMEI" requires both the device settings page and the dialer page to be met simultaneously, and the current information is insufficient, the system guides the user to complete the dialer process. The subsequent page information. In the standardized data returned by the system, the current_state field records the current dialogue progress and the information already obtained, the image_info field records the image recognition result, and the answer field outputs the guiding dialogue that meets the word limit.

[0212] Round 2: The user sends the number "1" along with a dial number. The screenshot shows the IMEI displayed as null. At this point, the system has collected all two conditions required to determine "missing serial number". Since the user's initial question was "Can it be recycled?", it triggered the preset additional question and reply rules, so it directly replied "Android phones with damaged basebands (missing serial number) can be recycled."

[0213] In the third round, when the user pressed for a decision, asking "Does it need to be judged?", the system, based on previously collected and verified evidence of unknown IMEIs on both pages, officially determined it to be a missing serial number and reiterated, "It has been determined to be a missing serial number; missing Android serial numbers can be recovered." The `image_info` field displayed "No new image; using previous evidence of unknown dual IMEIs," demonstrating the system's contextual memory capability, eliminating the need for the user to provide information repeatedly.

[0214] In the fourth round, the user continued to ask, "Is it okay to enter the serial number, teacher?" The system recognized that the user's intent had switched to "How to fill in the IMEI number," triggering another preset additional question and reply rule, and directly replied, "Just fill in the order number in the IMEI field~", demonstrating the system's ability to handle multiple related questions in the same conversation.

[0215] This interaction can strictly follow the aforementioned "dual-page verification" judgment rules to determine missing serial numbers, actively guide users to supplement information when it is incomplete, automatically trigger corresponding standardized responses for different user questions, and output all responses in a unified JSON format. It also has multimodal image recognition and context memory capabilities to meet the needs of real-time quality inspection and Q&A.

[0216] Figure 10 The schematic diagram of the quality inspection Q&A device based on human-computer interaction provided in this application is as follows: Figure 10 As shown, the human-computer interaction-based quality inspection Q&A device 300 provided in this embodiment includes:

[0217] The receiving module 301 is used to receive the first round of questions input by the user based on the chatbot during the quality inspection process of the target product; wherein, the chatbot includes multiple quality inspection question and answer modules, and each quality inspection question and answer module corresponds to a quality inspection item.

[0218] The determination module 302 is used to determine at least one target quality inspection question-and-answer module that matches the information from the first round of questions, based on the information from the first round of questions.

[0219] The judgment module 303 is used to determine the judgment result of the corresponding quality inspection item in the target quality inspection question and answer module through a multi-round dialogue based on the judgment rules preset in the target quality inspection question and answer module.

[0220] In one possible implementation, the determining module 302 is configured to:

[0221] The intent of the first round of questions is processed to identify the target quality inspection intent.

[0222] Based on the target quality inspection intent and the mapping relationship between the preset quality inspection intent and the quality inspection question and answer module, at least one target quality inspection question and answer module that matches the information in the first round of questions is determined.

[0223] In one possible implementation, the determining module 302 is configured to:

[0224] At least one of the following methods—regular expression matching, keyword matching, and semantic similarity matching—is used to match the first round of query information and determine the target quality inspection intent.

[0225] In one possible implementation, the decision rule is a tree-structured decision rule; the decision module 303 is used for:

[0226] For any target quality inspection question and answer module, according to the node logical order of the tree-shaped decision rules in the target quality inspection question and answer module, the user is guided to supplement and submit the multimodal condition information required for the judgment through multiple rounds of dialogue;

[0227] Based on multimodal conditional information, the judgment result of the corresponding quality inspection item in the target quality inspection question and answer module is determined.

[0228] In one possible implementation, the determination module 303 is used for:

[0229] In response to the fact that there are multiple target quality inspection Q&A modules, the judgment order of the corresponding quality inspection items for each target quality inspection Q&A module is determined;

[0230] Based on the judgment order, the judgment results of the corresponding quality inspection items in each target quality inspection question and answer module are determined through multiple rounds of dialogue, based on the pre-set judgment rules in each target quality inspection question and answer module.

[0231] In one possible implementation, the quality inspection question-and-answer module is generated by configuring exclusive system prompts for the multimodal visual language model. The system prompts include the judgment rules for the corresponding quality inspection item of the quality inspection question-and-answer module, as well as the image recognition area and feature requirements for the quality inspection item.

[0232] In one possible implementation, the quality inspection item is the determination of missing International Mobile Equipment Identity (IMEI) serial numbers, and the preset judgment rules in the target quality inspection question and answer module corresponding to the quality inspection item include:

[0233] Guide users to provide their local page and determine if the local page is working correctly;

[0234] If the local page is normal, it is determined that there is no missing IMEI serial number.

[0235] If the local page is abnormal, guide the user to provide the dial page and determine if the dial page is abnormal;

[0236] If the dialing page is normal, it is determined that there is no IMEI serial number missing anomaly.

[0237] If the dialing page is abnormal, it is determined that there is an IMEI serial number missing error, and the user is guided to provide the user identification SIM card reading status information;

[0238] If the card reading status information indicates that the card reading is normal, it is determined that there is an IMEI serial number missing abnormality; if the card reading status information indicates that the card reading is abnormal, it is determined that there is both an IMEI serial number missing abnormality and a SIM card abnormality.

[0239] In one possible implementation, the quality inspection item is a determination of discrepancies between the official website information and the target quality inspection Q&A module corresponding to the quality inspection item has preset judgment rules including:

[0240] Check if the target product's configuration information exists in the preset database;

[0241] If the configuration information for the target product exists in the database, and the configuration information is consistent with the user-provided local information, then it is determined that there is no discrepancy between the official website information and the local information; if the configuration information is inconsistent with the local information, then it is determined that there is a discrepancy between the official website information and the local information.

[0242] If the configuration information for the target product does not exist in the database, determine whether the target product supports querying from the official website;

[0243] If the target product does not support official website search, then it is determined that there is no discrepancy between the official website information and the target product information; if the target product supports official website search, then the user is guided to search the official website.

[0244] If the configuration information for the target product is not available on the official website, it is determined that there is an information discrepancy issue on the official website.

[0245] If the configuration information for the target product exists on the official website and the device model matches the machine, then if the configuration information does not display memory and storage information, it is determined that there is no discrepancy between the official website information and the configuration information; if the configuration information displays memory and storage information and the memory and storage information matches the machine, it is determined that there is no discrepancy between the official website information and the configuration information; if the configuration information displays memory and storage information but the memory and storage information does not match the machine, it is determined that there is a discrepancy between the official website information and the configuration information.

[0246] In one possible implementation, the quality inspection item is a new machine assessment, and the pre-set assessment rules in the target quality inspection question and answer module corresponding to the quality inspection item include:

[0247] Guide users to replenish the packaging of the target product;

[0248] If the packaging is opened, the device is considered not brand new.

[0249] If the packaging is unopened and the purchase channel is the preset channel, guide the user to provide the packaging box, product label and device page of the target product, and verify whether the IMEI serial numbers in the packaging box, product label and device page are consistent;

[0250] If the IMEI serial number is inconsistent, the device is determined to be not brand new;

[0251] If the IMEI serial numbers match, for preset brand models that support official website queries, verify whether the retention period of the target product displayed on the official website is greater than the preset threshold.

[0252] If the retention period of the target product is greater than the preset threshold, it is determined to be a non-brand new machine; if the retention period of the target product is less than or equal to the preset threshold, it is determined to be a brand new machine.

[0253] In one possible implementation, the device further includes a processing module for:

[0254] Based on the results of the target quality inspection items, the recycling price of the target product is determined.

[0255] The quality inspection and Q&A device based on human-computer interaction provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0256] Figure 11 A schematic diagram of the structure of the electronic device provided in this application. Figure 11 As shown, the electronic device 400 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the electronic device 400 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus. This electronic device can be the aforementioned server device.

[0257] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0258] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0259] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0260] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0261] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0262] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0263] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0264] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0265] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0266] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0267] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0268] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0269] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0270] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0271] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A quality inspection Q&A method based on human-computer interaction, characterized in that, include: During the quality inspection of the target product, the system receives the first round of questions from the user via a chatbot; the chatbot includes multiple quality inspection question-and-answer modules, each corresponding to a quality inspection item. Based on the information from the first round of questions, at least one target quality inspection question-and-answer module that matches the information from the first round of questions is identified; Based on the preset judgment rules in the target quality inspection question and answer module, the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module is determined through multi-round dialogue.

2. The method according to claim 1, characterized in that, The step of determining at least one target quality inspection question-answering module that matches the first round of question information includes: The intent recognition process is performed on the first round of questioning information to obtain the target quality inspection intent; Based on the target quality inspection intent and the preset mapping relationship between the quality inspection intent and the quality inspection question and answer module, at least one target quality inspection question and answer module that matches the first round of question information is determined.

3. The method according to claim 2, characterized in that, The process of performing intent recognition processing on the first round of questioning information to obtain the target quality inspection intent includes: The first round of question information is matched using at least one of regular expression matching, keyword matching, and semantic similarity matching to determine the target quality inspection intent.

4. The method according to claim 1, characterized in that, The judgment rule is a tree-structured decision rule; the determination of the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module based on the preset judgment rule in the target quality inspection question and answer module, through multi-round dialogue, includes: For any of the target quality inspection question and answer modules, the user is guided to supplement and submit the multimodal condition information required for the judgment through a multi-turn dialogue, according to the node logical order of the tree-shaped decision rules in the target quality inspection question and answer module. Based on the multimodal conditional information, the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module is determined.

5. The method according to claim 1, characterized in that, The method of determining the judgment result of the corresponding quality inspection item of the target quality inspection question and answer module through multi-round dialogue based on the preset judgment rules in the target quality inspection question and answer module also includes: In response to the presence of multiple target quality inspection question and answer modules, the judgment order of the corresponding quality inspection items for each target quality inspection question and answer module is determined; Based on the aforementioned judgment order, the judgment results of the corresponding quality inspection items of each target quality inspection question and answer module are determined sequentially based on the preset judgment rules in each target quality inspection question and answer module through a multi-round dialogue.

6. The method according to claim 1, characterized in that, The quality inspection question-and-answer module is generated by configuring exclusive system prompts for the multimodal visual language model. The system prompts include the judgment rules for the corresponding quality inspection items of the quality inspection question-and-answer module, as well as the image recognition area and feature requirements for the quality inspection items.

7. The method according to claim 1, characterized in that, The quality inspection item is the determination of missing International Mobile Equipment Identity (IMEI) serial numbers. The preset judgment rules in the target quality inspection question and answer module corresponding to the quality inspection item include: Guide users to provide their local page and determine if the local page is working correctly; If the local page is normal, it is determined that there is no IMEI serial number missing anomaly; If the local page is abnormal, the user is guided to provide the dial page and it is determined whether the dial page is abnormal. If the dialing page is normal, it is determined that there is no IMEI serial number missing anomaly. If the dialing page is abnormal, it is determined that there is an IMEI serial number missing error, and the user is guided to provide the user identification SIM card reading status information; If the card reading status information indicates that the card reading is normal, then it is determined that there is an IMEI serial number missing anomaly; if the card reading status information indicates that the card reading is abnormal, then it is determined that there is both an IMEI serial number missing anomaly and a SIM card anomaly.

8. The method according to claim 1, characterized in that, The quality inspection item is a determination of discrepancies between the official website information and the target quality inspection Q&A module corresponding to the quality inspection item has the following preset judgment rules: Query the preset database to see if the configuration information for the target product exists; If the configuration information of the target product exists in the database, and the configuration information is consistent with the local machine information provided by the user, it is determined that there is no official website information discrepancy; if the configuration information is inconsistent with the local machine information, it is determined that there is an official website information discrepancy. If the database does not contain configuration information for the target product, determine whether the target product supports online querying. If the target product does not support official website search, it is determined that there is no official website information discrepancy; if the target product supports official website search, the user is guided to search the official website. If the official website does not contain the configuration information for the target product, it is determined that there is an error due to mismatch between the official website information and the target product information. If the configuration information of the target product exists on the official website and the model is consistent with the local machine, then if the configuration information does not display memory and storage information, it is determined that there is no official website information mismatch anomaly; if the configuration information displays memory and storage information and the memory and storage information is consistent with the local machine, it is determined that there is no official website information mismatch anomaly; if the configuration information displays memory and storage information but the memory and storage information is inconsistent with the local machine, it is determined that there is an official website information mismatch anomaly.

9. The method according to claim 1, characterized in that, The quality inspection items are for new machines, and the preset judgment rules in the target quality inspection question and answer module corresponding to the quality inspection items include: Guide users to replenish the packaging of the target product; If the packaging is opened, the device is considered not brand new. If the packaging is unopened and the purchase channel is a preset channel, then guide the user to provide the packaging box, product label, and device page of the target product, and verify whether the IMEI serial numbers in the packaging box, product label, and device page are consistent. If the IMEI serial number is inconsistent, the device is determined to be not brand new; If the IMEI serial numbers match, for preset brand models that support official website queries, verify whether the retention period of the target product displayed on the official website is greater than the preset threshold. If the retention period of the target product is greater than a preset threshold, it is determined to be a non-brand new product; if the retention period of the target product is less than or equal to the preset threshold, it is determined to be a brand new product.

10. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-9.

Images