Evaluation method and device for understanding feedback semantics and performing intelligent feedback tracking based on NLP technology

Abstract

The invention provides an evaluation method and device for understanding feedback semantics and conducting intelligent feedback tracking based on the NLP technology, and the method comprises the steps: receiving a feedback text, carrying out the feature extraction of the feedback text, and obtaining a semantic feature and a classification result; determining a to-be-processed task according to the feedback text, the semantic feature and the classification result; according to the predicted processing time of each to-be-processed task, adjusting a processing node corresponding to the to-be-processed task; and distributing the to-be-processed task to the corresponding processing node, so that the processing node processes the to-be-processed task. Automatic feature extraction and classification are achieved through the NLP technology, the feedback processing speed is increased, and the system can quickly respond to a large amount of feedback information and respond to user requirements in time. The processing nodes are intelligently adjusted according to the task predicted processing time, so that the capabilities and load conditions of the tasks and the processing nodes are matched, and the overall processing efficiency and resource utilization rate of the system are improved.

Description

Technical Field

[0001] This application relates to the field of semantic processing technology, and in particular to an evaluation method and apparatus for understanding feedback semantics and performing intelligent feedback tracking based on NLP technology. Background Technology

[0002] In today's digital age, various products and services extensively collect user feedback to achieve continuous improvement. However, faced with massive amounts of feedback text data, traditional manual processing methods are inefficient and inaccurate, making it difficult to meet the needs for rapid response and precise processing. Summary of the Invention

[0003] This application aims to at least partially address one of the technical problems in the related art.

[0004] Therefore, this application proposes a method and apparatus.

[0005] One embodiment of this application proposes an evaluation method based on NLP technology to understand feedback semantics and perform intelligent feedback tracking, including: Receive feedback text, extract features from the feedback text, and obtain semantic features and classification results; The feedback text, semantic features, and classification results are stored in the blockchain, and the task to be processed is determined based on the feedback text, semantic features, and classification results. Adjust the processing node corresponding to each pending task according to the estimated processing time of each pending task; The task to be processed is assigned to the corresponding processing node so that the processing node can process the task to be processed.

[0006] Optionally, the step of extracting features from the feedback text to obtain semantic features and classification results includes: The feedback text is filtered to remove text of a preset type. The filtered feedback text is normalized to obtain a text vector; Feature extraction is performed on the text vector to obtain a fused feature vector; The classification is performed based on the fused feature vector to obtain the classification result, wherein the classification result represents the type of problem reflected in the feedback text.

[0007] Optionally, the step of extracting features from the text vector to obtain a fused feature vector includes: Extract the term frequency features and inverse document frequency features from the text vector; The fused feature vector is obtained by fusing the term frequency features and the inverse document frequency features.

[0008] Optionally, the method further includes: The first priority of each processing node is determined based on its historical accuracy and current load. The tasks to be processed corresponding to each processing node are determined according to the first priority.

[0009] Optionally, adjusting the processing node corresponding to each pending task based on its estimated processing time includes: Based on the problem type in historical task data, feedback text length, historical average processing time for similar problems, current load of processing nodes, and historical accuracy of processing nodes, the estimated processing time is predicted. Based on the estimated processing time and the load factor of the processing node, determine the second priority of the tasks to be processed in the processing node; Based on the load factor and the second priority of the tasks to be processed, at least a portion of the tasks to be processed are migrated to other processing nodes.

[0010] Optionally, the formula for calculating the second priority is:

[0011] Where S is the second priority, The estimated processing time is denoted as , and LoadFactor is the load factor.

[0012] Optionally, migrating at least a portion of the tasks to be processed to other processing nodes based on the load factor and the second priority of the tasks to be processed includes: If the load coefficient of the processing node is greater than or equal to a preset first coefficient threshold, the task with a second priority lower than a preset priority threshold in the processing node is determined as the target task. The target task is migrated to a processing node whose load coefficient is less than a preset second coefficient threshold.

[0013] Another embodiment of this application proposes an evaluation device for understanding feedback semantics and performing intelligent feedback tracking based on NLP technology, comprising: The classification module is used to receive feedback text, extract features from the feedback text, and obtain semantic features and classification results. The task determination module is used to store the feedback text, semantic features and classification results into the blockchain, and determine the task to be processed based on the feedback text, semantic features and classification results; The node determination module is used to adjust the processing node corresponding to each task to be processed according to the estimated processing time of each task to be processed. The processing module is used to assign the task to be processed to the corresponding processing node so that the processing node can process the task.

[0014] Another embodiment of this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method described in the foregoing aspect.

[0015] Another embodiment of this application proposes a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the foregoing aspect.

[0016] Another embodiment of this application proposes a chip including processing circuitry configured to perform the method described in one aspect above.

[0017] Another embodiment of this application proposes a computer program product that, when executed by a processor, implements the method described in the foregoing aspect.

[0018] The evaluation method, apparatus, electronic device, chip, and storage medium proposed in this application, which are based on NLP technology to understand feedback semantics and perform intelligent feedback tracking, have beneficial effects.

[0019] Improve feedback processing efficiency: By leveraging NLP technology to automate feature extraction and classification, the workload of manually processing feedback text is greatly reduced, and the speed of feedback processing is significantly improved, enabling the system to quickly respond to a large amount of feedback information and respond to user needs in a timely manner.

[0020] Enhancing data security and traceability: Utilizing blockchain to store feedback-related data not only ensures data security and prevents malicious tampering, but also allows for a clear view of the entire process from submission to processing through the traceability feature of blockchain, facilitating problem analysis and accountability.

[0021] Optimize task allocation rationality: Intelligently adjust processing nodes according to the estimated processing time of tasks, so that the capacity and load of processing nodes are matched with the tasks, effectively avoiding resource waste or task overload of processing nodes, and improving the overall processing efficiency and resource utilization of the system.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart illustrating an evaluation method based on NLP technology for understanding feedback semantics and performing intelligent feedback tracking, provided in an embodiment of this application; Figure 2 A flowchart illustrating an evaluation method based on NLP technology for understanding feedback semantics and performing intelligent feedback tracking, provided in an embodiment of this application; Figure 3 A schematic diagram of the structure of an evaluation device based on NLP technology for understanding feedback semantics and performing intelligent feedback tracking, provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a chip proposed in an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0025] With the popularization of digital services, the number of user feedback (such as product malfunction reports, experience suggestions, and data anomaly feedback) is growing exponentially. Traditional feedback tracking methods are no longer sufficient to meet the needs of efficient, accurate, and reliable processing. At the same time, the development of technologies such as NLP and blockchain has provided new paths to solve these pain points. The specific background is as follows: With the popularization of digital services, the number of user feedback (such as product malfunction reports, experience suggestions, and data anomaly feedback) is growing exponentially. Traditional feedback tracking methods are no longer sufficient to meet the needs of efficient, accurate, and reliable processing. At the same time, the development of technologies such as NLP and blockchain has provided new paths to solve these pain points. The specific background is as follows: (I) The core pain points of traditional feedback tracking methods 1. Low efficiency due to manual labor Traditional feedback processing relies on manual completion of the entire process of "text reading → problem classification → task assignment": the manual classification speed is about 50-100 items / hour. When the feedback volume exceeds 1000 items / day, classification delays are prone to occur (average response time ≥ 4 hours); task assignment depends on the administrator's experience, and there are problems such as "repeated assignment to high-load nodes" and "assigning professional problems to non-relevant personnel", which leads to a longer processing cycle (average processing time ≥ 24 hours), and the error rate of manual operation (classification error, assignment error) is as high as 15%-20%.

[0026] 2. Insufficient semantic understanding accuracy Traditional methods often use "keyword matching" for semantic parsing (e.g., if it contains "crash", it is classified as "functional failure"). This method cannot handle ambiguous text (e.g., "slow login" could be "performance problem" or "network problem"), multi-intent feedback (e.g., "APP crashes and data synchronization fails" contains two types of problems) and industry terms (e.g., "clearing delay" in the financial field and "data anonymization failure" in the medical field). As a result, the classification accuracy is generally less than 75%, and the subsequent processing direction deviates from the user's actual needs.

[0027] 3. Poor data reliability and traceability Feedback data is mostly stored in centralized databases, which poses risks such as "data tampering" (e.g., hiding negative feedback) and "progress fraud" (e.g., falsely reporting processing progress). Furthermore, the data of the entire feedback-processing process is scattered across different systems (e.g., feedback forms, work order systems, acceptance records), lacking a unified evidence storage mechanism. In case of disputes, it is difficult to trace responsibility, resulting in low user trust (the survey shows that only 38% of users believe that the feedback processing process is transparent).

[0028] 4. Blind task scheduling and time forecasting Traditional scheduling lacks the ability to predict processing time and only allocates tasks on a "first-come, first-served" basis, resulting in resource misallocation where "simple problems occupy high-capacity nodes" and "complex problems are assigned to low-efficiency nodes." At the same time, it cannot predict the processing cycle, so users cannot obtain clear feedback time, and the satisfaction score (NPS) is less than 50 out of 100.

[0029] (II) Current Status and Shortcomings of Related Technology Development 1. Limitations of NLP technology in feedback processing Traditional semantic encoder models (such as BERT) or neural network models (such as LSTM) have limitations. Using only one of these algorithms can easily expose the following problems: While the pure BERT model can understand context, it has poor adaptability to small sample industry data (e.g., when the vertical domain feedback data is less than 10,000, the accuracy drops to below 65%); the pure LSTM model cannot capture long text dependencies (e.g., in feedback exceeding 50 characters, the semantic extraction completeness drops by 40%); and most solutions do not integrate traditional features (such as TF-IDF), resulting in weak recognition ability for low-frequency industry terms, making it difficult to meet the needs of accurate classification.

[0030] 2. Application gaps of blockchain technology in the field of evidence preservation While blockchain has been used for data storage (such as electronic contracts and logistics records), it suffers from a "functional fragmentation" problem in feedback tracking scenarios: existing solutions only implement the on-chain recording of feedback text, without combining "semantic classification results, task allocation rules, and progress updates" with smart contracts, thus failing to trigger automated execution; moreover, most of them use public chains (such as Ethereum), which have slow transaction speeds (TPS≤15) and privacy leakage risks (feedback data is publicly visible), making them unsuitable for enterprise-level scenarios.

[0031] 3. Shortcomings in the accuracy of time prediction and scheduling technologies Traditional time prediction methods often use rule engines (such as "functional failure handling time = 24 hours" and "experience suggestion handling time = 12 hours"), without considering dynamic factors such as "problem complexity, node load, and historical processing efficiency", resulting in a prediction error rate of over 30%. Scheduling algorithms are mostly based on static load (such as only referring to the current number of tasks) without combining predicted processing time, failing to achieve a "load-efficiency" balance, and resulting in a node overload rate (load > load limit) as high as 25%.

[0032] In summary, traditional feedback tracing methods suffer from four major pain points: efficiency, accuracy, reliability, and scheduling. The individual application of existing NLP, blockchain, and predictive scheduling technologies cannot form a closed-loop solution. There is an urgent need to build an integrated technical solution of "semantic understanding - smart contracts - dynamic scheduling" to achieve full-process automation, intelligence, and reliability of feedback tracing.

[0033] The following description, with reference to the accompanying drawings, describes an evaluation method, apparatus, electronic device, chip, and storage medium based on NLP technology for understanding feedback semantics and performing intelligent feedback tracking, according to embodiments of this application.

[0034] Figure 1 This is a flowchart illustrating an evaluation method based on NLP technology for understanding feedback semantics and performing intelligent feedback tracking, as provided in an embodiment of this application.

[0035] As one implementation, the evaluation method based on NLP technology to understand feedback semantics and perform intelligent feedback tracking in this application embodiment can be configured in an evaluation device based on NLP technology to understand feedback semantics and perform intelligent feedback tracking. This evaluation device based on NLP technology to understand feedback semantics and perform intelligent feedback tracking can be applied to any electronic device so that the electronic device can perform the evaluation function based on NLP technology to understand feedback semantics and perform intelligent feedback tracking.

[0036] Among them, electronic devices can be any device with computing capabilities, such as mobile terminals, which can be hardware devices with various operating systems, touch screens and / or displays, such as mobile phones, tablets, personal digital assistants, wearable devices, etc.

[0037] As another implementation, the evaluation method based on NLP technology to understand feedback semantics and perform intelligent feedback tracking in this application embodiment can also be executed by a chip with processing capabilities. Chips include image signal processing chips (ISP), central processing units (CPU), application-specific integrated circuits (ASIC), digital signal processors (DSP), field-programmable gate arrays (FPGA), systems on a chip (SOC), reduced instruction set computers (RISC), etc., which will not be listed here.

[0038] It should be noted that all data collection operations related to users in this application are conducted with the user's authorization and in strict compliance with relevant laws and regulations such as privacy and security.

[0039] like Figure 1 As shown, the method may include the following steps: Step 101: Receive feedback text, extract features from the feedback text, and obtain semantic features and classification results; Step 102: Store the feedback text, semantic features, and classification results in the blockchain, and determine the task to be processed based on the feedback text, semantic features, and classification results; Step 103: Adjust the processing node corresponding to each task to be processed according to the estimated processing time of each task to be processed; Step 104: Assign the task to be processed to the corresponding processing node so that the processing node can process the task to be processed.

[0040] In this embodiment, in today's digital age, various products and services widely collect user feedback for continuous improvement. However, faced with massive amounts of feedback text data, traditional manual processing methods are inefficient and inaccurate, failing to meet the demands for rapid response and precise processing. An evaluation method based on NLP technology to understand feedback semantics and perform intelligent feedback tracking has emerged, aiming to leverage the powerful capabilities of natural language processing to achieve efficient interpretation and tracking of feedback information, thereby improving the quality of products and services.

[0041] Explanation of proper nouns: NLP (Natural Language Processing) technology: As an interdisciplinary field of computer science and linguistics, it is dedicated to enabling computers to understand, generate, and process human language, so that computers and humans can interact effectively through natural language.

[0042] Semantic features are abstract representations of the semantic information in feedback text. These features can accurately capture key information such as the meaning, emotion, and intention expressed in the text, providing a foundation for a deeper understanding of the feedback content.

[0043] Classification results: The classification is based on the nature and type of the problem reflected in the feedback text. By classifying, different feedback texts can be divided into corresponding categories, which makes it easier to formulate targeted processing strategies.

[0044] Blockchain: A distributed ledger technology that provides high security and reliability for data storage due to its decentralized, immutable, and traceable characteristics. In this method, it is used to store feedback text and its associated semantic features and classification results, ensuring the authenticity and integrity of the data.

[0045] Processing node: This can be understood as a unit in the system capable of processing specific tasks. It can be a server, computing device, or a specific software module. Each processing node differs in terms of processing capacity, resource configuration, and current load status.

[0046] Explanation of the principle: Feedback Text Reception and Feature Extraction: The system first receives feedback text from users, which may be presented in various forms, such as product reviews and service feedback. Then, advanced NLP techniques are used to perform in-depth feature extraction on the feedback text. Through a series of NLP techniques such as lexical analysis, syntactic analysis, and semantic understanding, semantic features are mined from the feedback text. Simultaneously, the text is classified based on predefined classification rules or machine learning models to obtain classification results. For example, for a feedback text about a mobile application, "The application takes a long time to load every time I open it, which greatly affects usability," feature extraction can yield semantic features such as "long application loading time," and the classification result might be "application performance problem."

[0047] Data storage and task determination: The extracted feedback text, semantic features, and classification results are stored together on the blockchain. The decentralized storage mechanism of the blockchain ensures that the data is backed up on multiple nodes and is tamper-proof, providing strong support for subsequent traceability and auditing. Based on this stored data, the system determines the tasks to be processed according to preset rules or algorithms. For example, if the classification result is "product functional defect," the corresponding task might be to arrange for the R&D team to conduct a detailed inspection and repair of that function.

[0048] Task Node Adjustment: Considering that the estimated processing time for different tasks may vary due to factors such as task nature and complexity, the system intelligently adjusts the corresponding processing nodes based on the estimated processing time of each task. Tasks with longer estimated processing times are tended to be assigned to processing nodes with stronger processing capabilities, more abundant resources, or lower current loads to optimize overall processing efficiency and avoid excessive backlog of tasks on a particular processing node.

[0049] Task Allocation and Processing: After the above adjustments, the system will accurately allocate tasks to the corresponding processing nodes. Upon receiving a task, the processing node will process it according to the established processing procedures and standards, thereby enabling the tracking and handling of feedback and ultimately achieving the goal of improving product or service quality.

[0050] Analysis of beneficial effects: Improve feedback processing efficiency: By leveraging NLP technology to automate feature extraction and classification, the workload of manually processing feedback text is greatly reduced, and the speed of feedback processing is significantly improved, enabling the system to quickly respond to a large amount of feedback information and respond to user needs in a timely manner.

[0051] Enhancing data security and traceability: Utilizing blockchain to store feedback-related data not only ensures data security and prevents malicious tampering, but also allows for a clear view of the entire process from submission to processing through the traceability feature of blockchain, facilitating problem analysis and accountability.

[0052] Optimize task allocation rationality: Intelligently adjust processing nodes according to the estimated processing time of tasks, so that the capacity and load of processing nodes are matched with the tasks, effectively avoiding resource waste or task overload of processing nodes, and improving the overall processing efficiency and resource utilization of the system.

[0053] Specific Implementation: Taking an online education platform as an example, a student submitted feedback text stating, "The course videos frequently buffer, severely impacting my learning progress." Upon receiving this feedback text, the system uses NLP technology to extract the semantic feature "course video buffering," classifying it as a "video playback problem." The feedback text, semantic feature, and classification result are stored in the blockchain. Based on this information, the task to be processed is determined to be checking the video playback system and resolving the buffering issue. Analysis indicates that this task is expected to take a long time, so the processing nodes are adjusted, and the task is assigned to the processing node corresponding to the professional team responsible for video technical support. After receiving the task, the team conducts a comprehensive check and optimization of the video playback system, resolving the buffering problem.

[0054] Optionally, the step of extracting features from the feedback text to obtain semantic features and classification results includes: The feedback text is filtered to remove text of a preset type. The filtered feedback text is normalized to obtain a text vector; Feature extraction is performed on the text vector to obtain a fused feature vector; The classification is performed based on the fused feature vector to obtain the classification result, wherein the classification result represents the type of problem reflected in the feedback text.

[0055] In this embodiment, the original feedback text often contains various complex information, and some of the pre-defined text types may be irrelevant to the core content of the feedback, or even interfere with subsequent feature extraction and classification. Therefore, it is necessary to screen and preprocess the feedback text before performing in-depth feature extraction to improve the accuracy and efficiency of subsequent processing.

[0056] Explanation of proper nouns: Preset text types: These refer to text types that are pre-defined based on business needs and common interfering factors, and are irrelevant to the core feedback content or have a negative impact on feature extraction. Examples include common greetings ("Hello," "Good morning," etc.), irrelevant advertising content ("Click here to learn more about offers"), repetitive or redundant expressions, etc.

[0057] Normalization: In text processing, normalization aims to convert data of different formats, sizes, or distributions into a unified standard form to facilitate subsequent analysis and processing by computers. For text, this typically involves converting the text into numerical text vectors, making different texts comparable within the same vector space.

[0058] Text vectors are digital representations of text in vector space. Each dimension of the vector represents a certain feature of the text. Text vectors can quantify and calculate the semantic information of text in numerical form, making it easier for computers to understand and process the text.

[0059] Fusion feature vector: A vector obtained by fusing multiple different features. This vector integrates information from multiple features and can more comprehensively and accurately describe the characteristics of the text, providing richer feature representations for subsequent tasks such as text classification.

[0060] Explanation of the principle: Text filtering: The system meticulously filters received feedback text, identifying and removing predefined text types based on preset rules and patterns. For example, it uses regular expression matching or machine learning-based text classification models to identify and remove greetings, advertising content, etc., from feedback text. Taking the feedback text "Hello, this software is really useful, but there are too many ads, which affects the user experience" as an example, the system will identify and remove "Hello" and possible advertising-related expressions, making the remaining text more focused on the core feedback content.

[0061] Normalization: The filtered feedback text is normalized, typically using word vector representation methods such as Word2Vec and GloVe, mapping each word in the text to a fixed-dimensional vector. By combining or operating on all word vectors, the entire text vector is obtained. This process transforms the text from natural language to a numerical vector form easily processed by computers, laying the foundation for subsequent feature extraction.

[0062] Feature Extraction: Feature extraction is performed on the obtained text vectors, focusing on word frequency (Word Frequency) and inverse document frequency (Inverse Document Frequency) features. Word frequency reflects how frequently each word appears in the text; words with higher frequencies are often closely related to the text's topic. Inverse document frequency measures the rarity of a word within the entire document set; words with higher rarity may have higher weight when distinguishing different text categories. For example, in a document set of feedback about tourist attractions, the word "attraction" has a high Word Frequency but a low Inverse Document Frequency; while the names of some specific attractions may have a low Word Frequency but a high Inverse Document Frequency.

[0063] Feature fusion and classification: The extracted word frequency features and inverse document frequency features are fused, for example, through concatenation or weighted summation, to obtain a fused feature vector. This fused feature vector combines information from both word frequency and inverse document frequency, providing a more comprehensive reflection of the text's features. Based on this fused feature vector, a pre-trained classification model (such as support vector machine, neural network, etc.) is used to classify the text, obtaining the classification result and clearly indicating the type of problem reflected by the text, such as "service quality issue" or "product function issue."

[0064] Analysis of beneficial effects: Improved feature extraction accuracy: By filtering text of preset types, interference information is effectively removed, making the semantic features extracted subsequently more accurately reflect the core content of the feedback text, thus improving the quality and accuracy of feature extraction.

[0065] Enhancing text processing efficiency: Normalization transforms text into a uniform vector form, simplifying subsequent processing and improving the computer's efficiency in processing text. Simultaneously, the standardized vector representation facilitates feature extraction and classification using various machine learning algorithms, further improving the overall efficiency of the text processing process.

[0066] Optimizing classification result quality: The fused feature vector obtained by combining word frequency features and inverse document frequency features integrates the advantages of both features, enabling a more comprehensive and accurate description of text characteristics. Classification based on this vector yields more precise results, helping to more accurately pinpoint the type of feedback issue and providing a clearer direction for subsequent processing.

[0067] Specific Implementation: For a feedback text about hotel accommodation, "Hello, the hotel is easy to find, and the rooms are very clean, but the breakfast selection is too limited; I hope you can increase it. Also, I saw that your hotel has promotional activities; could you please provide details?", the system first filters out "Hello" and the content about hotel promotional activities. The remaining text, "The hotel is easy to find, and the rooms are very clean, but the breakfast selection is too limited; I hope you can increase it," is normalized and converted into a text vector using Word2Vec technology. Word frequency features are extracted from the text vector, revealing that words such as "breakfast," "selection," and "limited" appear frequently. Inverse document frequency features are extracted to determine the rarity of these words in the hotel accommodation feedback document set. These two features are fused to obtain a fused feature vector, which is then used for classification using a pre-trained support vector machine model, resulting in the classification result "Hotel Catering Service Problem."

[0068] Optionally, the step of extracting features from the text vector to obtain a fused feature vector includes: Extract the term frequency features and inverse document frequency features from the text vector; The fused feature vector is obtained by fusing the term frequency features and the inverse document frequency features.

[0069] In this embodiment, term frequency (TF) and inverse document frequency (IVF) features play important roles in text feature extraction, reflecting the characteristics of the text from different perspectives. Using only one feature may not fully and accurately describe the essential features of the text; fusing both yields a more representative and discriminative fused feature vector, significantly improving the understanding and classification accuracy of the feedback text. **Definitions:** **Term Frequency (TF):** The frequency of each word in a given text. It intuitively reflects the importance of words in the text and their relevance to the text's topic. Generally, words closely related to the text's topic tend to appear frequently, thus TF features can capture the core content of the text. **Inverse Document Frequency (IVF):** An indicator used to measure the rarity of a word in the entire document set. If a word appears in most documents, its IVF is low, indicating a small contribution to distinguishing different document categories; conversely, if a word appears only in a few documents, its IVF is high, indicating a high weight in distinguishing document categories. Fusion Feature Vector: A vector generated by organically combining word frequency features and inverse document frequency features. It integrates information from both features, providing a more comprehensive and in-depth description of the semantic features of the text, and offering richer and more accurate feature representations for text classification and understanding. Explanation of the Principle: Word Frequency Feature Extraction: The system analyzes the text vector, counts the number of times each word appears in the text, and then converts it into word frequency features. For example, in the text "This phone has a powerful camera function, and the photo quality is also very good," the word "take photos" appears twice. After a certain normalization process (such as dividing by the total number of words in the text), the word frequency feature value of "take photos" is obtained. Word frequency features can intuitively reflect which words in the text appear frequently, thus helping us understand the core theme of the text. Inverse Document Frequency Feature Extraction: Calculate the inverse document frequency of each word in the context of the entire document set. Assuming there are N documents in the document set, and n documents contain a certain word w, then the inverse document frequency IDF(w) of that word = For example, in a collection of 100 mobile feedback documents, the word "take a photo" appears in 20 documents, then the inverse document frequency of "take a photo" is... Inverse document frequency (IVF) features can highlight words that are relatively rare in a document set, which may be important for distinguishing different types of text. Feature fusion: The extracted word frequency features and IVF features are fused to obtain a fused feature vector. Common fusion methods include concatenation, that is, concatenating the word frequency feature vector and the IVF feature vector end to end; or weighted summation, assigning different weights to word frequency features and IVF features according to actual needs, and then performing a summation operation. Through this fusion method, the fused feature vector contains both the frequency information of words in the text and the rarity information of words in the document set, which can more comprehensively and accurately describe the semantic features of the text. Beneficial effect analysis: Improved feature comprehensiveness: The fusion of word frequency features and IVF features allows the fused feature vector to comprehensively describe the characteristics of the text from different dimensions, making up for the limitations of single feature representation and more comprehensively reflecting the semantic information of the text. Enhanced Classification Accuracy: Since the fused feature vector contains richer and more accurate textual feature information, it can more effectively distinguish different types of feedback text when classifying text, thus significantly improving classification accuracy and providing a more reliable basis for subsequent feedback processing. Improved Text Understanding: By comprehensively considering word frequency and inverse document frequency, the system can more deeply understand the meaning and theme of feedback text, helping to uncover potential information in the text and better grasp the user's feedback intent. Specific Example: Taking the mobile application feedback text "This application's interface design is very simple and the operation is very convenient, but there are too many ads, affecting the user experience" as an example. First, word frequency features are extracted, such as the word frequencies of "application," "interface," "operation," and "ads." Then, the inverse document frequencies of these words are calculated in a set containing numerous mobile application feedback documents. Assuming that the word "ads" has a high inverse document frequency in the set, it indicates that it may play an important role in distinguishing different types of application feedback. The word frequency feature vector and the inverse document frequency feature vector are fused by weighted summation to obtain a fused feature vector. This fused feature vector integrates the information of word frequency and inverse document frequency, more comprehensively describing the characteristics of the feedback text. Based on this fused feature vector, a classification model can more accurately classify the feedback text as an "application advertising question".

[0070] Optionally, the method further includes: The first priority of each processing node is determined based on its historical accuracy and current load. The tasks to be processed corresponding to each processing node are determined according to the first priority.

[0071] In this embodiment, during the processing of feedback tasks, different processing nodes vary in processing capacity and quality due to factors such as their hardware configuration, personnel skill levels, and historical processing experience. Simultaneously, the current load of each processing node also affects its ability to handle new tasks. Therefore, to achieve efficient and accurate task processing, it is necessary to determine the priority of each processing node based on its historical accuracy and current load, and allocate tasks accordingly. Definitions: Historical Accuracy: Refers to the accuracy of a processing node in handling similar tasks in the past, usually measured as the ratio of the number of correctly processed tasks to the total number of processed tasks. Historical accuracy reflects the processing capacity and quality of a processing node; a higher accuracy indicates greater reliability in handling related tasks. Current Load: Represents the current workload or resource consumption of a processing node. A higher load indicates a busier processing node with a relatively weaker ability to handle new tasks; conversely, a lower load indicates a node's greater capacity to handle new tasks. First Priority: A priority determined based on a combination of the processing node's historical accuracy and current load, used to measure the priority of a processing node in handling tasks. The higher the first priority, the more frequently the processing node is assigned tasks. Explanation: First Priority Determination: The system collects historical data from each processing node, including the types and quantities of tasks processed, and the number of tasks processed correctly, to calculate the historical accuracy of each node. Simultaneously, the system monitors the current load of processing nodes in real time, for example, by monitoring metrics such as CPU utilization, memory usage, and task queue length. Then, based on a preset algorithm, the first priority of each processing node is determined by combining historical accuracy and current load. For example, a weighted summation method can be used, assigning different weights to historical accuracy and current load to calculate the first priority. If the weight of historical accuracy is 0.6, the weight of current load is 0.4, the historical accuracy of processing node A is 0.9, and the corresponding value for current load is 0.3 (the lower the load, the smaller the value), then the first priority of processing node A is 0.6 * 0.9 + 0.4 * 0.3 = 0.66. Task Allocation: Based on the determined first priority, the system allocates the tasks to be processed to each processing node. The highest priority processing nodes will be assigned tasks first. This ensures that tasks are prioritized for processing nodes with strong processing capabilities and relatively low current load, thereby improving the efficiency and accuracy of task processing. For example, for a new task to be processed, the system will first consider assigning it to the highest priority processing node. If that node can handle the task, it will be assigned to it; if that node is too overloaded, the next highest priority processing node will be considered, and so on.

[0072] Beneficial Effects Analysis: Optimized Task Allocation Efficiency: By comprehensively considering the historical accuracy and current load of processing nodes to determine the first priority, tasks can be preferentially assigned to the most suitable nodes for handling them, avoiding blind task allocation and improving the overall efficiency of task processing. Guaranteed Processing Quality: Assigning tasks to processing nodes with high historical accuracy helps ensure the quality of task processing, reduces the occurrence of error processing, and thus better meets user expectations for feedback processing.

[0073] Balanced processing node load: Prioritize tasks based on current load, enabling the system to dynamically adjust task allocation and avoid situations where some processing nodes are overloaded while others are idle. This achieves load balancing among processing nodes and improves the overall stability and resource utilization of the system.

[0074] Specific Implementation: Assume a feedback processing system on an e-commerce platform has three processing nodes: Node A, Node B, and Node C. Node A has a historical accuracy rate of 90% in handling product quality-related feedback tasks, and its current CPU utilization is 60%. Node B has a historical accuracy rate of 85%, and its current CPU utilization is 40%. Node C has a historical accuracy rate of 80%, and its current CPU utilization is 30%. Using a preset algorithm (e.g., historical accuracy weight 0.7, current load weight 0.3), the first priority of Node A is calculated as follows: 0.7 × 0.9 + 0.3 × (1 - 0.6) = 0.75 The first priority for processing node B is 0.7 × 0.85 + 0.3 × (1 - 0.4) = 0.745 The first priority for processing node C is 0.7 × 0.8 + 0.3 × (1 - 0.3) = 0.71. When a new product quality feedback task arrives, the system first assigns the task to processing node A. If node A is unable to handle the task for some reason (such as the task queue being full), it is then assigned to processing node B, and so on.

[0075] Optionally, adjusting the processing node corresponding to each pending task based on its estimated processing time includes: Based on the problem type in historical task data, feedback text length, historical average processing time for similar problems, current load of processing nodes, and historical accuracy of processing nodes, the estimated processing time is predicted. Based on the estimated processing time and the load factor of the processing node, determine the second priority of the tasks to be processed in the processing node; Based on the load factor and the second priority of the tasks to be processed, at least a portion of the tasks to be processed are migrated to other processing nodes.

[0076] In this embodiment, in a real feedback processing scenario, the estimated processing time for each task is influenced by various factors, including the problem type, feedback text length, historical average processing time for similar problems, the current load of the processing node, and historical accuracy. Simultaneously, the load factor of the processing node also affects the task priority. To allocate tasks more rationally and improve overall processing efficiency, these factors need to be comprehensively considered to predict the estimated processing time, and the corresponding processing nodes for each task need to be adjusted accordingly.

[0077] Explanation of proper nouns: Historical task data: Information related to past processed feedback tasks recorded by the system, including problem type, feedback text length, processing time, etc. This data provides a reference for predicting the estimated processing time of new tasks.

[0078] Estimated processing time: An estimate of the time required to process the current pending task, calculated by taking into account multiple factors, and used to reasonably arrange tasks and allocate processing nodes.

[0079] Load factor: A coefficient that measures the load level of a processing node. It reflects the current busyness of the processing node. The higher the load factor, the busier the processing node is.

[0080] Second priority: The task priority is determined by combining the estimated processing time and the load factor of the processing node. It is used to determine the processing order of tasks within the processing node and whether they need to be migrated to other processing nodes.

[0081] Explanation of the principle: Estimated Processing Time Prediction: The system extracts information related to the current task from historical task data, such as the problem type and feedback text length. For similar problems, it refers to the historical average processing time, combined with the current load and historical accuracy of the processing node, to predict the estimated processing time of the current task. For example, if the current task is feedback on software function optimization, and the historical average processing time for similar problems is 2 hours, and the current processing node has a high historical accuracy (meaning potentially faster processing speed) and a low current load (relatively abundant resources), then the estimated processing time may be relatively shorter; conversely, if the processing node has a high current load and a low historical accuracy, the estimated processing time may be longer.

[0082] Optionally, the formula for calculating the second priority is:

[0083] Where S is the second priority, The estimated processing time is denoted as , and LoadFactor is the load factor.

[0084] Based on the load factor and the second priority of the tasks, the system determines whether some tasks need to be migrated to other processing nodes. When the load factor of a processing node is greater than or equal to a preset first coefficient threshold, it indicates that the processing node is under high load. In this case, tasks with a second priority lower than a preset priority threshold in that processing node are identified as target tasks. These target tasks are then migrated to processing nodes with a load factor lower than the preset second coefficient threshold to balance the load of each processing node and improve overall processing efficiency.

[0085] Task Migration: Based on the load factor and the second priority of the tasks, the system determines whether some tasks need to be migrated to other processing nodes. When the load factor of a processing node is greater than or equal to the preset first coefficient threshold, it indicates that the processing node is under high load. At this time, tasks with a second priority lower than the preset priority threshold in that processing node are identified as target tasks. These target tasks are migrated to processing nodes with a load factor less than the preset second coefficient threshold to balance the load of each processing node and improve overall processing efficiency.

[0086] Analysis of beneficial effects: Accurate task time prediction: By comprehensively predicting the processing time based on multiple factors, the estimation of task processing time becomes more accurate, which helps to rationally allocate resources and plan task processing processes, thereby improving the system's scheduling capabilities.

[0087] Dynamic priority adjustment: The second priority is dynamically determined based on the expected processing time and load factor, which can reflect the relative importance and urgency of tasks within the processing node in real time, making it easier to optimize the task processing order and improve processing efficiency.

[0088] Load balancing optimization: Through the task migration mechanism, the backlog of tasks caused by excessive load on processing nodes is effectively avoided, and load balancing between processing nodes is achieved, thereby improving the overall stability and processing capacity of the system.

[0089] Specific Implementation: In an online customer service feedback processing system, a feedback message regarding an order payment issue is received. The feedback text is quite long. The system queries historical task data and finds that the average processing time for similar order payment issues is 1.5 hours, the historical accuracy rate of the current processing node is 85%, and the current load factor is 0.6 (assuming the load factor ranges from 0 to 1, with higher values ​​indicating higher load). Considering the length of the text and the high load of the current processing node, the estimated processing time is adjusted to 2 hours. The second priority is calculated using a formula. Assuming the preset first coefficient threshold is 0.5 and the preset priority threshold is 0.4, since the current processing node's load factor of 0.6 is greater than 0.5, and the task's second priority of 0.3125 is less than 0.4, this task is identified as the target task. The system finds another processing node with a load factor of 0.3 (less than the preset second coefficient threshold of 0.4), and therefore migrates the order payment issue feedback task to that processing node for processing.

[0090] A longer estimated processing time T means the task may require more resources and time to complete, thus lowering its priority when resources are limited. A higher load factor indicates a busier processing node, reducing the available resources for the task and also lowering its priority. For example, consider two tasks: Task A has an estimated processing time of 3 hours and a processing node load factor of 0.5; Task B has an estimated processing time of 2 hours and a processing node load factor of 0.3. The second priority of Task A can be calculated.

[0091] Second priority of task B

[0092] It is clear that task B has a higher priority.

[0093] Optionally, migrating at least a portion of the tasks to be processed to other processing nodes based on the load factor and the second priority of the tasks to be processed includes: If the load coefficient of the processing node is greater than or equal to a preset first coefficient threshold, the task with a second priority lower than a preset priority threshold in the processing node is determined as the target task. The target task is migrated to a processing node whose load coefficient is less than a preset second coefficient threshold.

[0094] In this embodiment, the load on processing nodes changes dynamically during actual task processing. When the load is too high, it may affect the processing efficiency and quality of the tasks. To ensure the efficient operation of the system, a mechanism is needed to identify low-priority tasks that are suitable for migration from processing nodes and transfer them to processing nodes with lower loads, thereby achieving load balancing and reasonable task allocation.

[0095] Explanation of proper nouns: Load factor: A value used to measure the current load level of a processing node. It is obtained by comprehensively evaluating the resource usage of the processing node (such as CPU utilization, memory usage, task queue length, etc.). The higher the load factor, the heavier the load on the processing node.

[0096] The preset first coefficient threshold is a pre-defined standard value for the load coefficient. When the load coefficient of a processing node reaches or exceeds this threshold, it indicates that the processing node is overloaded and task migration needs to be considered.

[0097] Second priority: The task priority is calculated based on the expected processing time and load factor, reflecting the relative importance and urgency of the task within the processing node.

[0098] Preset priority threshold: A pre-defined second priority standard value used to determine whether a task is a low-priority task. When the second priority of a task is lower than this threshold, the task may be regarded as the target task and migrated.

[0099] The preset second coefficient threshold is another pre-set standard value for the load coefficient. When the load coefficient of a processing node is less than this threshold, it indicates that the processing node has a low load and is capable of taking on tasks migrated from other nodes.

[0100] Target tasks: In the current processing node, tasks whose load coefficient is greater than or equal to the preset first coefficient threshold and whose second priority is lower than the preset priority threshold are identified as tasks that need to be migrated to other processing nodes.

[0101] Explanation of the principle: When the system detects that the load factor of a processing node is greater than or equal to a preset first threshold, it indicates that the node is under high load, which may affect task processing efficiency. At this time, the system will filter the tasks within that node according to a second priority, identifying tasks with a second priority lower than a preset priority threshold as target tasks. These target tasks, due to their low priority and the current high load on their respective nodes, are suitable for migration to other processing nodes. The system will then search for processing nodes with load factors lower than the preset second threshold and migrate the target tasks to those nodes. This reduces the burden on high-load processing nodes while making reasonable use of the resources of low-load nodes, achieving load balancing among processing nodes.

[0102] Analysis of beneficial effects: Load balancing optimization: By identifying and migrating low-priority tasks in high-load processing nodes through explicit rules, it effectively avoids task backlog and decreased processing efficiency caused by excessive load on processing nodes, achieves load balancing among processing nodes within the system, and improves the overall stability and processing capacity of the system.

[0103] Rational resource utilization: Migrating tasks from high-load nodes to low-load nodes allows for more rational allocation and utilization of system resources, avoiding excessive concentration of resources on high-load nodes and idle waste on low-load nodes, thereby improving resource utilization.

[0104] Improved task processing efficiency: Ensuring that processing nodes operate within an appropriate load range helps improve the task processing efficiency of each processing node, thereby improving the overall task processing efficiency of the system and enabling faster response and processing of user feedback.

[0105] Specific Implementation: In a task processing system of a cloud computing platform, the load factor of processing node X reaches 0.7 (the preset first threshold is 0.6), and this node has multiple tasks. After calculating the second priority of each task, the system finds that the second priority of task M is 0.3 (the preset priority threshold is 0.4), which is lower than the threshold. Therefore, task M is identified as the target task. Simultaneously, the system finds that the load factor of processing node Y is 0.3 (the preset second threshold is 0.4), indicating a lower load. Therefore, the system migrates task M from processing node X to processing node Y. In this way, the load on processing node X is alleviated, and processing node Y makes full use of its idle resources, thereby improving the overall task processing efficiency of the cloud computing platform.

[0106] In one possible embodiment, an evaluation system based on NLP technology to understand feedback semantics and perform intelligent feedback tracking is provided, comprising a "three-layer, two-chain" architecture: 1. Semantic parsing layer: Based on NLP technology, it realizes feedback text understanding and classification. 2. Smart Contract Layer: Blockchain-Driven Task Allocation and Progress Storage 3. Scheduling and Prediction Layer: Machine Learning-Driven Task Scheduling and Processing Time Prediction 4. Data Chain: Feedback on the entire lifecycle of data is stored and proven on the blockchain. 5. Execution Chain: An automated execution process of task allocation - processing - acceptance. (1) The role and significance of the three-layer technology module 1. Semantic parsing layer Function: Utilizes NLP technology to automatically parse unstructured feedback text submitted by users, while transforming messy raw text into machine-readable structured data, providing standardized input for subsequent intelligent scheduling and blockchain notarization.

[0107] Significance: Enabling intelligent access points lays the foundation for precise scheduling.

[0108] 2. Smart Contract Layer Function: It solidifies business rules in the form of code in the smart contract of the blockchain, records the "facts" of key links such as task creation, allocation, circulation and completion in an immutable way on the blockchain, forming a trusted "data chain", and automatically executes when conditions are met (such as automatic upgrade when the task times out), driving the operation of the "execution chain".

[0109] Significance: To establish transparency and credibility, and to automate processes.

[0110] 3. Scheduling and Prediction Layer: The System's "Decision Center" Function: Based on the results of semantic parsing (problem type, complexity) and real-time data (such as the skills of each processor, current workload, and historical performance), the system uses machine learning algorithms (such as reinforcement learning and optimization algorithms) to calculate the optimal task allocation scheme. At the same time, based on historical data, it predicts the possible processing time and probability of solving the task, and sets reasonable expectations and early warning mechanisms.

[0111] Significance: To optimize resources and improve user experience and management level.

[0112] (2) Overview of the problem handling and feedback process: 1. Intelligent parsing: After a user submits text feedback, the system uses NLP technology to automatically identify key information such as the type, subject, and urgency of the feedback, and transforms unstructured text into structured data.

[0113] 2. Generate evidence; structured data triggers a smart contract on the blockchain, automatically creating a new task, and permanently storing the key event of "task creation" as the first immutable record.

[0114] 3. Scheduling and allocation: Based on task information (such as urgency and type) and real-time resource status (such as personnel load), the system uses a machine learning model to calculate the optimal personnel and estimated completion time, and returns the allocation instruction to the smart contract.

[0115] 4. Execution Records: Smart contracts automatically assign tasks to designated personnel. All key nodes in the processing (such as start, collaboration, and completion) are recorded on the blockchain in real time, ensuring full transparency and traceability of the process.

[0116] 5. Acceptance and closure: After the task is completed, the smart contract will automatically update the status to "completed" after verification by the user or system, and archive the final result on the blockchain to form a reliable and complete processing file for query and analysis.

[0117] (II) Detailed Design of Core Technology Modules 1. NLP Semantic Understanding Module (Feedback Parsing and Classification) 1.1 Technical Principles By employing a three-tiered process of "text preprocessing → feature extraction → semantic classification," we achieve accurate understanding of feedback semantics and resolve the parsing problems of ambiguous text and multi-intent feedback.

[0118] 1.2. Key Algorithms and Formulas (1) Text preprocessing Noise removal: Removes special characters and stop words. The formula is as follows: Let the original feedback text be The stop word list is as follows: The cleaned text is

[0119] Semantic normalization: Word2Vec is used for word vector mapping, and word vectors... (Vector dimension, default 300), text vector

[0120] (2) Feature extraction (hybrid feature fusion) Traditional features: Shallow text feature extraction: using the Term Frequency-Inverse Document Frequency (TF-IDF) algorithm; Deep semantic feature extraction: using a pre-trained deep language model (BERT), calculated as follows: Frequency (TF)

[0121] in, Indicator In the text The number of times it appears in Representing text Total word count.

[0122] Inverse document frequency (IDF)

[0123] Where N is the total number of documents, and D is the number of words containing the term. A collection of documents, Indicates inclusion Number of documents TF-IDF

[0124] Deep features Using a BERT pre-trained model to extract contextual features, if a BERT base model is used, its output text features are: The final feature vector is formed by concatenating TF-IDF features and BERT features, i.e. (where k is the dimension of the TF-IDF feature).

[0125] (3) Semantic classification (BiLSTM-CRF model) Model structure: Bidirectional LSTM extracts sequence-dependent features, and a CRF layer optimizes label sequence prediction. Loss function: CRF negative log-likelihood loss, the formula is:

[0126] Where M is the number of samples, y_i is the true label (e.g., "functional failure", "experience optimization", "data error"), Y is the set of all labels, Score(y|x;θ) is the score of the label sequence y under the input x, and θ is the model parameter. Classification accuracy: Optimized by 5-fold cross-validation, the F1 score on the target test set is ≥0.92; The core target values ​​include: Weighted average F1 score ≥ 0.92 (primary optimization objective) Accuracy ≥ 0.90 (ensuring classification accuracy) Recall rate ≥ 0.88 (ensuring coverage) Macro average F1 score ≥ 0.89 (class balance) Differentiation target value: Emergency Complaint Category: Recall Rate ≥ 0.95 The standard deviation of performance for each category is ≤ 0.08 (balance). 1.3. Implementation Steps (1) Dataset Construction: Collect historical feedback data (≥100,000 records), label it with 8 core issue tags (functionality, data, experience, security, etc.), and divide the training / validation / test sets in an 8:1:1 ratio; (2) Preprocessing: Word segmentation is implemented based on jieba (Chinese) and NLTK (English), and a custom stop word list (including industry terms) is created; (3) Model training: Pre-trained Word2Vec (based on industry corpus, window size 5, minimum word frequency 15) Load the BERT-base pre-trained model, freeze the first 6 layers, and fine-tune the last 4 layers. Training BiLSTM-CRF (hidden layer dimension 256, dropout=0.3, learning rate 1e-4) The above model parameters will be adjusted appropriately during the model training process; (4) Semantic parsing interface: The model is encapsulated as a RESTful interface, which takes feedback text as input and outputs "question type + semantic keywords + confidence level".

[0127] 2. Blockchain Smart Contract Module (Feedback Storage and Task Allocation) 2.1. Technology Selection The underlying blockchain is a consortium blockchain (Hyperledger Fabric 2.4), and its nodes include "feedback initiator, processor, and regulator". Smart contract language: Go (Chaincode), supports transaction atomicity and access control. 2.2. Core Contract Logic and Formulas (1) Feedback data is uploaded to the blockchain Data structure: Feedback{ID, Content, SemanticResult, Timestamp, Hash} Anti-tampering hash: A unique identifier is calculated using SHA-256, with the following formula: Hash = SHA256(ID || Content || SemanticResult || Timestamp || Sign) where Sign is the digital signature of the initiator (based on the ECDSA algorithm). (2) Automatic task assignment contract Allocation rules: Priority is calculated based on "processor's historical accuracy + current load", using the following formula:

[0128] in: Acc_j: Accuracy of historical tasks processed by method j (Acc_j = number of correct processing / total number of processing) _x0005_ Load_j: The number of tasks currently pending in processor j. MaxLoad: The maximum load per node set by the system. α: Weighting coefficient (default 0.6, can be adjusted via regulatory contract) Contract Triggering: After the NLP module outputs semantic results, the allocation contract is automatically invoked, and the processor with the highest Priority_j is selected to assign the task. 2.3. Implementation Steps (1) Deployment of consortium blockchain: Set up three organizations (initiator, processor, and supervisor), each with one node, using the Raft consensus mechanism (3-node cluster). It is recommended that all three parties jointly maintain the sorting service: Total number of nodes: 3 sorting nodes (1 node from each organization); Node distribution: Sort Node 1: Maintained by the initiating organization Sort Node 2: Processor Organization Maintenance Sorting Node 3: Maintenance by the Regulatory Authority Configure channel policy: Feedback data is only visible to authorized nodes, and task allocation results are agreed upon across the entire network. (2) Smart contract development: Develop FeedbackChaincode: Implement interfaces for adding feedback to the chain (addFeedback), querying (queryFeedback), and verifying hashes (verifyHash). AddFeedback: Users can submit feedback, which is automatically generated as a content hash (tamper-proof) and records the creator's identity. QueryFeedback: Publicly discloses query feedback details to ensure information transparency. VerifyHash: Verifies whether the feedback content has been tampered with by comparing hashes, ensuring data authenticity. Develop TaskChaincode: Implement the interfaces for task assignment (assignTask), progress update (updateProgress), and acceptance confirmation (confirmAccept). AssignTask: The assignor creates a task, specifies the executor, and sets the initial progress to 0. UpdateProgress: Only the executor can update the progress (0-100); accepted tasks cannot be modified. ConfirmAccept: Only the assignor can accept the confirmation, and this action can only be taken when progress is 100% to ensure task completion. (3) Contract testing: Stress testing is conducted through the Fabric Test Network to ensure that ≥50 feedback requests are processed per second and the transaction success rate is ≥99.9%.

[0129] Response time testing is performed using stratified response and percentiles: Average response time: ≤500ms, P95 response time: ≤800ms Latency testing was conducted using blockchain-specific latency metrics and network latency testing: End-to-end latency: ≤1000ms (4) Data integration: The NLP module calls the contract interface through FabricSDK to realize the automatic uploading of semantic results to the blockchain. 3. Task scheduling and processing time prediction module 3.1. Technical Principles Based on historical task data, a machine learning model is built to predict processing time, and task priorities are dynamically adjusted in conjunction with real-time load to avoid node overload.

[0130] 3.2. Key Algorithms and Formulas (1) Processing time prediction (XGBoost regression model) Feature vector: X = [Type, ContentLen, HistTime, Load, Acc] (Type: Problem type code, ContentLen: Length of feedback text, HistTime: Historical average processing time (seconds) for similar problems, Load: Current load on the assigned node, Acc: Historical accuracy of the node) Note: Definition of Load: Unit of measurement: Standardized index, range [0, 1] Calculation formula: Load = 0.4 × Task ratio + 0.3 × Time pressure + 0.3 × Resource usage Data source: Real-time query of blockchain smart contracts + offline calculation cache Update frequency: refreshed every 5 minutes Prediction formula: Let the model output be the predicted processing time value \hat{T}, and calculate it using the additive model of XGBoost: Prediction accuracy: Mean absolute error (MAE) of the target test set ≤ 10% (2) Dynamic scheduling algorithm (priority adjustment based on prediction time) Scheduling priority:

[0131] in, (Load factor, range [0,1]), the larger S is, the higher the task priority. Scheduling trigger: When the node load factor is ≥0.8, low-priority tasks are automatically scheduled. Migrate to nodes with a load factor < 0.5 3.3. Implementation Steps (1) Construction of historical dataset: Collect task data from the past 6 months, including 12 features such as "problem type, processing time, node load, and accuracy", with a total of ≥50,000 samples. (2) Feature engineering: Category features (such as question type) are encoded using one-hot encoding. Continuous features (such as processing time and text length) are normalized using Z-score: x'=(x-μ) / σ (μ is the mean, σ is the standard deviation) (3) Model training and optimization: An XGBoost regression model was trained (100 trees, maximum depth 8, learning rate 0.1), using 5-fold cross-validation for parameter tuning. The dataset split ratio was 7:1.5:1.5. Compare LSTM and Random Forest models; select the model with the lowest MAE (predictive performance) for deployment. 1. LSTM Model LSTM models are suitable for sequence features (such as time-related node load changes), capturing long-term dependencies through gating mechanisms, and the output layer is a regression prediction. (ht represents the hidden state, ct represents the cell state, and W and b are parameters to be learned) 2. Random Forest Regression Model Random forest regression models are based on the ensemble of multiple decision trees, using bagging to reduce variance, and averaging the predictions of individual trees. Where M is the number of trees, Θm is the parameter of the m-th tree, and T is the decision tree prediction function.

[0132] 3. Decision rules when MAE is similar: Real-time systems: Prioritize systems with the fastest prediction speed (XGBoost). Batch processing system: Prioritize the most stable one (Random Forest). High interpretability requirements: Select features with clear importance (XGBoost / Random Forest). Limited operational resources: Choose the simplest option (Random Forest). (4) Development of the scheduling system: Real-time collection of node load data (updated every 10 seconds) Develop a task migration interface based on the predicted time and load factor to calculate scheduling priority. Deploy a monitoring panel to display task progress and predicted time deviation rate in real time.

[0133] Note: Time skew rate: Provides reliability weights (Acc feature) for the scheduling algorithm, detects the accuracy of the prediction model in real time, triggers retraining, dynamically adjusts the task allocation logic based on the skew feature, adds a safety margin to the prediction time, and improves the SLA achievement rate.

[0134] Deviation rate calculation rules: # Single task deviation rate bias_rate = abs(actual - predicted) / actual # Node historical accuracy (Acc) acc = 1 - average(bias_rates_last_100_tasks) # Scheduling Priority Calculation priority = predicted_time × (1 + load) × (1 + (1 - acc)×0.5) Figure 2 A flowchart illustrating an evaluation method based on NLP technology for understanding feedback semantics and performing intelligent feedback tracking, as provided in this application embodiment, is shown below. Figure 2 As shown, the entire process includes the following steps: 1. System initialization phase (1-2 weeks) Deploy a Hyperledger Fabric consortium blockchain and configure node permissions and consensus mechanisms. Train NLP semantic classification and time prediction models, and encapsulate API interfaces. 2. Feedback Processing Phase (Real-time) ① The feedback initiator submits the feedback text to the semantic parsing interface. ② The NLP module outputs "problem type + semantic keywords + confidence level" and calls the blockchain contract to complete the feedback on-chain. ③ The smart contract automatically calculates the priority of the processor, allocates tasks, and generates task IDs. ④ The scheduling system predicts the processing time and pushes the task to the processor's terminal. ⑤ The processor updates the task progress (e.g., "processing" or "pending acceptance"), and the progress data is automatically uploaded to the blockchain. ⑥ After the feedback initiator passes the acceptance, the contract records the "completed" status, and the entire process data is archived.

[0135] 3. Optimization and iteration phase (monthly) Incrementally train the NLP model based on new feedback data, and update the stop word list and classification labels. Analyze the reasons for time prediction deviations and optimize the feature vector (e.g., add a "problem urgency" feature). Adjusting the task allocation weight α in the smart contract improves overall processing efficiency. To achieve the above embodiments, this application also proposes an evaluation device based on NLP technology to understand feedback semantics and perform intelligent feedback tracking.

[0136] Figure 3This is a schematic diagram of the structure of an evaluation device that uses NLP technology to understand feedback semantics and perform intelligent feedback tracking, as provided in an embodiment of this application.

[0137] like Figure 3 As shown, the device may include: The classification module 310 is used to receive feedback text, extract features from the feedback text, and obtain semantic features and classification results; The task determination module 320 is used to store the feedback text, semantic features and classification results into the blockchain, and determine the task to be processed based on the feedback text, semantic features and classification results; The node determination module 330 is used to adjust the processing node corresponding to each task to be processed according to the estimated processing time of each task to be processed. The processing module 340 is used to assign the task to be processed to the corresponding processing node so that the processing node can process the task.

[0138] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of this embodiment, and will not be repeated here.

[0139] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing method embodiments.

[0140] To implement the above embodiments, this application also proposes a computer program product having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the foregoing method embodiments.

[0141] To implement the above embodiments, this application also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method described in the foregoing method embodiments.

[0142] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example, the electronic device 800 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0143] Reference Figure 4 The electronic device 800 may include one or more of the following components: processing component 802, memory 804, power component 806, multimedia component 808, audio component 810, input / output (I / O) interface 812, sensor component 814, and communication component 816.

[0144] Processing component 802 typically controls the overall operation of electronic device 800, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

[0145] Memory 804 is configured to store various types of data to support the operation of electronic device 800. Examples of such data include instructions for any application or method operating on electronic device 800, contact data, phonebook data, messages, pictures, videos, etc. Memory 804 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.

[0146] Power component 806 provides power to various components of electronic device 800. Power component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800.

[0147] Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When the electronic device 800 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0148] Audio component 810 is configured to output and / or input audio signals. For example, audio component 810 includes a microphone (MIC) configured to receive external audio signals when electronic device 800 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 804 or transmitted via communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

[0149] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0150] Sensor assembly 814 includes one or more sensors for providing state assessments of various aspects of electronic device 800. For example, sensor assembly 814 can detect the on / off state of electronic device 800, the relative positioning of components such as the display and keypad of electronic device 800, changes in position of electronic device 800 or a component of electronic device 800, the presence or absence of user contact with electronic device 800, orientation or acceleration / deceleration of electronic device 800, and temperature changes of electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 814 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0151] Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices. Electronic device 800 can access wireless networks based on communication standards, such as WiFi, 4G, or 5G, or combinations thereof. In one exemplary embodiment, communication component 816 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 816 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0152] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0153] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, which can be executed by a processor 820 of an electronic device 800 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0154] To implement the above embodiments, this application also proposes a chip, including: the chip includes a processing circuit configured to perform the methods provided in the foregoing embodiments.

[0155] Figure 5 This is a schematic diagram of the structure of a chip according to an embodiment of this application. See also... Figure 5 The diagram shown is a schematic representation of the structure of chip 1100, but it is not limited to this.

[0156] Chip 1100 includes processing circuitry 1101, which is configured to perform any of the above methods.

[0157] In some embodiments, chip 1100 further includes one or more interface circuits 1102. Optionally, the interface circuit 1102 is connected to memory 1103, and the interface circuit 1102 can be used to receive signals from memory 1103 or other devices, and the interface circuit 1102 can be used to send signals to memory 1103 or other devices. For example, the interface circuit 1102 can read instructions stored in memory 1103 and send the instructions to processing circuit 1101.

[0158] In some embodiments, the interface circuit 1102 performs at least one of the communication steps such as sending and / or receiving in the above method, while the processing circuit 1101 performs other steps.

[0159] In some embodiments, the terms interface circuit, interface, transceiver pin, transceiver, etc., can be used interchangeably.

[0160] In some embodiments, chip 1100 further includes one or more memories 1103 for storing instructions. Optionally, all or part of the memories 1103 may be located outside of chip 1100.

[0161] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0162] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0163] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0164] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0165] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0166] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0167] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0168] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. An evaluation method based on NLP technology to understand feedback semantics and perform intelligent feedback tracking, characterized in that, include: Receive feedback text, extract features from the feedback text, and obtain semantic features and classification results; The feedback text, semantic features, and classification results are stored in the blockchain, and the task to be processed is determined based on the feedback text, semantic features, and classification results. Adjust the processing node corresponding to each pending task according to the estimated processing time of each pending task; The task to be processed is assigned to the corresponding processing node so that the processing node can process the task to be processed.

2. The method according to claim 1, characterized in that, The step of extracting features from the feedback text to obtain semantic features and classification results includes: The feedback text is filtered to remove text of a preset type. The filtered feedback text is normalized to obtain a text vector; Feature extraction is performed on the text vector to obtain a fused feature vector; The classification is performed based on the fused feature vector to obtain the classification result, wherein the classification result represents the type of problem reflected in the feedback text.

3. The method according to claim 2, characterized in that, The step of extracting features from the text vector to obtain a fused feature vector includes: Extract the term frequency features and inverse document frequency features from the text vector; The fused feature vector is obtained by fusing the term frequency features and the inverse document frequency features.

4. The method according to claim 3, characterized in that, The method further includes: The first priority of each processing node is determined based on its historical accuracy and current load. The tasks to be processed corresponding to each processing node are determined according to the first priority.

5. The method according to claim 4, characterized in that, The step of adjusting the processing node corresponding to each pending task based on its estimated processing time includes: Based on the problem type in historical task data, feedback text length, historical average processing time for similar problems, current load of processing nodes, and historical accuracy of processing nodes, the estimated processing time is predicted. Based on the estimated processing time and the load factor of the processing node, determine the second priority of the tasks to be processed in the processing node; Based on the load factor and the second priority of the tasks to be processed, at least a portion of the tasks to be processed are migrated to other processing nodes.

6. The method according to claim 5, characterized in that, The formula for calculating the second priority is: Where S is the second priority, The estimated processing time is denoted as , and LoadFactor is the load factor.

7. The method according to claim 6, characterized in that, The step of migrating at least a portion of the tasks to be processed to other processing nodes based on the load factor and the second priority of the tasks to be processed includes: If the load coefficient of the processing node is greater than or equal to a preset first coefficient threshold, the task with a second priority lower than a preset priority threshold in the processing node is determined as the target task. The target task is migrated to a processing node whose load coefficient is less than a preset second coefficient threshold.

8. An evaluation device based on NLP technology to understand feedback semantics and perform intelligent feedback tracking, characterized in that, include: The classification module is used to receive feedback text, extract features from the feedback text, and obtain semantic features and classification results. The task determination module is used to store the feedback text, semantic features and classification results into the blockchain, and determine the task to be processed based on the feedback text, semantic features and classification results; The node determination module is used to adjust the processing node corresponding to each task to be processed according to the estimated processing time of each task to be processed. The processing module is used to assign the task to be processed to the corresponding processing node so that the processing node can process the task.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of the preceding claims 1-7.

Images