User opinion processing method and system

Abstract

The invention discloses a user opinion processing method and system.The user opinion processing method comprises the steps that user feedback opinions of multiple modalities are received, an identification model is called to analyze the user feedback opinions of the multiple modalities, and question categories corresponding to the user feedback opinions are obtained, determining a final problem category of the product from the problem categories corresponding to the feedback suggestions of the users; determining responsibility department information matched with the final problem category, and generating a to-be-processed work order according to a target user feedback opinion corresponding to the final problem category, an analysis result of the target user feedback opinion and the responsibility department information; and sending the to-be-processed work order to a service system corresponding to the responsible department information. According to the method and the device, the user opinion analysis accuracy is improved, automatic routing of multi-modal feedback opinions is realized, and the user feedback opinions are ensured to be transmitted to a proper responsible department, so that the processing efficiency of the user feedback opinions and the accuracy of a processing result are improved, and the user experience is improved.

Description

Technical Field

[0001] This disclosure relates to the field of large language model technology, and in particular to a method and system for processing user opinions. Background Technology

[0002] With the increasing intelligence and networking of kitchen appliances, users are increasingly demanding feedback through mini-programs or online platforms. For user feedback, manual screening and categorization of large amounts of text and image feedback by customer service or operations personnel before forwarding it to R&D, after-sales, or product design departments is not only inefficient and inaccurate, but may also result in untimely handling of issues affecting product safety. Alternatively, customer service robots can analyze problems through simple Q&A sessions with users before being manually forwarded to relevant departments. However, this approach also lacks accuracy and intelligence. Summary of the Invention

[0003] In view of the aforementioned defects or deficiencies in the existing technology, it is desirable to provide a user feedback processing method and system. Through a fully automated processing approach that automatically identifies the problem category of multimodal user feedback, automatically maps it to the responsible department, and automatically sends work orders, this approach not only improves the accuracy of problem identification and reduces the workload of manual screening, but also improves the accuracy of user feedback parsing by automatically identifying multimodal user feedback using an identification model. Furthermore, it achieves automatic routing of multimodal feedback to ensure that user feedback is delivered to the appropriate responsible department, thereby improving the efficiency of user feedback processing, the accuracy of the processing results, and the user experience.

[0004] Firstly, this application provides a method for processing user feedback. The method includes: Receive user feedback from multiple modalities, call the recognition model to parse the user feedback from multiple modalities to obtain the problem category corresponding to each user feedback, and determine the final problem category of the product from the problem categories corresponding to each user feedback; Determine the responsible department information that matches the final problem category, and generate a work order to be processed based on the target user feedback corresponding to the final problem category, the parsing result of the target user feedback, and the responsible department information; Send the pending work order to the business system corresponding to the responsible department information.

[0005] In conjunction with the first aspect, in one possible implementation, each piece of user feedback corresponds to a parsing result, which includes a problem category and product problem information. Determining the final problem category of the product from the problem categories corresponding to each piece of user feedback includes: Based on the problem categories and product fault information included in each of the analysis results, the similarity between the analysis results is determined. If there are parsing results with a similarity higher than the similarity threshold, then the final problem category is determined based on the parsing results with a similarity higher than the similarity threshold; If the similarity is lower than the similarity threshold, the confidence level of each user feedback opinion is calculated, and the final problem category is determined based on the parsing result of the user feedback opinion with the highest confidence level.

[0006] In conjunction with the first aspect, in one possible implementation, generating a work order to be processed based on the target user feedback corresponding to the final problem category, the parsing result of the target user feedback, and the responsible department information includes: Identify the danger characteristics in each of the target user feedback opinions, and determine the risk level of the target user feedback opinions based on the danger characteristics information; The pending work order is generated based on the target user feedback corresponding to the final problem category, the analysis results of the target user feedback, the responsible department information, and the risk level.

[0007] In conjunction with the first aspect, in one possible implementation, identifying hazard characteristic information in each of the target feedback opinions and determining the risk level of the target feedback opinions based on the hazard characteristic information includes: Identify dangerous image features and / or dangerous keywords in the target user feedback; The danger rating of the dangerous image features and / or the dangerous keywords is determined according to the rating mapping relationship; the rating mapping relationship is used to characterize the danger rating corresponding to different dangerous image features and the danger rating corresponding to different dangerous keywords. The risk level of the target feedback opinion is determined based on the hazard score of the hazard image features and / or the hazard keywords.

[0008] In conjunction with the first aspect, in one possible implementation, determining the responsible department information matching the final problem category includes: From the pre-built enterprise knowledge graph, find the name of the responsible department corresponding to the final problem category of the product, and determine the information of the responsible department based on the found responsible department name; The enterprise knowledge graph is used to represent the problem categories of different products and the names of the responsible departments for handling different problem categories.

[0009] In conjunction with the first aspect, in one possible implementation, the method further includes: Receive the processing result of the pending work order; If the processing result indicates that the information of the responsible department does not match the actual information of the responsible department, then the enterprise knowledge graph is adjusted according to the information of the actual responsible department.

[0010] In conjunction with the first aspect, in one possible implementation, sending the pending work order to the business system corresponding to the responsible department information includes: If the pending work order is not processed within the preset time period, new user feedback will be received. If the issue category corresponding to the new user feedback is the same as the final issue category, then an expedited processing identifier is added to the pending work order, and the pending work order with the expedited processing identifier is sent to the business system.

[0011] In conjunction with the first aspect, in one possible implementation, the method further includes: The system obtains the processing status of the pending work orders from the business system and reports the processing status to the user client.

[0012] In conjunction with the first aspect, in one possible implementation, the method further includes: Record the entire process data of user opinion processing and write the entire process data of user opinion processing into the training dataset; the training dataset is used to perform incremental training of the recognition model periodically; The data for the entire user feedback processing process includes the target user feedback, the analysis results, the responsible department information, and the confirmation information of the processing results of the pending work orders.

[0013] Secondly, this application also provides a user feedback processing system. This system includes: a central platform, a user terminal, and multiple business systems. The user terminal is used to receive user feedback from multiple modalities and send the user feedback from multiple modalities to the central platform. The central platform is used to call the recognition model to parse the user feedback opinions of the multiple modalities and obtain the problem category corresponding to each user feedback opinion; The central platform is also used to determine the final problem category of the product from the problem categories corresponding to each user feedback opinion, determine the responsible department information that matches the final problem category, and generate a pending work order based on the target user feedback opinion corresponding to the final problem category, the parsing result of the target user feedback opinion, and the responsible department information; and send the pending work order to the business system corresponding to the responsible department information. The business system is used to receive and process the pending work orders.

[0014] This application provides a user feedback processing method and system. The method first parses user feedback from multiple modalities to obtain the corresponding problem categories for each feedback. Then, it determines the final problem category of the product from these categories. Next, it identifies the responsible department information matching the final problem category. Based on the target user feedback corresponding to the final problem category, the parsing results of the target user feedback, and the responsible department information, it generates a work order and sends it to the business system corresponding to the responsible department. This fully automated processing method—automatic identification of problem categories, automatic mapping of responsible departments, and automatic work order sending for multimodal user feedback—not only improves the accuracy of problem identification and reduces manual screening, but also improves the accuracy of user feedback parsing by automatically identifying multimodal user feedback using a recognition model. Furthermore, it achieves automatic routing of multimodal feedback, ensuring that user feedback is delivered to the appropriate responsible department, thereby improving the processing efficiency and accuracy of the processing results, and enhancing the user experience. Attached Figure Description

[0015] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is one of the schematic diagrams of a user feedback processing system in one embodiment; Figure 2 This is one of the flowcharts illustrating a user feedback processing method in one embodiment; Figure 3 This is a schematic diagram of the user feedback processing procedure in one embodiment; Figure 4 This is a second flowchart illustrating a user feedback processing method in one embodiment; Figure 5 This is the third flowchart illustrating a user feedback processing method in one embodiment; Figure 6 This is a flowchart of a user feedback processing method in one embodiment, number four. Figure 7 This is the fifth flowchart illustrating a user feedback processing method in one embodiment; Figure 8 This is a schematic diagram of the spectrum optimization process in one embodiment; Figure 9 This is a flowchart of a user feedback processing method in one embodiment, number six. Figure 10 This is a schematic diagram of the work order processing tracking and user feedback process in one embodiment; Figure 11 This is the seventh flowchart illustrating a user feedback processing method in one embodiment; Figure 12 This is a second schematic diagram of a user feedback processing system in one embodiment; Figure 13 This is a schematic diagram of the closed-loop processing of user feedback in one embodiment; Figure 14 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0016] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present application will now be described in detail with reference to the accompanying drawings and embodiments. Furthermore, the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The terms "first" and "second," etc., in the specification and claims of the embodiments of this application are used to distinguish different objects, not to describe a specific order of objects.

[0018] With the increasing intelligence and networking of kitchen appliances, users are increasingly providing feedback through mini-programs or online platforms. This feedback includes: product usage experience (such as noise, smoke extraction power, and design); installation and after-sales issues (such as improper installation or slow repair response); design improvement suggestions (such as sharp edges on range hoods that could cause injury); and safety hazards (such as electrical leaks, gas leaks, and sharp parts). Therefore, effectively handling diverse user feedback is crucial.

[0019] In related technologies, user feedback can be handled by human customer service or operations staff who manually screen and categorize large amounts of text and image feedback before forwarding it to other departments such as R&D, after-sales, or product design. However, this approach has the following problems: 1) Low efficiency: Relying on manual screening and classification is time-consuming and labor-intensive; 2) Poor accuracy: Manual classification is prone to errors, resulting in user feedback not being delivered to the appropriate department; 3) Delayed response: Poor user experience, and in severe cases, problems affecting the safe use of the product cannot be handled quickly.

[0020] Alternatively, customer service robots can analyze problems by engaging in simple Q&A or sentiment analysis with users, and then manually forward them to relevant departments for processing. However, this approach is not very accurate or intelligent.

[0021] Low efficiency not only results in low efficiency and poor accuracy, but may also lead to problems affecting the safe use of products not being handled in a timely manner; alternatively, customer service robots can analyze problems by conducting simple question-and-answer sessions with users, and then manually transfer them to relevant departments for processing. However, the accuracy and intelligence of this approach are also low.

[0022] To address the aforementioned technical problems, this application provides a user feedback processing method and system, wherein the user feedback processing method is applied to... Figure 1 The user feedback processing system shown here executes the user feedback processing method through the following entity: Figure 1 The central platform in the user feedback processing system shown is a digital platform that provides enterprises with resource integration, collaborative support, and comprehensive management services, and possesses at least user management and internal operation functions. Furthermore, this central platform can be located within an electronic device or independently outside of an electronic device; the electronic device can be a personal computer, server, embedded system, or other devices. This application does not specifically limit its scope. The following description uses the central platform as an example to illustrate the user feedback processing method.

[0023] Below, in conjunction with Figures 2 to 14 This application describes a method for processing user feedback. To facilitate understanding of this method, several exemplary embodiments will be provided below for detailed explanation. It is understood that these exemplary embodiments can be combined with each other, and similar concepts or processes may not be repeated in some embodiments.

[0024] Reference Figure 2 This is a flowchart illustrating the user feedback processing method provided in an embodiment of this application. Figure 2 As shown, the user feedback processing method includes the following steps 101 to 103.

[0025] Step 101: Receive user feedback from multiple modalities, call the recognition model to parse the user feedback from multiple modalities to obtain the problem category corresponding to each user feedback, and determine the final problem category of the product from the problem categories corresponding to each user feedback.

[0026] In this context, multimodal user feedback can be understood as product usage feedback submitted by users in at least two of the following formats: text, images, voice, and video. This ensures that comprehensive user feedback on the product is obtained.

[0027] For example, the two modalities of user feedback are text feedback and image feedback. Text feedback is a natural language description, such as "The corners of the range hood are too sharp and easy to bump into." Image feedback is a visual language description, such as a picture of the corners of the range hood taken by the user. The number of image feedbacks can be one or more, such as receiving nine pictures of the corners of the range hood taken according to the platform settings.

[0028] User feedback for each modality can be understood as the user's feedback on the product's usage experience using the corresponding modality; the product here can be kitchen appliances, such as range hoods, gas stoves, microwave ovens, ovens, steam cookers, etc.

[0029] The recognition model can be understood as a pre-trained multimodal recognition model capable of processing multimodal user feedback. This model possesses text parsing, image parsing, audio / video parsing, and comprehensive judgment capabilities. Furthermore, it can categorize the input multimodal user feedback into question categories and perform a comprehensive judgment on the parsed question categories to output the final question category for the product. For example, the multimodal recognition model can be a cross-modal neural network model / vector concatenation model combined with a Softmax classifier.

[0030] Alternatively, the recognition model can be understood as including a large natural language model for processing text feedback, an image detection model for processing image feedback, an audio / video recognition model for processing audio / video feedback, and a classifier. The classifier can comprehensively determine the problem categories identified by the large natural language model, the image detection model, and / or the audio / video recognition model, thereby outputting the final problem category of the product. For example, the large natural language model is a pre-trained large model containing BERT / RoBERTa architecture and named entity recognition architecture, and the image detection model is a pre-trained object detection network (such as the YOLOv8 / DETR model). The YOLOv8 / DETR model is an end-to-end object detector that supports object detection, classification, and real-time segmentation tasks.

[0031] Each user feedback item can be categorized into at least one issue category. Each issue category can be one of the following: product functional defects, product installation problems, product user experience, installation and after-sales issues, design improvement suggestions, and security risk issues.

[0032] The final problem category can be understood as a unified product problem classification determined by comprehensively analyzing feedback from multiple users, used to solve the problem of misjudgment that may occur with feedback from a single modality.

[0033] Specifically, users can submit user feedback in multiple modalities to the central platform through a client or front-end mini-program. The client or front-end mini-program can collect user feedback in different modalities, package the collected user feedback in multiple modalities into a data package, and send it to the central platform.

[0034] The central platform receives a data packet of feedback and then synchronously parses the multimodal user feedback in the data packet by calling the recognition model. After that, it can obtain the final problem category output by the recognition model, or it can make a comprehensive judgment on the problem categories corresponding to each user feedback output by the recognition model to obtain the final problem category of the product.

[0035] For example, if the data packet of feedback received by the central platform includes both text feedback and image feedback, and if a security risk issue is identified from both the text feedback and the image feedback through the recognition model, the final issue category of the product can be determined to be a security risk issue.

[0036] Step 102: Determine the responsible department information that matches the final problem category, and generate a work order to be processed based on the target user feedback corresponding to the final problem category, the analysis results of the target user feedback, and the responsible department information.

[0037] The responsible department information can be understood as the department responsible for handling issues in relation to the company's organizational structure. This can be obtained by searching the pre-built mapping relationship between the responsibilities of different departments and different issue categories, thereby ensuring the accurate distribution of the final issue category for the product. For example, installation / repair issues correspond to the after-sales service department, design defect / safety risk issues correspond to the R&D department, and appearance / experience suggestion issues correspond to the product design department.

[0038] The target user feedback corresponding to the final issue category can be user feedback from multiple modalities, or it can be one of the user feedback from multiple modalities.

[0039] When the target user feedback is from multiple modalities, the parsing result of the target user feedback is the parsing result of each of the multiple modalities; or, when the target user feedback is from one of the multiple modalities, the parsing result of the target user feedback is the parsing result of that single user feedback.

[0040] When the analysis results of target user feedback include both text feedback and image feedback, the analysis results for text feedback include the product model, the name of the involved component (e.g., the corner of a range hood), the problem category (e.g., safety risk), and the sentiment (e.g., negative / neutral / positive). The analysis results for image feedback include the product model, the name of the involved component (e.g., panel, knob, corner), appearance defects (e.g., damage, sharpness, cracks, abnormal appearance), and the problem category. Furthermore, the analysis results for image feedback can be output in the form of tags, such as: Product = Range Hood, Component = Corner, Feature = Sharp, Problem = Safety Risk. Thus, using user-uploaded image feedback for product and specific component identification, combined with image recognition for specific component localization, can supplement text information and improve the accuracy and completeness of product safety issue identification.

[0041] For example, when user feedback in multiple modalities includes both text and image feedback, and the product is a range hood, the following can be referenced: Figure 3 The diagram shown illustrates the user feedback processing procedure. Figure 3 As shown, the central platform analyzes the text feedback, and the analysis results include range hood, edges and corners, sharp points, neutral, and safety risks. The analysis results of the image feedback include: product = range hood, component = edges and corners, feature = sharp points, and problem = safety risk. At this point, it can be determined that the final problem category of the range hood is a safety risk problem.

[0042] In addition, when the central platform uses the object detection network to analyze the image feedback, it first checks whether the image feedback contains kitchen appliances. If the image feedback does not contain kitchen appliances, the image is ignored. Otherwise, if the image feedback contains kitchen appliances, the specific components (such as panels, knobs, corners, etc.) are further located. Then, the defect features (such as damage, sharpness, cracks, abnormal appearance, etc.) are identified for the located specific components.

[0043] To identify abnormal appearances, the appearance of the identified specific component can be compared with the appearance of the corresponding standard component. If the appearance does not match the standard component (e.g., the similarity is too low), the appearance of the identified specific component is determined to be abnormal. Abnormal appearances can be understood as caused by other abnormal situations such as the corresponding kitchen appliance being dropped from a height or deformed by heavy objects.

[0044] A pending work order can be understood as a structured task document containing problem details, analysis results, and instructions from the responsible department. For example, the content of a pending work order may include text feedback, image feedback, the analysis results of the text feedback, the analysis results of the image feedback, and the responsible department information.

[0045] Specifically, for the final problem category of a product, the central platform can find the name of the responsible department corresponding to the final problem category of the product from the pre-built mapping relationship between different department responsibilities and different problem categories, and then determine the name of the responsible department as the responsible department information, so as to generate a work order to be processed based on the target user feedback corresponding to the final problem category, the analysis results of the target user feedback, and the responsible department information.

[0046] Step 103: Send the pending work order to the business system corresponding to the responsible department information.

[0047] Specifically, the central platform can directly push the generated pending work orders to the business system corresponding to the responsible department (such as Enterprise Resource Planning (ERP) or Customer Relationship Management (CRM) system) through the enterprise service bus, thereby triggering the corresponding responsible department to execute the pending work order.

[0048] The user feedback processing method provided in this application first parses user feedback from multiple modalities to obtain the problem category corresponding to each user feedback, and then determines the final problem category of the product from the problem categories corresponding to each user feedback. Next, it determines the responsible department information matching the final problem category, and generates a work order to be processed based on the target user feedback corresponding to the final problem category, the parsing result of the target user feedback, and the responsible department information, and sends it to the business system corresponding to the responsible department information. In this way, through a fully automated processing method of automatically identifying problem categories, automatically mapping responsible departments, and automatically sending work orders for multimodal user feedback, it not only improves the accuracy of problem identification and reduces the workload of manual screening, but also improves the accuracy of user feedback parsing by automatically identifying multimodal user feedback using a recognition model, and achieves automatic routing of multimodal feedback to ensure that user feedback is delivered to the appropriate responsible department, thereby improving the processing efficiency of user feedback, the accuracy of the processing results, and the user experience.

[0049] Based on the above Figure 2 In one example embodiment of the method shown, considering that each modality of user feedback can yield a parsing result after identification and analysis, and that each parsing result can contain the corresponding product's problem category and product problem information, in order to improve the accuracy of parsing and identification, the parsing result that is closest to the user's actual intention can be selected from multiple parsing results to determine the final product problem category. Based on this, the specific process of determining the final product problem category from the problem categories corresponding to each user feedback in step 101 in this embodiment can be achieved through... Figure 4 Steps 201 to 203 shown are implemented.

[0050] Step 201: Based on the problem categories and product fault information included in each analysis result, determine the similarity between each analysis result.

[0051] Step 202: If there are parsing results with similarity higher than the similarity threshold, then determine the final problem category based on the parsing results with similarity higher than the similarity threshold.

[0052] Step 203: If the similarity is lower than the similarity threshold, calculate the confidence level of each user's feedback and determine the final problem category based on the analysis result of the user's feedback with the highest confidence level.

[0053] The similarity threshold can be understood as a critical value used to determine whether the parsing results have sufficient consistency. Specifically, it can be implemented by using a preset numerical range or a dynamic adjustment algorithm. For example, the similarity threshold can be set to 0.8 or dynamically calculated based on historical data.

[0054] Confidence level can be understood as a quantitative evaluation metric of the reliability or credibility of a single user's feedback, that is, the credibility of a user's feedback using each modality. Confidence level can be used to select the most credible user feedback as the basis for deciding the final question category when there is insufficient similarity among multiple parsing results.

[0055] Specifically, in determining the final problem category of a product, the problem category and product fault information of each parsing result are first structured to form a comparable result data vector. Then, the similarity between each result data vector is calculated using algorithms such as cosine similarity or Euclidean distance. Finally, the similarity is compared with a similarity threshold.

[0056] If the similarity is higher than the similarity threshold, then all the multiple parsing results are considered to be the parsing results that are closest to the user's actual intention. In this case, the question category contained in any of the multiple parsing results can be determined as the final question category of the product.

[0057] Conversely, if the similarity is below the similarity threshold, it is assumed that among the multiple parsing results, there is at least one parsing result that most closely reflects the user's true intent and at least one parsing result that least closely reflects the user's true intent. In this case, the user feedback with the highest confidence level can be selected from the user feedback from multiple modalities, and the parsing result of the user feedback with the highest confidence level is determined as the parsing result that most closely reflects the user's true intent. The problem category contained in this parsing result is then determined as the final problem category of the product. For example, when multiple parsing results include parsing results of text feedback and parsing results of image feedback, and the confidence level of the text feedback is 0.7 and the confidence level of the image feedback is 0.9, then the problem category of the product in the parsing result of the image feedback is determined as the final problem category of the product.

[0058] It should be noted that the confidence level of each user feedback opinion can be determined using a pre-trained opinion confidence level determination model. Furthermore, the confidence level can also be used to select the most reliable analytical result from multiple results when the similarity between them is insufficient, thus serving as the basis for the final problem category decision. In this way, the confidence level of each analytical result can be determined, and the final product category can be determined based on the analytical result with the highest confidence level.

[0059] For example, when multiple parsing results include both text feedback and image feedback, and the text feedback includes the problem category and product problem information (such as product model, name of involved parts, and sentiment), while the image feedback includes the problem category and product problem information (such as product model, name of involved parts, and appearance defects), the confidence level of each parsing result can be determined using a weighted confidence calculation formula or a pre-trained confidence level.

[0060] Furthermore, it should be noted that when multiple parsing results are the closest to the user's actual intent, the confidence of each parsing result can be increased to enhance the accuracy of the subsequent self-learning of the recognition model.

[0061] The user feedback processing method provided in this application introduces a two-layer decision-making mechanism of similarity threshold screening and confidence assessment, which can accurately identify the core problem categories of a product. Especially when multiple user feedback opinions are contradictory or scattered, the method avoids the problem classification errors caused by judging user opinions from a single modality by quantitatively analyzing the consistency and reliability of user feedback opinions, thereby improving the accuracy of product problem classification and processing efficiency.

[0062] Based on the above Figure 2In one example embodiment of the method shown, considering the varying urgency of the user's true intent, the processing priority of the generated work order will also differ. This allows for the generation of pending work orders based on the risk level of the target user feedback that best reflects the user's true intent. Based on this, in step 102, pending work orders are generated according to the target user feedback corresponding to the final problem category, the analysis results of the target user feedback, and the responsible department information. The specific process in this embodiment can be achieved through… Figure 5 Steps 301 and 302 shown are implemented.

[0063] Step 301: Identify the hazard characteristics in the feedback from each target user, and determine the risk level of the feedback based on the hazard characteristics.

[0064] Step 302: Generate a pending work order based on the target user feedback corresponding to the final problem category, the analysis results of the target user feedback, the responsible department information, and the risk level.

[0065] Among them, hazard characteristic information can be understood as key information related to product safety extracted from the feedback of each target user.

[0066] Risk level can be understood as a priority index based on the severity of all dangerous characteristics, thereby determining the urgency of the user's true intentions.

[0067] Specifically, during the work order generation process, a pre-trained feature recognition model can first identify hazardous features from the feedback of each target user. Then, all identified hazardous features are input into a risk level determination model. After concatenating all hazardous features, they are mapped to risk level probabilities. Based on the relationship between the risk level probabilities and preset probability thresholds, the risk level of each target user's feedback is determined and output. Alternatively, all identified hazardous features can be input into a scoring calculation engine to generate a comprehensive risk score according to predefined scoring rules. For example, identifying two medium-risk features or one high-risk feature can trigger a high-risk level determination.

[0068] At this point, the target user feedback corresponding to the final problem category of the product, the parsed results of the target user feedback, the responsible department information, and the corresponding risk level can be filled into a pre-set document format work order template, and a risk level label can be embedded to obtain a work order to be processed. In this way, high-risk work orders to be processed can be prioritized for processing. For example, the content of a work order to be processed may include text feedback, image feedback, the parsed results of text feedback, the parsed results of image feedback, the R&D department, and high risk.

[0069] The user feedback processing method provided in this application establishes standardized rules for identifying hazardous features and a user feedback priority scheduling mechanism based on safety levels. This enables the objective identification of safety hazards in multimodal user feedback, achieving automated identification and quantitative assessment of hazardous feature information in target user feedback. This facilitates the priority processing of high-risk feedback (such as sharpness, damage, electrical hazards, etc.), solving the problem of delayed emergency problem handling caused by low efficiency of manual screening, thereby improving product safety.

[0070] Based on the above Figure 5 In one example embodiment of the method shown, the identified hazard feature modalities differ depending on the modality of the user feedback. Therefore, when multimodal user feedback includes both text and image feedback, text keywords and hazard image features can be identified and comprehensively scored to determine the risk level. Based on this, the specific process of identifying hazard feature information in each target feedback and determining the risk level of the target feedback based on the hazard feature information in step 301 can be implemented in this embodiment through... Figure 6 Steps 401 to 403 shown are implemented.

[0071] Step 401: Identify dangerous image features and / or dangerous keywords in the feedback from target users.

[0072] Step 402: Determine the hazard scores of hazardous image features and / or hazardous keywords based on the scoring mapping relationship; the scoring mapping relationship is used to characterize the hazard scores corresponding to different hazardous image features and the hazard scores corresponding to different hazardous keywords.

[0073] Step 403: Determine the risk level of the target feedback based on the hazard rating of hazard image features and / or hazard keywords.

[0074] Among them, dangerous image features can be understood as abnormal visual elements with safety hazards extracted from image feedback through image recognition technology. Specifically, convolutional neural network models can be used to extract features and classify and identify image feedback, such as identifying abnormal visual elements such as burn marks on circuit boards or gas leak flames on gas stoves.

[0075] Dangerous keywords can be understood as abnormal words related to product safety in text feedback. Specifically, they can be identified by matching natural language processing technology with a predefined text keyword library, such as words like "leakage", "smoke", "fire", and "danger".

[0076] The rating mapping relationship can be understood as a pre-established correspondence rule between dangerous keywords, dangerous image features and quantitative scores. Specifically, expert experience or historical data analysis can be used to determine the dangerous scores corresponding to different dangerous image features and different dangerous keywords. For example, the dangerous image feature "burnt circuit board" can be mapped to 80 points, and the dangerous keyword "leakage" can be mapped to 90 points.

[0077] Specifically, for the target user feedback corresponding to the final problem category of the product, there can be at least one modality of user feedback. When the target user feedback includes both text and image feedback, an image feature recognition model can be used to analyze the content of the image feedback and extract visual features that indicate potential product safety hazards, thereby obtaining at least one hazardous image feature. Simultaneously, a text keyword recognition model is used to scan and analyze the text feedback and match it with a pre-set hazardous keyword database until at least one hazardous keyword is identified in the text feedback. In this way, all hazardous image features and / or all hazardous keywords in the target user feedback can be identified.

[0078] At this point, based on the pre-configured rating mapping relationship, all identified dangerous image features and / or all dangerous keywords are converted into corresponding danger scores; then, all danger scores are summed and averaged to obtain the comprehensive score value of the target user's feedback.

[0079] Considering that the higher the urgency of user feedback, the more comprehensive and prominent the dangerous keywords and / or images reflecting high urgency will be in the feedback, and the higher the processing priority of the corresponding work order will be, a comprehensive scoring threshold can be preset. This threshold can be understood as a critical value used to determine whether the feedback should be processed with the highest priority. Based on this, the comprehensive score of the feedback can be compared with the preset threshold. If the comprehensive score is higher than the threshold, the feedback is determined to be high-risk, with the highest processing priority, and must be processed immediately. Conversely, if the comprehensive score is lower than the threshold, the feedback is determined to be low-to-medium risk, with a low processing priority, and can be processed slowly or temporarily.

[0080] It should be noted that, to ensure more accurate risk levels, the first risk level corresponding to the scoring interval to which the above comprehensive score belongs can be determined based on the pre-set scoring intervals for different risk levels. Furthermore, if at least one of the first risk level and the risk level determined based on the comprehensive scoring threshold is high risk, then the risk level of the target user's feedback is considered high risk.

[0081] For example, when the first risk level is high risk and the comprehensive score is lower than the preset comprehensive score threshold, the risk level of the target user's feedback is high risk; or, when the first risk level is medium to low risk and the comprehensive score is higher than the preset comprehensive score threshold, the risk level of the target user's feedback is also high risk.

[0082] The user feedback processing method provided in this application standardizes the risk assessment process through an automated feature recognition and quantitative scoring mechanism, avoiding misjudgments or omissions caused by human factors. It achieves accurate identification and graded processing of safety hazards in user feedback, ensuring that urgent issues related to product safety can be automatically marked as high priority, shortening the response time for product safety issues, and improving the user experience.

[0083] Based on the above Figure 2 In one example embodiment of the method shown, to ensure that the user's true intent can be quickly and specifically transferred to the corresponding responsible department, the central platform can pre-build an enterprise knowledge graph to automatically route to the appropriate responsible department based on product information and problem category information. Based on this, the process of determining the responsible department information matching the final problem category in step 102 can be implemented through the following steps in this embodiment.

[0084] From the pre-built enterprise knowledge graph, find the name of the responsible department that corresponds to the final problem category of the product, and determine the responsible department information based on the found responsible department name.

[0085] Among them, the enterprise knowledge graph is used to represent the types of problems existing in different products and the names of the responsible departments for handling different types of problems.

[0086] An enterprise knowledge graph can be understood as a relational database built through a structured data model, specifically implemented using graph database technology. This enterprise knowledge graph stores the mapping rules between product issue categories and responsible departments through the relationships between nodes and edges, enabling the central platform to quickly retrieve matching results based on semantic associations.

[0087] The name of the responsible department can be understood as the identifier of the functional organization set up within the enterprise. Specifically, it can be implemented using a unified coding system, such as coding the after-sales service department as D01 and the R&D department as D02; automatic matching is achieved by binding the code with nodes in the knowledge graph.

[0088] Specifically, once the central platform determines the final problem category of a product, it can access the enterprise knowledge graph and traverse the product problem nodes within the graph until it locates the responsible department node directly associated with the final problem category of the product.

[0089] For example, if the final issue category is "circuit board overheating," and the enterprise knowledge graph pre-defines this issue as belonging to the R&D department, the central platform automatically extracts the corresponding code and contact information for the R&D department as the responsible department information. During this process, the enterprise knowledge graph allows administrators or the central platform to adjust and update node relationships in real time according to the enterprise's organizational structure. That is, when an enterprise expands its product lines or adjusts its departmental structure, only the node relationships in the enterprise knowledge graph need to be updated to maintain efficient processing of user feedback, without needing to redevelop the allocation logic, thus improving the central platform's flexibility in adapting to business changes. For example, when a new security protection issue is added to the product line, simply adding an edge linking the issue category to the quality supervision department in the enterprise knowledge graph completes the configuration.

[0090] It should be noted that, in order to improve the parsing and classification accuracy of the recognition model, historical user feedback and its corresponding historical processing results can be collected and used as training data to regularly optimize or incrementally train the recognition model, thereby helping to improve the accuracy of enterprise knowledge image routing responsibility department information. The user feedback processing method provided in this application constructs an enterprise knowledge graph containing multi-dimensional relationships, forming an extensible semantic network of departmental responsibilities and product problem categories. This enables the responsibility matching process to achieve a response time within seconds without manual intervention, automatically and accurately matching the responsible department for handling user feedback. This reduces the error rate of work order transfer caused by manual allocation and shortens the response time for handling user feedback.

[0091] In one example embodiment, considering that the department responsible for automatic routing via the enterprise knowledge graph is not necessarily the department actually responsible for handling user feedback, it is necessary for the responsible departments to communicate and negotiate to determine the department actually responsible for handling user feedback. Then, the enterprise knowledge graph is optimized based on the information of the department actually handling user feedback, thereby improving the routing accuracy of the enterprise knowledge graph. Based on this, the user feedback processing method provided in this application embodiment may further include a graph optimization process. In this embodiment, the graph optimization process can be achieved through… Figure 7 Steps 501 and 502 shown are implemented.

[0092] Step 501: Receive the processing results of the pending work orders.

[0093] Step 502: If the processing result indicates that the information of the responsible department does not match the actual information of the responsible department, then the enterprise knowledge graph is adjusted according to the actual information of the responsible department.

[0094] The processing result can be understood as the feedback data generated after the pending work order has been processed. This result data can be returned to the user's mini-program or client. Specifically, it can be implemented by marking the work order processing status or by manual confirmation information, and can be used to verify the accuracy of the matching of the responsible department's information.

[0095] The information on the actual responsible department can be understood as the identification information of the responsible department that actually completes the pending work orders or has the ability to handle the feedback from the target users. It can be used as benchmark data for optimizing the enterprise knowledge graph.

[0096] Specifically, the central platform can receive processing results from business systems. These results represent the processing outcomes of work orders completed by the corresponding responsible departments within the business systems. When the responsible department information in the processing results returned by the business systems differs from the responsible department information routed through the enterprise knowledge graph, it can be determined that the previously routed responsible department information is not the actual responsible department information. This triggers a graph update process. At this point, the enterprise knowledge graph can be adjusted based on the responsible department information (i.e., the actual responsible department information) contained in the processing results. Specifically, by parsing the actual responsible department code in the processing results, the responsible department node corresponding to the original problem category is retrieved in the enterprise knowledge graph, and the association between that node and the problem category node is modified. For example, when the after-sales service department reports that a certain product design issue should be handled by the R&D department, and the R&D department returns the processing results to the central platform, the central platform automatically reduces the association weight between the problem category node and the after-sales department in the enterprise knowledge graph, while simultaneously establishing a new association with the R&D department. This forms a closed-loop feedback mechanism, enabling the enterprise knowledge graph to continuously optimize the responsible department matching rules based on the actual processing results.

[0097] For example, refer to Figure 8 The diagram shown illustrates the spectrum optimization process, as follows: Figure 8 As shown, when the central platform routes the after-sales service department based on the final problem category of circuit board overheating through the enterprise knowledge graph, it will send the pending work order to the business system A corresponding to the after-sales service department. After determining that the R&D department should handle it through linear mediation or other methods, the pending work order will be sent to the business system B corresponding to the R&D department. Then, business system B will send the R&D department information to the central platform, and the central platform will optimize the enterprise knowledge graph. The user feedback processing method provided in this application establishes an automated correction mechanism through processing result feedback. It can automatically identify situations where the responsible department is incorrectly matched, and by correcting the relationships in the knowledge graph, it can realize the dynamic updating of the enterprise knowledge graph, avoid the recurrence of similar problems in incorrect allocation, and significantly improve the accuracy and processing efficiency of work order flow.

[0098] Based on the above Figure 2 In one example embodiment of the method shown, considering that the user's true intention in submitting feedback is to resolve product security issues as quickly as possible, but if no progress is seen or relevant personnel are contacted within a certain period after submitting feedback, feedback will be submitted again for the same product issue to remind the user that the product security issue needs to be resolved as soon as possible. Based on this, in step 103, a pending work order is sent to the business system corresponding to the responsible department information. The specific implementation process in this embodiment can be achieved through... Figure 9 Steps 601 and 602 shown are implemented.

[0099] Step 601: If the pending work order is not processed within the preset time period, receive new user feedback.

[0100] Step 602: If the problem category corresponding to the new user feedback is the same as the final problem category, add an expedited processing mark to the pending work order and send the pending work order with the expedited processing mark to the business system.

[0101] The preset duration can be understood as a pre-set threshold for determining whether a work order is to be processed earliest. Specifically, it can be implemented using a fixed time interval or a dynamically adjusted time window. This preset duration can be used to establish a monitoring benchmark for the work order processing status and trigger subsequent feedback verification processes.

[0102] New user feedback can be understood as feedback data submitted proactively by users after a preset time period, which can be obtained through mini-programs, online platforms, or voice interaction; this new user feedback can be used to verify the persistence and urgency of the aforementioned final issue categories.

[0103] The expedited processing identifier can be understood as a marker used to increase the priority of a work order. Specifically, it can be implemented in the form of a numeric label, color code, or metadata field. This expedited processing identifier can force the allocation of processing resources by changing the work order sorting rules or triggering an alarm mechanism.

[0104] Specifically, when a business system receives a pending work order from the central platform, it can track the work order's processing status and directly or indirectly provide feedback to the user's mini-program. This feedback can be sent directly to the user's mini-program or through the central platform. The work order processing status includes "accepted," "processing," and "completed." If the business system fails to track the work order's processing status within a preset timeframe, the user will not be able to view the progress through the mini-program. In this case, the user can submit new feedback to the central platform regarding the same product security issue to remind the relevant departments to handle it promptly. The central platform will then receive this new user feedback.

[0105] The central platform analyzes new user feedback and matches the analyzed problem categories with the final problem categories of the aforementioned products. If both belong to the same product security issue, it infers that the user urgently needs the relevant responsible department to handle the product security issue. In this case, an expedited processing identifier is embedded in the pending work order, and then the pending work order with the expedited processing identifier is sent to the business system to remind the corresponding responsible department to prioritize the handling of this pending work order.

[0106] The user feedback processing method provided in this application forms a closed-loop processing mechanism by automatically monitoring the timeliness of work order processing and combining it with the objective verification of new user feedback. This avoids subjective errors in human judgment and can forcibly increase the processing priority through expedited processing indicators, ensuring that critical issues affecting product safety receive priority processing resources and reducing safety risks caused by processing delays.

[0107] Based on the above Figure 1 In one example embodiment of the method shown, considering that users cannot know the processing status of work orders during the user feedback processing process, resulting in information opacity and a poor user experience, the processing status of work orders can be tracked in real time and feedback can be provided to users throughout the entire process from work order generation to work order completion. Based on this, the embodiments of this application can also provide a status tracking and user feedback process, which can be implemented in this embodiment through the following steps.

[0108] The system retrieves the processing status of pending work orders from the business system and sends the processing status feedback to the user client.

[0109] The work order processing status can include received, processing, and completed.

[0110] Specifically, the central platform obtains the processing status of pending work orders from the business system to ensure that users can promptly understand the progress of their feedback. This acquisition process can be implemented in various ways. For example, the central platform can periodically query the processing status of currently pending work orders by calling the Application Programming Interface (API) provided by the business system; alternatively, the business system can proactively notify the central platform through message queues or callback mechanisms when the work order processing status changes; furthermore, the central platform can deploy a listening service to continuously monitor the work order processing status or logs in the business system. Once an update to the work order processing status is detected, it is captured and then fed back to the user's client. The purpose is to present the obtained work order processing status to the user clearly and promptly, thereby improving the user experience. For instance, the central platform can send status updates to users in real time through the application's (APP) push notification function; or, it can provide a "My Work Orders" or "Processing Progress" module on the user's logged-in web page or mini-program interface for users to actively query; or, it can send work order processing status change notifications to users via SMS or email.

[0111] It's important to note that the business system can track the processing status of pending work orders, such as "received," "processing," and "completed." Furthermore, the business system can provide real-time feedback on the processing status to the user's client or front-end mini-program, allowing users to view the processing progress and forming a complete closed loop. For example, when the business system determines that the responsible department has completed the processing of the pending work order, it can also provide feedback on the processing result to the user's client or front-end mini-program (e.g., "The R&D department has confirmed that the product design will be optimized"). This closed-loop feedback tracking process improves the user experience and constructs a complete user feedback processing chain. For example, refer to Figure 10 The diagram shown illustrates the work order processing tracking and user feedback process, as follows: Figure 10 As shown, the business system can track the processing status of pending work orders and feed back the "processing" status to the central platform, which then relays it back to the user's end so that the user can view the processing progress.

[0112] Therefore, it can be understood that by integrating the acquisition and feedback mechanism of work order processing status into the user feedback processing flow, a closed loop of information flow is achieved. After the central platform receives user feedback from multiple modalities, and processes it through identification model analysis, problem category determination, responsible department matching, and generation of pending work orders, the pending work order is not only sent to the corresponding business system for processing, but also continuously retrieves the latest processing status from the business system. Once the work order status changes, such as from "pending" to "processing" or "completed," the central platform immediately transmits these updated work order processing statuses to the client used by the user who initiated the feedback. This mechanism ensures that users can understand the entire processing progress of their feedback in real time and transparently, thus effectively solving the problem of information asymmetry. The user feedback processing flow can also extend from front-end intelligent identification and work order generation to back-end processing status tracking and feedback, forming a complete, efficient, and user-friendly service chain, significantly improving users' perception of service quality and satisfaction.

[0113] The user feedback processing method provided in this application effectively solves the problem of information opacity during the user feedback processing process by obtaining the processing status of pending work orders from the business system and promptly providing feedback to the user client. Users can keep track of the latest progress of work orders in real time, thereby significantly improving user experience and satisfaction. This proactive feedback mechanism not only enhances users' trust in the service process but also avoids user anxiety caused by information delays, making the entire user feedback processing process more humanized and efficient. This achieves intelligent management of the entire chain from problem discovery to resolution feedback, greatly optimizing the interactive experience between users and product services.

[0114] Based on the above Figure 1 In one example embodiment of the method shown, considering that the recognition model may experience a decline in accuracy due to a lack of continuous training, making it unable to adapt to new types of user feedback and resulting in degraded processing performance, a closed-loop data feedback and model optimization mechanism can be constructed to effectively address the problems of decreased accuracy and insufficient adaptability of the recognition model. Based on this, the user feedback processing method provided in this embodiment may further include a model self-learning optimization process, which can be implemented in this embodiment through the following steps.

[0115] Record the entire process of user feedback processing data and write the entire process data into the training dataset; the training dataset is used to incrementally train the recognition model periodically.

[0116] The data for the entire user feedback processing process includes target user feedback, analysis results, responsible department information, and confirmation information on the processing results of pending work orders.

[0117] The processing result confirmation information can indicate that the processing result of the pending work order has been finally confirmed, so as to remind the central platform that the corresponding pending work order has been processed.

[0118] Specifically, after receiving user feedback from multiple modalities, the central platform uses a recognition model to analyze this feedback, determine the final problem category of the product, identify the matching responsible department, and generate a pending work order to be sent to the corresponding business system. Based on this, the entire process of user feedback processing is comprehensively recorded, including the target user feedback, the analysis results of the target user feedback, the responsible department information, and the confirmation information of the pending work order processing results. This complete process data is systematically written into a continuously updated training dataset. This training dataset serves as the "learning material" for the recognition model, used for periodic incremental training (active learning) to continuously improve classification and recognition accuracy. This allows the recognition model to continuously learn and correct itself from actual user feedback processing instances. For example, if the recognition model initially misclassifies a problem, but the final confirmation information of the pending work order reveals the correct classification and responsible department, this information will be included in the training dataset and used in subsequent incremental training to correct the classification logic of the recognition model. This continuous feedback mechanism, based on real processing results, enables the identification model to constantly adapt to new user feedback patterns, product problem types, and changes in business processes, thereby maintaining and improving its identification accuracy and robustness.

[0119] The user feedback processing method provided in this application effectively solves the problems of decreased accuracy and inability to adapt to new feedback types caused by a lack of continuous training in the recognition model by constructing a closed-loop data feedback and model optimization mechanism. Through comprehensive recording and utilization of data throughout the entire user feedback processing process, the central platform can provide the recognition model with continuous, authentic, and validated training samples. This closed-loop incremental training mechanism enables the recognition model to continuously learn and evolve, adapting promptly to product updates, changes in user feedback habits, and emerging problem types. This significantly improves the model's accuracy in recognizing and classifying user feedback, ultimately ensuring that work orders are more accurately assigned to the correct responsible departments. This improves the overall efficiency of user feedback processing and user satisfaction, and avoids resource waste and processing delays caused by model performance degradation.

[0120] For example, refer to Figure 11 The flowchart illustrating the user feedback processing method is shown below. Figure 11As shown, users submit text and image feedback via a mini-program. The central platform performs text parsing and image recognition, comprehensively judging the product's problem category from the parsing results to obtain the final problem category. This final problem category is then mapped to the corresponding responsible department information, generating and distributing a work order. When the final problem category is a security risk issue, the corresponding responsible department is instructed to prioritize handling the work order. After the responsible department completes the processing, the business system tracks the work order processing status and provides feedback on the status and result to the user. The central platform can record the entire user feedback process—from target user feedback to parsing results, responsible department processing, and final confirmation—for self-learning optimization of the recognition model. By combining the multimodal user feedback (text and image feedback) for opinion classification and department mapping, automatic routing of user feedback is achieved, ensuring that user feedback is automatically delivered to the appropriate responsible department. The specific processes involved can be referred to in the aforementioned embodiment, and will not be elaborated further here.

[0121] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0122] Reference Figure 12 This application also provides a user feedback processing system, including: a central platform, a user terminal, and multiple business systems, wherein: The user terminal is used to receive user feedback from multiple modalities and send user feedback from multiple modalities to the central platform.

[0123] The central platform is used to call the recognition model to parse user feedback from multiple modalities and obtain the problem category corresponding to each user feedback.

[0124] The central platform is also used to determine the final problem category of the product from the problem categories corresponding to various user feedback opinions, determine the responsible department information that matches the final problem category, and generate pending work orders based on the target user feedback opinions corresponding to the final problem category, the analysis results of the target user feedback opinions, and the responsible department information; and send the pending work orders to the business systems corresponding to the responsible department information. The business system is used to receive and process the pending work order.

[0125] The user-side can be understood as a front-end interactive module that collects multimodal feedback from users. Specifically, it can be implemented using a mini-program or application programming interface that integrates text input, image upload, and voice collection functions, and supports unified access to multimodal data.

[0126] It should be noted that, for user feedback received from multiple modalities, the central platform can refer to the aforementioned method implementation embodiment with the central platform as the execution subject to determine the final problem category of the product and generate pending work orders.

[0127] In one possible implementation, the central platform is also used to obtain the processing results of pending work orders from the business system and send the processing results to the user terminal.

[0128] Specifically, after the business system completes the work order processing, it can proactively report the processing result to the central platform. Alternatively, the central platform can proactively retrieve the processing result through a predefined data interface, such as sending a result query request to the business system at fixed intervals or triggering a result callback notification when the business system completes processing. The obtained processing result can flow back to the user's mini-program through the message channel registered on the user's end, forming a closed loop. Therefore, users do not need to actively query or wait for manual feedback; they can continuously check the processing progress after submitting feedback. For example, refer to... Figure 13 The diagram shown illustrates the closed-loop processing of user feedback. Figure 13 As shown, the central platform can not only receive the work order processing status from the business system, but also the processing results of pending work orders from the business system, and then feed back the work order processing status or results to the user terminal so that the user can view the processing progress and results.

[0129] It should be noted that, in addition to sending processing results to the user terminal, the central platform can also receive the work order processing status tracked and reported by the business system, and then send the work order processing status to the user terminal so that the user can check the work order processing progress, forming a complete closed loop. The specific process for checking the work order processing status can be referred to the aforementioned embodiment. It will not be repeated here.

[0130] Figure 14 A schematic diagram of a hardware architecture suitable for implementing embodiments of this application is shown, such as... Figure 14 As shown, the electronic device specifically includes a processor 701, a communication interface 702, a memory 703, and a communication bus 704. The processor 701, communication interface 702, and memory 703 communicate with each other via the communication bus 704. The memory 703 stores programs that can be executed by the processor 701. The processor 701 executes the programs stored in the memory 703 to implement… Figure 2 The steps of the method shown.

[0131] The communication bus 704 mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 704 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 14 The bus is represented by a single thick line, but this does not imply that there is only one bus or one type of bus. Communication interface 702 is used for communication between the aforementioned electronic device and other devices. Memory 703 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor 701. The aforementioned processor 701 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc., or it may be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0132] In another embodiment of this application, a computer-readable storage medium is provided, which stores a computer program that, when run on a computer, causes the computer to perform the actions described in the above embodiments. Figure 2 The steps of the method shown.

[0133] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape, etc.), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0134] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for processing user feedback, characterized in that, The method includes: Receive user feedback from multiple modalities, call the recognition model to parse the user feedback from multiple modalities to obtain the problem category corresponding to each user feedback, and determine the final problem category of the product from the problem categories corresponding to each user feedback; Determine the responsible department information that matches the final problem category, and generate a work order to be processed based on the target user feedback corresponding to the final problem category, the parsing result of the target user feedback, and the responsible department information; Send the pending work order to the business system corresponding to the responsible department information.

2. The method according to claim 1, characterized in that, Each piece of user feedback corresponds to a parsing result, which includes the problem category and product problem information. Determining the final problem category of the product from the problem categories corresponding to each piece of user feedback includes: Based on the problem categories and product fault information included in each of the analysis results, the similarity between the analysis results is determined. If there are parsing results with a similarity higher than the similarity threshold, then the final problem category is determined based on the parsing results with a similarity higher than the similarity threshold; If the similarity is lower than the similarity threshold, the confidence level of each user feedback opinion is calculated, and the final problem category is determined based on the parsing result of the user feedback opinion with the highest confidence level.

3. The method according to claim 1, characterized in that, The step of generating a pending work order based on the target user feedback corresponding to the final problem category, the parsing results of the target user feedback, and the responsible department information includes: Identify the danger characteristics in each of the target user feedback opinions, and determine the risk level of the target user feedback opinions based on the danger characteristics information; The pending work order is generated based on the target user feedback corresponding to the final problem category, the analysis results of the target user feedback, the responsible department information, and the risk level.

4. The method according to claim 3, characterized in that, The step of identifying hazard characteristic information in each of the target feedback opinions and determining the risk level of the target feedback opinions based on the hazard characteristic information includes: Identify dangerous image features and / or dangerous keywords in the target user feedback; The danger rating of the dangerous image features and / or the dangerous keywords is determined according to the rating mapping relationship; the rating mapping relationship is used to characterize the danger rating corresponding to different dangerous image features and the danger rating corresponding to different dangerous keywords. The risk level of the target feedback opinion is determined based on the hazard score of the hazard image features and / or the hazard keywords.

5. The method according to any one of claims 1 to 4, characterized in that, The determination of the responsible department information matching the final problem category includes: From the pre-built enterprise knowledge graph, find the name of the responsible department corresponding to the final problem category of the product, and determine the information of the responsible department based on the found responsible department name; The enterprise knowledge graph is used to represent the problem categories of different products and the names of the responsible departments for handling different problem categories.

6. The method according to claim 5, characterized in that, The method further includes: Receive the processing result of the pending work order; If the processing result indicates that the information of the responsible department does not match the actual information of the responsible department, then the enterprise knowledge graph is adjusted according to the information of the actual responsible department.

7. The method according to any one of claims 1 to 4, characterized in that, Sending the pending work order to the business system corresponding to the responsible department information includes: If the pending work order is not processed within the preset time period, new user feedback will be received. If the issue category corresponding to the new user feedback is the same as the final issue category, then an expedited processing identifier is added to the pending work order, and the pending work order with the expedited processing identifier is sent to the business system.

8. The method according to any one of claims 1 to 4, characterized in that, The method further includes: The system obtains the processing status of the pending work orders from the business system and reports the processing status to the user client.

9. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Record the entire process data of user opinion processing and write the entire process data of user opinion processing into the training dataset; the training dataset is used to perform incremental training of the recognition model periodically; The data for the entire user feedback processing process includes the target user feedback, the analysis results, the responsible department information, and the confirmation information of the processing results of the pending work orders.

10. A user feedback processing system, characterized in that, This includes: a central platform, user terminals, and multiple business systems. The user terminal is used to receive user feedback from multiple modalities and send the user feedback from multiple modalities to the central platform. The central platform is used to call the recognition model to parse the user feedback opinions of the multiple modalities and obtain the problem category corresponding to each user feedback opinion; The central platform is also used to determine the final problem category of the product from the problem categories corresponding to each user feedback, determine the responsible department information that matches the final problem category, and generate a work order to be processed based on the target user feedback corresponding to the final problem category, the parsing result of the target user feedback, and the responsible department information. Send the pending work order to the business system corresponding to the responsible department information; The business system is used to receive and process the pending work orders.

Images