Improvement method and device for user feedback and computer readable storage medium

Abstract

The invention provides an improvement method and device for user feedback and a computer readable storage medium. In the scheme, an initial knowledge graph is constructed, and nodes of the initial knowledge graph comprise function points, key problems and improvement suggestions; obtaining user feedback information, analyzing the user feedback information by using a large language model, and updating the initial knowledge graph according to an analysis result to obtain an updated knowledge graph; according to relevant information of the updated knowledge graph, characteristic parameters of a large language model are adjusted, an adjusted large language model is obtained, the adjusted large language model is applied to analysis to obtain improvement suggestions corresponding to user feedback information, and the relevant information comprises a first data pair and a second data pair; the first data pair comprises user feedback information and a function point corresponding to the user feedback information, and the second data pair comprises a key problem associated with the function point and an improvement suggestion corresponding to the key problem. According to the scheme, the problem of relatively low efficiency of user feedback analysis in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology. Specifically, it relates to an improvement method, device, computer-readable storage medium, and electronic device for user feedback. Background Art

[0002] In the current field of user experience, manually analyzing user feedback to identify sentiment tendencies and key issues, and then proposing system optimization suggestions based on this, is an important link in enhancing the customer experience. However, this process has significant limitations, including a long data processing chain, time-consuming and costly analysis, and the easy introduction of subjective biases when relying on personal experience for analysis, thus increasing the risk of misjudgment. The accuracy of manual analysis is not high, and the efficiency is low. Therefore, a more efficient and objective analysis method is needed.

[0003] With the in-depth application of large language models in natural language understanding and generation, it provides in-depth technical assistance for user experience analysis. Large language models can quickly process a large amount of text data, significantly shortening the generation time from user feedback to system optimization suggestions, while reducing labor costs. However, the analysis of large language models is mainly based on their training datasets, which results in lower accuracy in specific domain analysis, difficulty in accurately focusing on specific concerns, and difficulty in fully capturing and understanding the subtle and domain-specific context in user feedback, thus affecting the analysis efficiency. Summary of the Invention

[0004] The main objective of this application is to provide an improvement method, device, computer-readable storage medium, and electronic device for user feedback, so as to at least solve the problem of low efficiency in user feedback analysis in the prior art.

[0005] To achieve the above object, according to one aspect of the present application, an improvement method for user feedback is provided, including: constructing an initial knowledge graph, where the nodes of the initial knowledge graph include function points, key issues, and improvement suggestions, and the function points are functional modules that users have when using the product; obtaining user feedback information, analyzing the user feedback information using a large language model to obtain an analysis result, and updating the initial knowledge graph according to the analysis result to obtain an updated knowledge graph, where the update operation includes at least one of the following: adjusting node weights, adjusting the relationship weights between nodes, adding nodes, and the relationship weights between nodes represent the degree of association between nodes; adjusting the feature parameters of the large language model according to the relevant information of the updated knowledge graph to obtain an adjusted large language model, and applying the adjusted large language model to analyze and obtain improvement suggestions corresponding to the user feedback information, where the relevant information includes a first data pair and a second data pair, the first data pair includes the user feedback information and the corresponding function point, and the second data pair includes the key issues associated with the function point and the improvement suggestions corresponding to the key issues.

[0006] According to another aspect of the present application, an improvement device for user feedback is provided, including: a construction unit for constructing an initial knowledge graph, where the nodes of the initial knowledge graph include function points, key issues, and improvement suggestions, and the function points are functional modules that users have when using the product; an update unit for obtaining user feedback information, analyzing the user feedback information using a large language model to obtain an analysis result, and updating the initial knowledge graph according to the analysis result to obtain an updated knowledge graph, where the update operation includes at least one of the following: adjusting node weights, adjusting the relationship weights between nodes, adding nodes, and the relationship weights between nodes represent the degree of association between nodes; an adjustment unit for adjusting the feature parameters of the large language model according to the relevant information of the updated knowledge graph to obtain an adjusted large language model, and applying the adjusted large language model to analyze and obtain improvement suggestions corresponding to the user feedback information, where the relevant information includes a first data pair and a second data pair, the first data pair includes the user feedback information and the corresponding function point, and the second data pair includes the key issues associated with the function point and the improvement suggestions corresponding to the key issues.

[0007] According to still another aspect of the present application, a computer-readable storage medium is provided, where the computer-readable storage medium includes a stored program, and when the program runs, it controls the device where the computer-readable storage medium is located to execute any one of the improvement methods for user feedback.

[0008] According to another aspect of the present application, an electronic device is provided, including: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include those for executing any one of the improvement methods for user feedback.

[0009] Applying the technical solution of the present application, an initial knowledge graph is constructed. The nodes of the initial knowledge graph include function points, key issues, and improvement suggestions. The function points are the functional modules that users have when using the product; user feedback information is obtained, and the large language model is used to analyze the user feedback information to obtain an analysis result, and the initial knowledge graph is updated according to the analysis result to obtain an updated knowledge graph, where the update operation includes at least one of the following: adjusting the node weights, adjusting the relationship weights between nodes, and adding new nodes. The relationship weights between nodes represent the degree of association between nodes; the characteristic parameters of the large language model are adjusted according to the relevant information of the updated knowledge graph to obtain an adjusted large language model, and the adjusted large language model is applied to analyze and obtain improvement suggestions corresponding to the user feedback information. Among them, the relevant information includes the first data pair and the second data pair. The first data pair includes the user feedback information and the corresponding function point, and the second data pair includes the key issues associated with the function point and the corresponding improvement suggestions. In this solution, by injecting the answers output by the large language model into the initial knowledge graph, continuously expanding and improving the knowledge graph relationship network, and then adjusting the large language model through the more perfect knowledge graph, the answers output by the large language model are made more accurate and comprehensive, thereby improving the efficiency of user feedback analysis and solving the problem of low efficiency of user feedback analysis in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The specification drawings forming a part of the present application are used to provide a further understanding of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0011] Figure 1 A hardware structure block diagram of a mobile terminal for executing an improvement method for user feedback provided in an embodiment of the present application is shown;

[0012] Figure 2 A flowchart of an improvement method for user feedback provided in an embodiment of the present application is shown;

[0013] Figure 3 An initial knowledge graph schematic diagram of an improvement method for user feedback provided in an embodiment of the present application is shown;

[0014] Figure 4The flowchart of an improved method for user feedback provided by an embodiment of the present application is shown;

[0015] Figure 5 The schematic diagram of a knowledge graph constructed based on historical data for an improved method for user feedback provided by an embodiment of the present application is shown;

[0016] Figure 6 The structural block diagram of an improved device for user feedback provided by an embodiment of the present application is shown.

[0017] Among them, the above-mentioned drawings include the following reference numerals:

[0018] 102, processor; 104, memory; 106, transmission device; 108, input / output device. Detailed implementation manners

[0019] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so as to describe the embodiments of the present application here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or device.

[0022] As introduced in the background art, the prior art's user analysis method based on large language models is difficult to fully capture and understand the subtle and domain-specific context in user feedback, thus affecting the analysis efficiency. To solve the problem of low efficiency in user feedback analysis in the prior art, the embodiments of the present application provide an improved method, device, computer-readable storage medium and electronic device for user feedback.

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0024] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 is a hardware structure block diagram of a mobile terminal for an improved method for user feedback according to an embodiment of the present invention. As Figure 1 shown, the mobile terminal may include one or more ( Figure 1 only one is shown in Figure 1 a processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Among them, the above mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above mobile terminal. For example, the mobile terminal may further include more or fewer components than Figure 1 shown in

[0025] The memory 104 can be used to store computer programs. For example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implements the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories may be connected to the mobile terminal through a network. Examples of the above networks include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one instance, the transmission device 106 includes a network adapter (abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (abbreviated as RF) module, which is used to communicate with the Internet wirelessly.

[0026] In this embodiment, an improved method for user feedback running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0027] Figure 2 It is a schematic flowchart of an improved method for user feedback according to an embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0028] Step S201, construct an initial knowledge graph. The nodes of the above initial knowledge graph include function points, key problems, and improvement suggestions. The above function points are functional modules that users have when using the product;

[0029] Specifically, the Knowledge Graph is an important branch technology of artificial intelligence. It is a structured semantic knowledge base used to describe concepts in the physical world and their interrelationships in symbolic form. Its basic unit of composition is the "entity-relationship-entity" triple, as well as entity and its related attribute-value pairs. Entities are connected to each other through relationships to form a networked knowledge structure. The user feedback knowledge graph takes the function points, key problems, and analyzed improvement suggestions refined from user feedback as nodes and connects them to each other as a network, which can achieve accurate traceability of relationships.

[0030] In this embodiment, the main purpose of constructing the initial knowledge graph is to create a structured framework for storing and associating all functional modules (i.e., function points) in the software system, the key issues mentioned in user feedback, and the improvement suggestions proposed for these issues. Each node in this initial knowledge graph represents an independent function point in the software system, a key issue reflected by users, or a suggestion for solving these issues. Among them, user feedback refers to the feedback of users on products and services, including comments, expectations, preferences, etc., which is obtained through sources such as software evaluations in the app store, system user feedback, questionnaires, and internal employee feedback. Function points refer to the functional modules that users use in the product, that is, each functional module that provides services externally in the software system, such as payment functions, message notifications, account management, etc. All user feedback is centered around function points. Key issues are the most critical and core information or problems clustered and refined from a large amount of user feedback, such as "payment delay", "unfriendly interface", etc. These key issue nodes, by establishing relationships with function point nodes, clarify which function point has caused user dissatisfaction or attention. Improvement suggestions are a series of specific improvement measures and suggestions aimed at improving aspects such as the user interface, interaction logic, and performance based on user feedback, data analysis, and industry best practices, providing clear optimization directions and references for system developers or designers, such as "optimize the payment interface", "increase user guidance information", etc. Improvement suggestion nodes, by establishing connections with key issue nodes, illustrate how to solve problems or meet user needs.

[0031] By constructing the above initial knowledge graph, it is possible to quickly locate the key issues under a certain function point and the existing improvement suggestions for these issues, so as to provide more accurate and rich prompts to the large language model in subsequent analysis, assisting it in making more reasonable analysis decisions. Generally speaking, constructing the initial knowledge graph is to establish a basic platform for efficient and accurate feedback analysis and improvement opinion generation, which can automatically and intelligently process user feedback, greatly accelerating the feedback conversion process and improving the conversion quality and efficiency.

[0032] Step S202: Obtain user feedback information, analyze the above user feedback information using a large language model to obtain an analysis result, and update the above initial knowledge graph according to the above analysis result to obtain an updated knowledge graph, where the update operation includes at least one of the following: adjusting the node weight, adjusting the relationship weight between nodes, adding nodes, and the relationship weight between the above nodes represents the degree of association between nodes;

[0033] Specifically, the obtained user feedback information is input into a large language model for in-depth analysis to identify the deep meaning, sentiment tendency, functional requirements, and potential key issues in the user feedback information. Leveraging its powerful natural language understanding and generation capabilities, the large language model can quickly process complex text information, extract valuable data from it, and form an analysis result.

[0034] Next, the above initial knowledge graph is dynamically updated based on these analysis results. The update operations mainly include three types: adjusting the node weights, adjusting the relationship weights between nodes, and adding nodes. Specifically, the node weight reflects the importance of function points, key issues, or improvement suggestions. The relationship weight between nodes measures the association strength between function points and key issues, and between key issues and improvement suggestions. In addition, when the large language model identifies new key issues or improvement suggestions based on the function points extracted from the user feedback information, corresponding nodes will be added to the initial knowledge graph to expand the information coverage of the initial knowledge graph, resulting in an updated knowledge graph. This mechanism of dynamically adding nodes ensures that the knowledge graph can promptly reflect the new changes in the software system and the new needs of users.

[0035] Through the above update operations, the large language model feeds the analysis result back to the initially constructed initial knowledge graph, and the initial knowledge graph is continuously updated, making the updated knowledge graph more powerful and accurate. For example, when a key issue is first output by the large language model, the key issue is added to the knowledge graph and connected to the relevant function points. If subsequent user feedback continues to mention this key issue, then the node weight of this key issue and the relationship weight with the corresponding function points will increase, reflecting the importance of this key issue. At the same time, once an effective improvement suggestion is found, the relationship weight between this key issue and the corresponding improvement suggestion will also increase, making it easier to be associated and recommended by the large language model in future analysis.

[0036] In summary, by analyzing user feedback information with the large language model and real-time updating the initial knowledge graph, the self-improvement and optimization of the knowledge graph are achieved, ensuring that it can always reflect the latest and most relevant information.

[0037] In step S203, the feature parameters of the above large language model are adjusted according to the relevant information of the above updated knowledge graph, obtaining an adjusted large language model. The adjusted large language model is applied to analyze and obtain improvement suggestions corresponding to the above user feedback information. Among them, the above relevant information includes the first data pair and the second data pair. The first data pair includes the above user feedback information and its corresponding function points. The second data pair includes the key issues associated with the above function points and the improvement suggestions corresponding to the above key issues.

[0038] Specifically, the updated knowledge graph is utilized to further optimize the large language model, aiming to enhance the accuracy and comprehensiveness of the large language model in analyzing user feedback, thereby improving the analysis efficiency. This process is achieved by adjusting the feature parameters of the large language model. Feature parameters are important parameters within the large language model that affect the prediction results. Adjusting the feature parameters enables the large language model to better understand and process specific types of data. Specifically, the feature parameters of the large language model are also known as model parameters or hyperparameters. These parameters are automatically adjusted by algorithms during the training process, defining the structure and function of the model and directly influencing the model's understanding of input data and the accuracy of the predicted output. Feature parameters mainly include weights, biases, the number of hidden layer units, the number of layers, the learning rate, activation functions, attention mechanism parameters, etc. Among them, weights are one of the most fundamental parameters in the model, representing the connection strength between nodes in the neural network. In the large language model, weights can determine the impact of a word or phrase on the overall meaning of a sentence. Through fine-tuning, the model can learn more specific weight assignments. For example, when processing user feedback, which words or phrases are more relevant to function points or key issues. Biases work together with weights to adjust the activation level of neural network nodes. In the large language model, biases can help the model lean towards a more appropriate output when facing a specific input sequence, enhancing the model's expressive ability. The number of hidden layer units determines the capacity or complexity of the model. More hidden layer units mean the model can capture more complex data patterns, but it may also increase the risk of overfitting. The number of layers in a deep learning model determines the level of abstraction it can learn. More layers allow the model to learn deeper semantic relationships, but for a specific task, too many layers may lead to training difficulties and waste of computing resources. The learning rate is the speed at which the model adjusts its parameters during training. An appropriate learning rate can accelerate training while reducing the likelihood of getting stuck in local optima. During the fine-tuning stage, adjusting the learning rate can ensure that the model can quickly adapt to new data and converge stably. Activation functions, such as ReLU, tanh, or sigmoid, determine how neurons convert signals into outputs. Different activation functions affect the model's non-linear expressive ability and thus its ability to understand and generate language. In modern language models such as Transformer, the attention mechanism allows the model to focus on certain parts of the sequence when processing sequential data. Adjusting the parameters of the attention mechanism can optimize the model's judgment of the importance of information when analyzing user feedback.

[0039] The process of adjusting feature parameters is usually iterative. By continuously experimenting and evaluating the performance of the model on specific tasks (such as the accuracy of user feedback analysis), a set of parameter combinations that best suit the current requirements can be found. In this embodiment, the updated knowledge graph provides additional supervision signals to help guide the adjustment of feature parameters, enabling the large language model to more effectively understand and convert user feedback into improvement suggestions.

[0040] Furthermore, the relevant information after updating the knowledge graph, namely the first data pair and the second data pair, is used to guide the adjustment of the feature parameters of the large language model. Among them, the first data pair includes user feedback information and its corresponding function point. For example, if the user feedback is about the delay problem of the payment function, then "payment delay" and "payment function" form a pair of first data pairs, which helps the large language model learn how different user feedback texts should be mapped to specific function modules of the software system. The second data pair includes key issues associated with the function point (such as the delay problem in the payment function) and improvement suggestions for these key issues (such as optimizing the payment interface), and the second data pair guides the large language model on how to generate reasonable and effective improvement suggestions starting from the identified key issues.

[0041] The first data pair and the second data pair are used to adjust the large language model, that is, to fine-tune the large language model. Fine-tuning is a machine learning technique that, based on a pre-trained model, uses domain-specific data for additional training to make the model better adapt to the requirements of that domain. In this process, the large language model adjusts its internal feature parameters according to the association between user feedback and function points, the nature of key issues, and the effectiveness of improvement suggestions in the first data pair and the second data pair, such as modifying the weights of word embeddings, changing the configuration of the attention mechanism, etc., to more accurately capture the intention of user feedback and generate appropriate improvement suggestions based on this.

[0042] The adjusted large language model can, when receiving new user feedback information, more quickly identify its associated function points, more accurately extract key issues, and generate improvement suggestions that are highly relevant and targeted to the key issues based on the rich improvement suggestions in the knowledge graph. This process not only improves the efficiency of feedback analysis but also ensures the quality of improvement suggestions, helping the software development team quickly locate and solve problems and continuously optimize the user experience.

[0043] In short, by extracting key semantic and relational information from the updated knowledge graph and making targeted adjustments to the large language model, the model can better understand problems and more accurately generate solutions when processing user feedback.

[0044] Through this embodiment, by constructing and maintaining an initial knowledge graph that includes functional points, key issues, and improvement suggestions, and combining the analysis capabilities of large language models, intelligent analysis of user feedback information is achieved. It can automatically identify the functional modules mentioned in user feedback, extract key issues, and update the initial knowledge graph based on the analysis results of large language models. Among them, adjusting the node weights reflects the relative importance of functional points, issues, or suggestions, adjusting the relationship weights between nodes strengthens the association between various parts, and adding new nodes expands the coverage of the knowledge graph. This dynamic optimization ensures that the knowledge graph can continuously reflect the real needs of users and the latest status of the product, thereby improving the efficiency and accuracy of feedback analysis. The update of the knowledge graph not only enriches the training data of large language models but also enables large language models to better understand the context and intention of user feedback by adjusting the feature parameters of large language models. At the same time, the analysis results of large language models are used for the expansion and optimization of the knowledge graph, forming a closed feedback loop. This two-way empowerment of the model and the graph improves the flexibility and intelligence of user analysis, making the analysis results more accurate and the improvement suggestions more effective. In addition, since the update of the knowledge graph and the fine-tuning of large language models are carried out after receiving new user feedback, this method supports continuous learning and iteration. Over time, it will be able to better understand and respond to user feedback, continuously optimize its analysis capabilities and the quality of improvement suggestions, forming a positive cycle and continuously enhancing the product and service.

[0045] Through the above series of steps, first, construct an initial knowledge graph. The nodes of the initial knowledge graph include functional points, key issues, and improvement suggestions. The functional points are the functional modules that users have when using the product. Then, obtain user feedback information, use a large language model to analyze the user feedback information to obtain analysis results, and update the initial knowledge graph according to the analysis results to obtain an updated knowledge graph. The update operations include at least one of the following: adjusting node weights, adjusting relationship weights between nodes, and adding new nodes. The relationship weights between nodes represent the degree of association between nodes. Finally, adjust the feature parameters of the large language model according to the relevant information of the updated knowledge graph to obtain an adjusted large language model, and apply the adjusted large language model to analyze and obtain improvement suggestions corresponding to the user feedback information. Among them, the relevant information includes the first data pair and the second data pair. The first data pair includes user feedback information and its corresponding functional point, and the second data pair includes the key issues associated with the functional point and the improvement suggestions corresponding to the key issues. That is, by injecting the answers output by the large language model into the initial knowledge graph, continuously expanding and improving the knowledge graph relationship network, and then adjusting the large language model through a more perfect knowledge graph, the answers output by the large language model are made more accurate and comprehensive, thereby improving the efficiency of user feedback analysis.

[0046] Regarding the construction of the above initial knowledge graph, first, historical user feedback information and relevant improvement suggestions are collected. Through natural language processing techniques, especially using large language models, information related to product functions is extracted from these historical user feedback information, which is the function points. At the same time, the large language model identifies the main problems in the historical user feedback information, that is, the key problems. For example, from the user feedback "Every time I use the payment function, I have to wait a long time to complete the transaction", the "payment function" can be extracted as the function point, and "delay during payment" as the key problem, and relevant improvement suggestions for the key problem are obtained. After obtaining the function points, key problems, and improvement suggestions, in order to quantify and measure the importance of each node and its mutual relationship, initial node weights and initial relationship weights between nodes are set for these nodes according to preset rules. The above preset rules refer to setting the initial weight q for each node according to the function priority and the number of repetitions of the key problem m , node connections are established according to the corresponding relationship between the function points, key problems, and improvement suggestions to form a relationship network between the function points and key problems, and between the key problems and improvement suggestions. At the same time, the initial relationship weight Q between the function point node k and the key problem node m is set according to the number of times the improvement suggestion is adopted km , as well as the initial relationship weight Q between the key problem node m and the improvement suggestion node r mr。The node weight reflects the relative importance of a node in the knowledge graph. Initial weights are set for each key issue node according to functional priorities and the number of times key issues are repeated. For example, a high weight is assigned to the "transaction function" because, in terms of functional priorities, the "transaction function" directly involves the core services of the product, while the weight of "color selection on the settings page" is relatively low because it has a relatively small impact on user satisfaction. Another example is that in historical user feedback data, certain problems or defects are repeatedly mentioned by multiple users. The statistics of the number of times key issues are repeated reflect the universality and urgency of this issue. Therefore, the more times a problem is reported, the higher its node weight in the knowledge graph, which ensures that key issues that have a wide impact and are strongly reflected by users can be prioritized for processing and resolution. At the same time, the initial relationship weight between the function point node and the key issue node is set according to the number of times improvement suggestions are adopted. The number of times an improvement suggestion is adopted reflects the effectiveness and implementation value of this improvement suggestion. If an improvement suggestion is adopted and successfully implemented multiple times, it indicates that the improvement suggestion has significant effects in solving key problems and enhancing the user experience. Therefore, the more times an improvement suggestion is adopted, the higher its node weight, which helps to quickly retrieve and recommend effective solutions when facing similar problems. For example, if there are multiple associations between the "transaction function" node and the "payment delay" key issue node in historical data, and the improvement measure "optimize the payment interface" is adopted and proven effective multiple times, then the relationship path from "transaction function" to "payment delay" to "optimize the payment interface" will be assigned a higher weight. By setting the initial relationship weight between the function point node and the key issue node based on functional priorities, the number of times key issues are repeated, and the number of times improvement suggestions are adopted, modules with high functional priorities and key issues with a large number of repetitions are given higher initial weights due to their importance in user experience and business operations. This ensures that in the analysis and optimization process, these key information can be prioritized, high-frequency problems can be quickly identified and responded to, ineffective analysis can be reduced, and the efficiency and accuracy of problem-solving can be improved. The number of times an improvement suggestion is adopted reflects its effectiveness and popularity in past feedback analysis and optimization practices. Suggestions with a large number of adoptions have a high initial weight in the knowledge graph, which helps to tend to propose solutions that have been proven effective or widely praised when generating improvement suggestions, thereby enhancing the pertinence and practical value of the suggestions. This setting ensures that the knowledge graph not only reflects the importance of each node but also the intensity of the interaction between entities, enabling the graph to more accurately guide model analysis and the generation of optimization suggestions.

[0047] By setting the initial node weights and the initial relationship weights between nodes through the above preset rules, the knowledge graph can become an intelligent and dynamically adjustable decision-making support tool, helping the large language model to quickly focus on important functions, urgent problems, and effective solutions when analyzing new user feedback, thereby improving the efficiency and accuracy of the entire feedback analysis process.

[0048] Through the above steps, an initial knowledge graph containing function points k, key problems m, and improvement suggestions r is constructed, as Figure 3 shown. This initial knowledge graph not only structurally represents the product functions, common problems in user feedback, and suggestions for solving these problems, but also reflects the importance and relevance of each part in actual applications through the weight mechanism.

[0049] All in all, the process of constructing the initial knowledge graph lays the foundation for the intelligent analysis of subsequent user feedback and the generation of improvement suggestions. It ensures that when the large language model analyzes new feedback, it can be based on the existing structured knowledge and historical experience, so as to more accurately understand the feedback content, extract key problems, and propose targeted improvement measures.

[0050] In this embodiment, in order to improve the ability of the large language model to process user feedback in a specific field, a mechanism for cross-domain knowledge fusion and transfer learning can also be introduced. For example, in addition to the knowledge graph based on user feedback, one or more cross-domain knowledge graphs can be constructed, including professional knowledge and cases in related fields (such as computer vision, natural language processing, etc.), for supplementing the domain knowledge of the large language model. Through transfer learning technology, the professional knowledge and cases in the cross-domain knowledge graph are transformed into a form that the large language model can understand, and the parameters of the large language model are fine-tuned, so that it can be more accurate and professional when processing user feedback in a specific field. The above mechanism increases the analysis ability of the large language model in a specific field, enables it to better understand professional terms and complex situations in the field, and improves the pertinence and practical application value of improvement suggestions.

[0051] When elaborating on the relevant content of the first data pair and the second data pair, it is mentioned that "by using specific domain data for additional training on the basis of a pre-trained model". The "pre-trained model" here means that before using the large language model to analyze user feedback information, it is necessary to pre-train the large language model so that the large language model has the ability to analyze user feedback and improve the sensitivity of the large language model to specific domain information. Pre-training the large language model includes: using historical user feedback information and relevant improvement suggestions as training data, establishing data pairs of user feedback with function points, key problems, and improvement suggestions; using the above training data to adjust the feature parameters of the above large language model so that the above large language model has the ability to analyze user feedback.

[0052] It is understandable that historical user feedback information needs to be collected and sorted out first. This historical user feedback information can come from various channels, including in-app evaluations, app stores, questionnaires, social media discussions, customer service records, etc. Then, the historical user feedback information is paired with relevant improvement suggestions to establish the connection between user feedback and function points, as well as between key issues and improvement suggestions. This connection is presented in the form of data pairs. For example, the association between the "payment delay" problem and the "payment function", and the association between the "payment delay" problem and the improvement suggestion of "optimizing the payment interface".

[0053] Use the above data pairs to adjust the feature parameters of the large language model, which is the fine-tuning mentioned above. Although the large language model has been pre-trained on a large scale in the general field and has powerful natural language processing capabilities, in order to make the large language model perform better in specific user feedback analysis tasks, it is necessary to let the large language model learn the specific knowledge and rules in this field through fine-tuning. The fine-tuned large language model can understand the semantics in user feedback, including the expression of negative emotions, the description of function usage, and the specific symptoms of problems; the fine-tuned large language model can accurately map the information in user feedback to the specific function points in the user's use of the product; the fine-tuned large language model can also identify the core problems behind user feedback and generate reasonable improvement suggestions according to the nature of the problems.

[0054] The importance of pre-training is that it enables the large language model to have a certain amount of domain knowledge and context understanding ability before contacting actual user feedback. Therefore, in the subsequent analysis process, it can process the newly received feedback more quickly and accurately, and generate optimization suggestions that are more in line with user needs and the current situation of the product. In other words, by using historical data for fine-tuning, it is to let the large language model "re-learn" and learn how to better serve the improvement of the user experience.

[0055] After pre-training, the large language model has acquired knowledge and context understanding capabilities in the field of user feedback analysis, and this large language model can be put into use. The large language model is used to analyze the obtained user feedback information to obtain an analysis result, and the above initial knowledge graph is updated according to the above analysis result to obtain an updated knowledge graph, including: obtaining the above user feedback information, using the above large language model to extract relevant function points from the above user feedback information, and matching relevant key questions of the above relevant function points from the above initial knowledge graph; calculating the weighted values of each of the above relevant key questions, arranging the above relevant key questions in descending order of their weighted values to obtain a set of relevant key questions, and selecting some key questions from the above set of relevant key questions according to the first preset screening rule to obtain a final set of key questions; supplementing the above final set of key questions into the prompt words of the above large language model for analysis to obtain the final key questions; matching the above final key questions from the above initial knowledge graph, if a match is found, updating the node weights of the relevant key questions associated in the above initial knowledge graph and the node relationship weights between the above associated key questions and the corresponding function points, if no match is found, adding a new node and setting an initial weight to obtain a first updated knowledge graph.

[0056] Specifically, first, collect user feedback information about the product from multiple sources, and this information includes users' opinions, complaints or suggestions. Then, use the large language model to perform semantic analysis on each piece of feedback collected to identify the function modules mentioned in the feedback, that is, "relevant function points". For example, if the user feedback involves the usage experience of the payment function, the large language model will identify "payment function" as a relevant function point. The large language model matches the identified "relevant function points" with all the "relevant key questions" associated in the initial knowledge graph, and these relevant key questions have been identified and classified in previous user feedback, such as payment delay, payment security issues, etc. Then, use the formula X(m n )=q mn ×Q kmn to calculate the weighted value of each relevant key question, where n represents the total number of relevant key questions, m n represents the nth relevant key question, X(m n ) represents the weighted value of the relevant key question m n , q mn represents the initial weight of the relevant key question m n , and Q kmn represents the node relationship weight between the relevant key question m n and the corresponding function point. Arrange the calculated weighted values of each relevant key question in descending order to form a set of relevant key questions M={m1,m2,...,m n}, select some key issues from the above set of relevant key issues according to the first preset screening rule. The first preset screening rule is to select the nodes with the top i weighted values from the set of relevant key issues M = {m1, m2,..., m n} to obtain {m1, m2,..., m i}, {m1, m2,..., m i} as the final set of key issues.

[0057] Supplement {m1, m2,..., m i} into the prompt words of the large language model. Prompt words are a crucial input mechanism used to guide the model to generate specific types of responses or content. This input can be in text form, such as a question, a description, or a conversation history, etc., aiming to trigger the model to output in a certain specific way or style. The prompt words in the large language model can not only improve the accuracy and efficiency of the model but also make the output of the model more in line with the expectations and needs of users. By skillfully designing and optimizing the prompt words, users can more effectively utilize the large language model to generate text content that meets specific tasks or scenarios. At the same time, the prompt words also have a certain degree of flexibility, allowing users to guide the model in different ways, thus achieving flexible adaptation to different tasks and scenarios. In short, the role of prompt words is to guide the large language model to conduct more focused and in-depth analysis. By adding the final set of key issues to the prompt words, the large language model can more accurately identify and understand the core issues in the user feedback, which is the final key issue m'.

[0058] According to the final key issue m' identified by the large language model, check whether there is a corresponding key issue node in the initial knowledge graph. If a matching key issue node m' is found, that is, the final key issue m' has been included in the initial knowledge graph, then increase the weight of the final key issue m' by 1 on the original basis, and increase the relationship weight between the final key issue m' and its corresponding function point by 1 on the original basis; if the final key issue m' identified by the large language model is not matched in the initial knowledge graph, it means that the final key issue m' is a new issue. At this time, add the final key issue m' as a new key issue node to the initial knowledge graph, set the initial weight of the new key issue node to 1, and set the initial node relationship weight between the new key issue and its corresponding function point to 1.

[0059] After the above update process is completed, the first updated knowledge graph is obtained. The nodes and their relationship weights in the first updated knowledge graph are optimized or expanded, better reflecting the latest status and key issues of the user feedback.

[0060] Through the above update operations, not only can user feedback be analyzed quickly and accurately to identify key issues therein, but also the knowledge graph can be dynamically maintained and optimized to always maintain the latest and most comprehensive understanding of product feature points, problems encountered by users, and their solutions.

[0061] After obtaining the first updated knowledge graph, use the above first updated knowledge graph to match the associated improvement suggestions for the above final key issues, calculate the weighted values of the above associated improvement suggestions, and arrange the above associated improvement suggestions in descending order of their weighted values to obtain an associated improvement suggestion set, and select some improvement suggestions from the above associated improvement suggestion set according to the second preset screening rule to obtain a final improvement suggestion set; supplement the above final improvement suggestion set into the prompts of the above large language model for analysis to obtain an updated improvement suggestion set; match each improvement suggestion in the above updated improvement suggestion set with the above first updated knowledge graph. If a match is found, update the node weights of the related improvement suggestions in the above first updated knowledge graph and the node relationship weights between the above related improvement suggestions and the corresponding key issues. If no match is found, add a new node and set the initial weight.

[0062] Specifically, obtain the improvement suggestions associated with the final key issue m' obtained through the analysis of the large language model from the first updated knowledge graph, and use the formula X(r n ) = q m' ×(Q km' +Q m'r ) to calculate the weighted values of these improvement suggestions, where n represents the number of improvement suggestions associated with the final key issue m', X(r n ) represents the weighted value of the improvement suggestion, q m' represents the weight of the final key issue m', Q km' represents the node relationship weight between the final key issue m' and the corresponding feature point, and Q m'r represents the node relationship weight between the final key issue m' and the corresponding improvement suggestion. Arrange them in descending order of the weighted value to obtain an associated improvement opinion set R = {r1, r2,..., r n}, and select some improvement suggestions from the above associated improvement suggestion set according to the second preset screening rule to obtain a final improvement suggestion set. The second preset screening rule refers to screening out the first i nodes {r1, r2,..., r n} with the highest weighted values from the associated improvement opinion set R = {r1, r2,..., r i}, which is the final improvement suggestion set R = {r1, r2,..., r i}, supplement the set of final improvement suggestions into the prompt words of the large language model. The large language model analyzes the complete improvement plan for user feedback and outputs an updated set of improvement suggestions R' = {r1', r2',..., r n '}.

[0063] For each improvement suggestion in the obtained updated set of improvement suggestions R' = {r1', r2',..., r n '}, check whether there is already a node in the first updated knowledge graph. If the node of the improvement suggestion already exists, increase the weight of the improvement suggestion by 1 on the original basis, and at the same time increase the weight of the node relationship between the improvement suggestion and the corresponding key issue by 1 on the original basis. If the node of this improvement suggestion is not in the first updated knowledge graph, it means that this is a new improvement suggestion. Add the improvement suggestion as a newly added improvement suggestion node to the first updated knowledge graph, set the initial weight of the newly added improvement suggestion node to 1, establish the relationship between the newly added improvement suggestion and the corresponding key issue at the same time, and set its initial node relationship weight to 1.

[0064] Add the unmatched final key issues to the initial knowledge graph and set the initial weights, which is equivalent to learning new knowledge. Over time, the knowledge graph can continuously improve itself and form a dynamic learning framework. This continuous learning ability enables the system to adapt to the changing user needs and product environment, and maintain the timeliness and effectiveness of its analysis ability and improvement suggestion generation.

[0065] Through the above steps, not only can accurate and effective improvement suggestions be generated based on the current user feedback, but also the knowledge graph can be continuously optimized and expanded to ensure the comprehensiveness and timeliness of the graph content. This method creates a dynamic learning system in which the knowledge graph and the large language model promote each other and jointly improve the quality of feedback analysis and solution generation. Specifically, as a structured knowledge base, the knowledge graph provides rich background information and historical cases for the large language model, making the large language model analysis more accurate and comprehensive. The analysis results of the large language model, including the final key issues and the updated improvement suggestions, in turn enrich the content and structure of the knowledge graph, improving the intelligence and adaptability of the knowledge graph. This dynamic update and maintenance mechanism enables the knowledge graph to quickly adapt to the changes in user needs and the iteration of product functions, while ensuring that the analysis ability of the large language model is always in the optimal state, thus forming a positive cycle in terms of user experience improvement and product optimization.

[0066] To further enhance the functionality and intelligence level of this application, a context-aware mechanism can be introduced in user feedback analysis. Natural language processing technology is used to extract context information from user feedback, such as time, location, user behavior, etc. This information helps to more deeply understand the background of user needs and problems. Based on the extracted context features, the analysis parameters of the large language model are dynamically adjusted to create a personalized analysis model, enabling more personalized and accurate improvement suggestions to be provided according to different user contexts. By matching with the context features in the knowledge graph, solved problems and improvement suggestions similar to the current user feedback context can be given priority, further improving the analysis efficiency and the pertinence of the solutions.

[0067] The following is a specific embodiment of the improvement method for user feedback applying this application, as Figure 4 shown Figure 4The cross-analysis model in it includes a large language model and a knowledge graph. Collect user voices, that is, obtain user feedback information. Extract function points from the user feedback information through a pre-trained and fine-tuned large language model (LLM), and match them in the voice knowledge graph (initial knowledge graph) to construct a relationship network between function points and key issues, identify all key issues associated with the function points, calculate the weighted value of each key issue, sort the key issues according to their weighted values to form a set of key issues, and further use the large language model to analyze several issues with the highest weighted values in the set of key issues to output specific key issues as the "final key issues" of this analysis. This step ensures the depth and accuracy of the feedback analysis and focuses on the issues that most need to be solved. According to the obtained "final key issues", update the weights of relevant nodes and the relationship weights in the knowledge graph to construct a relationship network between optimization suggestions and key issues. If the key issue node already exists in the graph, its weight will be updated, otherwise a new node will be added. Next, obtain all improvement suggestions associated with the "final key issue" node, calculate the weighted values of the improvement suggestions, and sort them to form a set of improvement suggestions. This step provides a basis for generating subsequent improvement suggestions. Based on the obtained set of improvement suggestions, request the large language model to further analyze and output a complete system improvement plan. This plan considers several suggestions with the highest weighted values in the set of improvement suggestions to ensure the comprehensiveness and pertinence of the improvement suggestions. Match the improvement suggestions in the output improvement plan with the knowledge graph and update the weights of the improvement suggestion nodes and the relationship weights in the graph. If the improvement suggestion already exists in the graph, its weight increases, otherwise a new improvement suggestion node is added, so that the knowledge graph is updated. This step continuously optimizes the content of the graph to ensure that it covers the latest improvement suggestions and optimization ideas. Adjust the characteristic parameters of the large language model according to the relevant information of the updated knowledge graph to obtain an adjusted large language model, apply the adjusted large language model to analyze and obtain improvement suggestions corresponding to the user feedback information, and integrate and output a final improvement plan including immediate improvement measures and long-term optimization strategies.

[0068] Through this series of steps, leveraging the interaction between the large language model and the knowledge graph, automatically and efficiently analyze user feedback, generate improvement suggestions, and continuously optimize the large language model and the knowledge graph to form a closed-loop and adaptive user feedback conversion system. This process can not only quickly respond to user needs but also continuously learn and adjust according to the actual effects to ensure the continuous improvement and innovation of the optimization plan.

[0069] The following will be specifically described with embodiments:

[0070] The knowledge graph constructed based on historical data is as Figure 5As shown, the functional points include face recognition and account movement notifications; the key issues include frequent failures, insecure name display, no notification for deductions, automatic activation of charges, and missing functions; the improvement suggestions include optimizing the face recognition algorithm, developing value-added service management functions, immediately suspending the charges for unauthorized value-added services, and adding the function of face recognition for transfers.

[0071] Collect user feedback data from multiple sources and analyze one of the data items. Feedback 1: "I have two savings cards. Since I don't use them often, I negotiated repeatedly last year to cancel the SMS notification service. Today, without informing me, 30 yuan of SMS fees were deducted from each card to forcibly activate the SMS notification service. Here, I strongly request the return of the deducted SMS fees of 60 yuan and the cancellation of the SMS notification service! Otherwise, I reserve the right to appeal!"

[0072] The main function point "transaction notice" in feedback 1 is extracted by the large language model and matched from the knowledge graph to obtain all the key issues associated with the function point and calculate the weighted value of each key issue. The key issue set {"automatic activation of charges", "no notice of deduction", "unsafe display of name"} is obtained by arranging them in descending order of weighted value; the first two nodes {"automatic activation of charges", "no notice of deduction"} in the key issue set are supplemented to the prompt word, and the large language model is requested to analyze the key issues of user feedback and output the final key issue "automatic activation of charges"; the key issue nodes are matched from the knowledge graph and the node weights are updated, and the relationship weights between the function point and the key issues are updated; all the improvement opinion nodes associated with the key issue "automatic activation of charges" are obtained from the knowledge graph and the weighted values are calculated, and the improvement opinion set {"immediately suspend the value-added service charges without consent", "develop a value-added service management function"} is obtained by arranging them in descending order of weighted value; the first two nodes in the improvement opinion set are supplemented to the prompt word, and the large language model is requested to analyze the complete system improvement plan for user feedback; the improvement plan output by the large language model is extracted as a new improvement opinion set {"improve the user agreement to ensure that all service terms are clear and easy to understand"} and matched in the knowledge graph. If not matched, new nodes are added and the initial weights are set, and the relationship with the "automatic activation of charges" node is established and the relationship weight is set; the key issues, improvement opinion sets output by the large language model are paired with user feedback as the data for the next round of fine-tuning training of the large language model to continuously improve the large model's analysis ability for user feedback; the system improvement plan generated by the large language model is integrated and output. For example, improvement opinions: immediately suspend all value-added service charges that have not been explicitly consented to by users; develop a value-added service management function to allow users to manage their own value-added services; redesign the user agreement to ensure that all service terms are clear and easy to understand; regularly collect user feedback and continuously optimize the service. Implementation steps: 1) Suspend all value-added service deductions that have not been consented to by users. 2) Send a notice to affected users to explain the situation and provide refund options. 3) Develop and launch a value-added service management function to allow users to customize value-added services. 4) Simplify the language of the user agreement and highlight important terms. 5) Establish a user feedback channel and regularly analyze feedback data for product and service improvement. 6) Regularly check and adjust value-added service packages to better match user needs.

[0073] The embodiments of the present application also provide an improvement device for user feedback. It should be noted that the improvement device for user feedback in the embodiments of the present application can be used to execute the improvement method for user feedback provided by the embodiments of the present application. The device for implementing the above embodiments and preferred implementation manners has been described and will not be repeated here. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0074] The following introduces the improvement device for user feedback provided by the embodiments of the present application.

[0075] Figure 6 is a structural block diagram of the improvement device for user feedback according to the embodiments of the present application. As Figure 6 shown, the device includes a construction unit 10, an update unit 20, and an adjustment unit 30. Among them, the construction unit 10 is used to construct an initial knowledge graph. The nodes of the above initial knowledge graph include function points, key problems, and improvement suggestions. The above function points are function modules that users have when using the product; the update unit 20 is used to obtain user feedback information, analyze the above user feedback information using a large language model to obtain an analysis result, and update the above initial knowledge graph according to the above analysis result to obtain an updated knowledge graph, where the update operation includes at least one of the following: adjusting the node weight, adjusting the relationship weight between nodes, and adding new nodes. The above relationship weight between nodes represents the degree of association between nodes; the adjustment unit 30 is used to adjust the feature parameters of the above large language model according to the relevant information of the above updated knowledge graph to obtain an adjusted large language model, and use the above adjusted large language model to analyze and obtain improvement suggestions corresponding to the above user feedback information. Among them, the above relevant information includes a first data pair and a second data pair. The above first data pair includes the above user feedback information and its corresponding function point, and the above second data pair includes the key problem associated with the above function point and the improvement suggestion corresponding to the above key problem.

[0076] Specifically, the construction unit is used to create a structured framework for storing and associating all function modules (i.e., function points) in the software system, the key problems mentioned in the user feedback, and the improvement suggestions for these problems. Each node in the initial knowledge graph represents an independent function point in the software system, a key problem reflected by the user, or a suggestion for solving these problems.

[0077] By constructing the above initial knowledge graph, it is possible to quickly locate the key issues under a certain function point and the existing improvement suggestions for these issues, so as to provide more accurate and rich prompts to the large language model in subsequent analysis and assist it in making more reasonable analysis decisions. Generally speaking, constructing the initial knowledge graph is to establish a basic platform for efficient and accurate feedback analysis and improvement suggestion generation, which can automatically and intelligently process user feedback, greatly accelerating the feedback conversion process and improving the conversion quality and efficiency.

[0078] The update unit is used to input the obtained user feedback information into the large language model for in-depth analysis to identify the deep meaning, sentiment tendency, function requirements, and potential key issues in the user feedback information. With its powerful natural language understanding and generation capabilities, the large language model can quickly process complex text information, extract valuable data from it, and form an analysis result.

[0079] Then, the above initial knowledge graph is dynamically updated according to these analysis results. The update operations mainly include three types: adjusting the node weights, adjusting the relationship weights between nodes, and adding nodes. Specifically, the node weight reflects the importance of the function point, key issue, or improvement suggestion. The relationship weight between nodes measures the association strength between the function point and the key issue, and between the key issue and the improvement suggestion. In addition, when the large language model identifies new key issues or improvement suggestions based on the function points extracted from the user feedback information, corresponding nodes will be added to the initial knowledge graph to expand the information coverage of the initial knowledge graph and obtain the updated knowledge graph. This mechanism of dynamically adding nodes ensures that the knowledge graph can promptly reflect the new changes in the software system and the new needs of users.

[0080] Through the above update operations, the large language model feeds back the analysis results to the initially constructed initial knowledge graph, and the initial knowledge graph is continuously updated, making the updated knowledge graph more powerful and accurate. For example, when a key issue is first output by the large language model, this key issue will be added to the knowledge graph and connected to the relevant function points. If subsequent user feedback continues to mention this key issue, then the node weight of this key issue and the relationship weight with the corresponding function point will increase, reflecting the importance of this key issue; at the same time, once an effective improvement suggestion is found, the relationship weight between this key issue and the corresponding improvement suggestion will also increase, making it easier to be associated and recommended by the large language model in future analysis.

[0081] In summary, the update unit is used to analyze user feedback information through the large language model and update the initial knowledge graph in real time, realizing the self-improvement and optimization of the knowledge graph, and ensuring that it can always reflect the latest and most relevant information.

[0082] The adjustment unit is used to further optimize the large language model using the updated knowledge graph, so as to improve the accuracy and comprehensiveness of the large language model in analyzing user feedback, thereby enhancing the analysis efficiency. This process is achieved by adjusting the feature parameters of the large language model. Feature parameters are important parameters within the large language model that affect the prediction results. Adjusting the feature parameters can enable the large language model to better understand and process specific types of data. The process of adjusting feature parameters is usually iterative. By continuously experimenting and evaluating the performance of the model on specific tasks (such as the accuracy of user feedback analysis), a set of parameter combinations that best suit the current requirements can be found. In this embodiment, the updated knowledge graph provides additional supervision signals to help guide the adjustment of feature parameters, enabling the large language model to more effectively understand and convert user feedback into improvement suggestions.

[0083] Further, the relevant information after using the updated knowledge graph, that is, the first data pair and the second data pair, is used to guide the adjustment of the feature parameters of the large language model. Among them, the first data pair includes user feedback information and its corresponding function point. For example, if the user feedback is about the delay problem of the payment function, then "payment delay" and "payment function" constitute a pair of first data pairs. The first data pair helps the large language model learn how different user feedback texts should be mapped to specific function modules of the software system. The second data pair includes key issues associated with the function point (such as the delay problem in the payment function) and improvement suggestions for these key issues (such as optimizing the payment interface). The second data pair guides the large language model on how to generate reasonable and effective improvement suggestions starting from the identified key issues.

[0084] The first data pair and the second data pair are used to adjust the large language model, that is, to fine-tune the large language model. Fine-tuning is a machine learning technique that, based on a pre-trained model, uses data in a specific domain for additional training, so that the model can better meet the requirements of that domain. In this process, the large language model adjusts its internal feature parameters according to the association between user feedback and function points, the nature of key issues, and the effectiveness of improvement suggestions in the first data pair and the second data pair, such as modifying the weights of word embeddings, changing the configuration of the attention mechanism, etc., to more accurately capture the intention of user feedback and generate appropriate improvement suggestions based on this.

[0085] The adjusted large language model can, when receiving new user feedback information, more quickly identify its associated function points, more accurately refine key issues, and generate improvement suggestions that are highly relevant and targeted to the key issues based on the rich improvement suggestions in the knowledge graph. This process not only improves the efficiency of feedback analysis but also ensures the quality of improvement suggestions, helps the software development team quickly locate and solve problems, and continuously optimize the user experience.

[0086] In short, the adjustment unit is used to specifically adjust the large language model by extracting key semantic and relational information from the updated knowledge graph, enabling the model to better understand questions and generate more accurate solutions when processing user feedback.

[0087] Through this embodiment, by constructing and maintaining an initial knowledge graph containing function points, key issues, and improvement suggestions, and combining the analysis capabilities of the large language model, intelligent analysis of user feedback information is achieved. It can automatically identify the function modules mentioned in user feedback, extract key issues, and update the initial knowledge graph based on the analysis results of the large language model. Among them, adjusting the node weights reflects the relative importance of function points, issues, or suggestions, adjusting the relationship weights between nodes strengthens the associations between various parts, and adding new nodes expands the coverage of the knowledge graph. This dynamic optimization ensures that the knowledge graph can continuously reflect the real needs of users and the latest state of the product, thereby improving the efficiency and accuracy of feedback analysis. The update of the knowledge graph not only enriches the training data of the large language model but also enables the large language model to better understand the context and intention of user feedback by adjusting the feature parameters of the large language model. At the same time, the analysis results of the large language model are used for the expansion and optimization of the knowledge graph, forming a closed feedback loop. This two-way empowerment of the model and the graph improves the flexibility and intelligence of user analysis, making the analysis results more accurate and the improvement suggestions more effective. In addition, since both the update of the knowledge graph and the fine-tuning of the large language model are carried out after receiving new user feedback, this method supports continuous learning and iteration. Over time, it will be able to better understand and respond to user feedback, continuously optimize its analysis capabilities and the quality of improvement suggestions, forming a positive cycle and continuously enhancing the product and service.

[0088] In this embodiment, by injecting the answers output by the large language model into the initial knowledge graph, continuously expanding and improving the knowledge graph relationship network, and then adjusting the large language model through the more perfect knowledge graph, the answers output by the large language model are made more accurate and comprehensive, thereby improving the efficiency of user feedback analysis.

[0089] In the specific implementation process, the above-mentioned construction unit includes a first acquisition module, a setting module, and a first establishment module. Among them, the first acquisition module is used to acquire historical user feedback information and related improvement suggestions, and extract the above-mentioned function points and the above-mentioned key issues from the above-mentioned historical user feedback information; the setting module is used to set initial node weights and initial relationship weights between nodes for the above-mentioned function points, the above-mentioned key issues, and the above-mentioned improvement suggestions according to preset rules; the first establishment module is used to establish node connections according to the corresponding relationships of the above-mentioned function points, the above-mentioned key issues, and the above-mentioned improvement suggestions, form a relationship network between the above-mentioned function points and the above-mentioned key issues, and between the above-mentioned key issues and the above-mentioned improvement suggestions, and obtain the above-mentioned initial knowledge graph.

[0090] Regarding the construction of the above initial knowledge graph, first, historical user feedback information and relevant improvement suggestions are collected. Through natural language processing techniques, especially using large language models, information related to product functions is extracted from these historical user feedback information, which are the functional points. At the same time, the large language model identifies the main problems in the historical user feedback information, that is, the key problems. For example, from the user feedback "Every time I use the payment function, I have to wait a long time to complete the transaction", the "payment function" can be extracted as the functional point, and "delay during payment" as the key problem, and relevant improvement suggestions for the key problem are obtained.

[0091] To quantify and measure the importance of each node and its interrelationships, the above-mentioned setting module includes a setting sub-module, which is used to set an initial weight for each node according to the function priority and the number of times a key issue appears, establish node connections based on the correspondence between function points, key issues, and improvement suggestions to form a relationship network among function points, key issues, and improvement suggestions. At the same time, set the initial weight of the relationship between the function point node and the key issue node, and the initial weight of the relationship between the key issue node and the improvement suggestion node according to the number of times the improvement suggestion is adopted. The node weight reflects the relative importance of the node in the knowledge graph. Set the initial weight for each key issue node according to the function priority and the number of times the key issue appears. For example, the "transaction function" is given a high weight because the "transaction function" directly involves the core service of the product according to the function priority, while the weight of the "color selection on the setting page" is relatively low because it has a relatively small impact on user satisfaction. Another example is that in the historical user feedback data, some problems or defects are repeatedly mentioned by multiple users. The statistics of the number of times a key issue appears reflect the universality and urgency of this issue. Therefore, the more times a problem is reported, the higher the node weight of it in the knowledge graph, which ensures that key issues that have a wide impact and are strongly reflected by users can be prioritized for processing and resolution. At the same time, set the initial weight of the relationship between the function point node and the key issue node according to the number of times the improvement suggestion is adopted. The number of times an improvement suggestion is adopted reflects the effectiveness and implementation value of this improvement suggestion. If an improvement suggestion is adopted and successfully implemented multiple times, it indicates that this improvement suggestion has a significant effect in solving key issues and improving the user experience. Therefore, the more times an improvement suggestion is adopted, the higher its node weight, which helps to quickly retrieve and recommend effective solutions when facing similar problems. The number of times an improvement suggestion is adopted reflects its effectiveness and popularity in past feedback analysis and optimization practices. Suggestions with a large number of adoptions have a high initial weight in the knowledge graph, which helps to tend to propose solutions that have been proven effective or widely praised when generating improvement suggestions, thus enhancing the pertinence and practical value of the suggestions. This setting ensures that the knowledge graph not only reflects the importance of each node but also embodies the intensity of the interaction between entities, so that the graph can more accurately guide model analysis and the generation of optimization suggestions.

[0092] By setting the initial node weights and the initial weights of the relationships between nodes through the above preset rules, the knowledge graph can become an intelligent and dynamically adjustable decision support tool, helping the large language model to quickly focus on important functions, urgent issues, and effective solutions when analyzing new user feedback, thereby improving the efficiency and accuracy of the entire feedback analysis process.

[0093] In summary, the process of constructing the initial knowledge graph lays the foundation for generating intelligent analysis and improvement suggestions for subsequent user feedback. It ensures that when analyzing new feedback, the large language model can be based on existing structured knowledge and historical experience, so as to more accurately understand the feedback content, extract key issues, and propose targeted improvement measures.

[0094] Furthermore, the above device further includes a second establishment module and an adjustment module. Among them, the second establishment module is used to use the above historical user feedback information and the above relevant improvement suggestions as training data to establish data pairs of user feedback and function points, key issues and improvement suggestions; the adjustment module is used to adjust the characteristic parameters of the above large language model by using the above training data, so that the above large language model has the ability to analyze user feedback.

[0095] It can be understood that first, it is necessary to collect and sort out historical user feedback information. Then, pair the historical user feedback information with relevant improvement suggestions to establish the connection between user feedback and function points, key issues and improvement suggestions. This connection is presented in the form of data pairs, such as the association between the "payment delay" problem and the "payment function", and the association between the "payment delay" problem and the improvement suggestion of "optimizing the payment interface".

[0096] Use the above data pairs to adjust the characteristic parameters of the large language model, which is the fine-tuning mentioned above. Although the large language model has been pre-trained on a large scale in the general field and has powerful natural language processing capabilities, in order to make the large language model perform better in specific user feedback analysis tasks, it is necessary to let the large language model learn the specific knowledge and rules in this field through fine-tuning. The fine-tuned large language model can understand the semantics in user feedback, including the expression of negative emotions, the description of function usage, and the specific symptoms of problems; the fine-tuned large language model can accurately map the information in user feedback to the specific function points in the user's use of the product; the fine-tuned large language model can also identify the core issues behind user feedback and can generate reasonable improvement suggestions according to the nature of the problems.

[0097] After pre-training, the large language model has the knowledge and context understanding ability in the field of user feedback analysis, and this large language model can be put into use. The above update unit includes a second acquisition module, a first calculation module, a first supplement module, and a first matching module. Among them, the second acquisition module is used to acquire the above user feedback information, extract relevant function points from the above user feedback information by using the above large language model, and match relevant key questions of the above relevant function points from the above initial knowledge graph; the first calculation module is used to calculate the weighted values of each of the above relevant key questions, arrange the relevant key questions in descending order according to the weighted values of each of the above relevant key questions to obtain a set of relevant key questions, and select some key questions from the above set of relevant key questions according to the first preset screening rule to obtain a final set of key questions; the first supplement module is used to supplement the above final set of key questions into the prompt words of the above large language model for analysis to obtain the final key questions; the first matching module is used to match the above final key questions from the above initial knowledge graph. If a match is found, the node weights of the relevant key questions associated in the above initial knowledge graph and the node relationship weights between the above associated key questions and the corresponding function points are updated. If no match is found, new nodes are added and initial weights are set to obtain a first updated knowledge graph.

[0098] Specifically, the second acquisition module is used to collect user feedback information on the product from multiple sources, and this information includes user opinions, complaints, or suggestions. Then, the large language model is used to perform semantic analysis on each piece of feedback collected to identify the functional modules mentioned in the feedback, that is, "relevant function points". For example, if the user feedback involves the usage experience of the payment function, the large language model will identify "payment function" as the relevant function point. The large language model matches the identified "relevant function points" with all the "relevant key questions" associated in the initial knowledge graph, and these relevant key questions have been identified and classified in previous user feedback, such as payment delay, payment security issues, etc. The first calculation module is used to calculate the weighted value of each relevant key question. The calculated weighted values of each relevant key question are arranged in descending order to form a set of relevant key questions, and some key questions are selected from the above set of relevant key questions according to the first preset screening rule to obtain a final set of key questions.

[0099] The first supplementary module is used to supplement the set of final key questions into the prompts of the large language model. Prompts are a crucial input mechanism for guiding the model to generate specific types of responses or content. Such input can be in text form, such as a question, a description, or a conversation history, etc., aiming to trigger the model to output in a certain specific way or style. Prompts in the large language model can not only improve the accuracy and efficiency of the model but also make the output of the model more in line with the expectations and needs of users. By skillfully designing and optimizing the prompts, users can more effectively utilize the large language model to generate text content that meets specific tasks or scenarios. At the same time, the prompts also have a certain degree of flexibility, allowing users to guide the model in different ways, thus achieving flexible adaptation to different tasks and scenarios. In short, the role of prompts is to guide the large language model to conduct more focused and in-depth analysis. By adding the set of final key questions to the prompts, the large language model can more accurately identify and understand the core issues in user feedback, which are the final key questions.

[0100] The first matching module is used to search in the initial knowledge graph to see if there are corresponding key question nodes for the final key questions identified by the large language model. If a matching key question node is found, that is, the final key question has been included in the initial knowledge graph, then the weight of the final key question is incremented by 1 on the original basis, and the relationship weight between the final key question and its corresponding function point is incremented by 1 on the original basis.

[0101] If the final key questions identified by the large language model are not matched in the initial knowledge graph, it means that the final key questions are new questions. At this time, the final key questions are added to the initial knowledge graph as new key question nodes, and the initial weight of the new key question nodes is set to 1. At the same time, the initial node relationship weight between the new key questions and their corresponding function points is set to 1.

[0102] After the above update process is completed, the first updated knowledge graph is obtained. The nodes and their relationship weights in the first updated knowledge graph are optimized or expanded, better reflecting the latest status and key issues of user feedback. Through the above update operations, not only can user feedback be analyzed quickly and accurately to identify the key issues therein, but also the knowledge graph can be dynamically maintained and optimized to always maintain the latest and most comprehensive understanding of product function points, problems encountered by users, and their solutions.

[0103] The above device further includes a second calculation module, a second supplement module, and a second matching module. Among them, the second calculation module is used to, after obtaining the first updated knowledge graph, match the associated improvement suggestions for the above final key questions by applying the above first updated knowledge graph, calculate the weighted values of the above associated improvement suggestions, and arrange the above associated improvement suggestions in descending order of the weighted values to obtain an associated improvement suggestion set, and select some improvement suggestions from the above associated improvement suggestion set according to the second preset screening rule to obtain a final improvement suggestion set; the second supplement module is used to supplement the above final improvement suggestion set into the prompt words of the above large language model for analysis to obtain an updated improvement suggestion set; the second matching module is used to match each improvement suggestion in the above updated improvement suggestion set with the above first updated knowledge graph. If a match is found, update the node weights of the related improvement suggestions in the above first updated knowledge graph and the node relationship weights between the above related improvement suggestions and the corresponding key questions. If no match is found, add a new node and set an initial weight.

[0104] Specifically, obtain the improvement suggestions associated with the final key questions obtained through the analysis of the large language model from the first updated knowledge graph, and calculate the weighted values of these improvement suggestions. Arrange them in descending order of the weighted values to obtain an associated improvement suggestion set, select some improvement suggestions from the above associated improvement suggestion set according to the second preset screening rule to obtain a final improvement suggestion set, supplement the final improvement suggestion set into the prompt words of the large language model, and the large language model analyzes the complete improvement plan for the user feedback to output an updated improvement suggestion set.

[0105] For each improvement suggestion in the obtained updated improvement suggestion set, check whether there is already a node in the first updated knowledge graph. If the node of the improvement suggestion already exists, add 1 to the weight of the improvement suggestion on the original basis, and at the same time add 1 to the node relationship weight between the improvement suggestion and the corresponding key question on the original basis. If the node of the improvement suggestion is not in the first updated knowledge graph, it means that this is a new improvement suggestion. Add the improvement suggestion as a newly added improvement suggestion node to the first updated knowledge graph, set the initial weight of the newly added improvement suggestion node to 1, establish the relationship between the newly added improvement suggestion and the corresponding key question at the same time, and set its initial node relationship weight to 1.

[0106] Adding the unmatched final key questions to the initial knowledge graph and setting the initial weights is equivalent to learning new knowledge. Over time, the knowledge graph can continuously improve itself, forming a dynamic learning framework. This continuous learning ability enables the system to adapt to the ever-changing user needs and product environment, and maintain the timeliness and effectiveness of its analysis ability and improvement suggestion generation.

[0107] Through the above steps, not only can accurate and effective improvement suggestions be generated based on the current user feedback, but also the knowledge graph can be continuously optimized and expanded to ensure the comprehensiveness and timeliness of the graph content. This method creates a dynamic learning system in which the knowledge graph and the large language model promote each other and jointly improve the quality of feedback analysis and solution generation. Specifically, as a structured knowledge base, the knowledge graph provides rich background information and historical cases for the large language model, making the analysis of the large language model more accurate and comprehensive. The analysis results of the large language model, including the final key issues and updated improvement suggestions, in turn enrich the content and structure of the knowledge graph, improving the intelligence and adaptability of the knowledge graph. This dynamic update and maintenance mechanism enables the knowledge graph to quickly adapt to changes in user needs and product function iterations, while ensuring that the analysis ability of the large language model is always in the optimal state, thus forming a positive cycle in terms of user experience improvement and product optimization.

[0108] The above-mentioned improvement device for user feedback includes a processor and a memory. The above-mentioned construction unit, update unit, adjustment unit, etc. are all stored in the memory as program units, and the corresponding functions are implemented by the processor executing the above-mentioned program units stored in the memory. The above-mentioned modules are all located in the same processor; or, the above-mentioned each module is located in different processors in any combination form.

[0109] The memory may include non-permanent memory in a computer-readable medium, forms such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0110] The embodiment of the present invention provides a computer-readable storage medium. The above-mentioned computer-readable storage medium includes a stored program, wherein when the above-mentioned program runs, it controls the device where the above-mentioned computer-readable storage medium is located to execute the above-mentioned improvement method for user feedback.

[0111] The embodiment of the present invention provides a processor. The above-mentioned processor is used to run a program, wherein when the above-mentioned program runs, it executes the above-mentioned improvement method for user feedback.

[0112] The embodiment of the present invention provides an electronic device, including a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-mentioned improvement method for user feedback.

[0113] The present application also provides a computer program product, which is suitable for executing a program initialized with the steps of the above-mentioned improvement method for user feedback when executed on a data processing device.

[0114] Obviously, those skilled in the art should understand that the various modules or steps of the present invention described above can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed over a network composed of multiple computing devices. They can be implemented by program code executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order than here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. In this way, the present invention is not limited to any specific combination of hardware and software.

[0115] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0116] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or multiple flows and / or blocks

[0117] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions specified in Figure 1 one or more of the flows Figure 1 or multiple flows and / or blocks

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the functions in the flowFigure 1 one or more processes and / or blocks Figure 1 steps of functions specified in one or more blocks.

[0119] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0120] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash RAM. The memory is an example of computer-readable media.

[0121] Computer-readable media includes permanent and non-permanent, removable and non-removable media and can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes or other magnetic storage devices, or any other non-transitory media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0122] It should also be noted that the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0123] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An improved method for user feedback, characterized in that, Including: Construct an initial knowledge graph, where the nodes of the initial knowledge graph include function points, key issues, and improvement suggestions. The function points are the functional modules that users have when using the product; Obtain user feedback information, analyze the user feedback information using a large language model to obtain an analysis result, and update the initial knowledge graph according to the analysis result to obtain an updated knowledge graph. The update operations include at least one of the following: adjusting the node weights, adjusting the relationship weights between nodes, and adding new nodes. The relationship weights between nodes represent the degree of association between nodes; Adjust the feature parameters of the large language model according to the relevant information in the updated knowledge graph to obtain an adjusted large language model, and use the adjusted large language model to analyze and obtain improvement suggestions corresponding to the user feedback information. Among them, the relevant information includes the first data pair and the second data pair. The first data pair includes the user feedback information and its corresponding function point, and the second data pair includes the key issues associated with the function point and the improvement suggestions corresponding to the key issues.

2. The method according to claim 1, wherein Obtain user feedback information, analyze the user feedback information using a large language model to obtain an analysis result, and update the initial knowledge graph according to the analysis result to obtain an updated knowledge graph, including: Obtain the user feedback information, use the large language model to extract relevant function points from the user feedback information, and match the relevant key issues of the relevant function points from the initial knowledge graph; Calculate the weighted values of each relevant key issue, arrange the relevant key issues in descending order of their weighted values to obtain a set of relevant key issues, and select some key issues from the set of relevant key issues according to the first preset screening rule to obtain a final set of key issues; Supplement the final set of key issues into the prompt words of the large language model for analysis to obtain the final key issues; Match the final key issues from the initial knowledge graph. If a match is found, update the node weights of the relevant key issues in the initial knowledge graph and the relationship weights between the relevant key issues and the corresponding function points. If no match is found, add a new node and set the initial weight to obtain a first updated knowledge graph.

3. The method according to claim 2, characterized in that, After matching the final key issues from the initial knowledge graph, if a match is found, update the node weights of the relevant key issues in the initial knowledge graph and the relationship weights between the relevant key issues and the corresponding function points. If no match is found, add a new node and set the initial weight. The method includes: Use the first updated knowledge graph to match the associated improvement suggestions for the final key issues, calculate the weighted values of the associated improvement suggestions, arrange the associated improvement suggestions in descending order of their weighted values to obtain a set of associated improvement suggestions, and select some improvement suggestions from the set of associated improvement suggestions according to the second preset screening rule to obtain a final set of improvement suggestions; Supplement the set of the final improvement suggestions into the prompts of the large language model for analysis to obtain an updated set of improvement suggestions; Match each improvement suggestion in the updated set of improvement suggestions with the first updated knowledge graph. If a match is found, update the node weights of the related improvement suggestions and the node - to - node relationship weights between the related improvement suggestions and the corresponding key issues in the first updated knowledge graph. If no match is found, add a new node and set the initial weight.

4. The method according to claim 2, wherein If no match is found, add a new node and set the initial weight to obtain the first updated knowledge graph, including: If the final key issue is not found in the initial knowledge graph, add the final key issue as a node to the initial knowledge graph; Set the initial weight of the final key issue and set the initial relationship weight between the final key issue and the corresponding function point node to obtain the first updated knowledge graph.

5. The method according to claim 1, wherein Construct an initial knowledge graph. The nodes of the initial knowledge graph include function points, key issues, and improvement suggestions. The function points are the functional modules that users have when using the product, including: Obtain historical user feedback information and related improvement suggestions, and extract the function points and the key issues from the historical user feedback information; Set initial node weights and initial node - to - node relationship weights for the function points, the key issues, and the improvement suggestions according to preset rules; Establish node connections according to the corresponding relationships among the function points, the key issues, and the improvement suggestions to form a relationship network between the function points and the key issues, and between the key issues and the improvement suggestions, to obtain the initial knowledge graph.

6. The method according to claim 5, wherein Set initial node weights and initial node - to - node relationship weights for the function points, the key issues, and the improvement suggestions according to preset rules, including: Set the node weights and the node - to - node relationship weights according to function priorities, the number of times key issues occur, and the number of times improvement suggestions are adopted.

7. The method according to claim 5, characterized in that, Before using the large language model to analyze the user feedback information, the method further includes: Use the historical user feedback information and the related improvement suggestions as training data to establish data pairs of user feedback, function points, key issues, and improvement suggestions; Use the training data to adjust the feature parameters of the large language model so that the large language model has the ability to analyze user feedback.

8. An improved device for user feedback, characterized in that, including: A construction unit for constructing an initial knowledge graph. The nodes of the initial knowledge graph include function points, key issues, and improvement suggestions. The function points are the functional modules that users have when using the product; An update unit for obtaining user feedback information, using a large language model to analyze the user feedback information to obtain an analysis result, and updating the initial knowledge graph according to the analysis result to obtain an updated knowledge graph, where the update operations include at least one of the following: adjusting node weights, adjusting node - to - node relationship weights, adding new nodes. The node - to - node relationship weight represents the degree of association between nodes; An adjustment unit is configured to adjust the feature parameters of the large language model according to the relevant information of the updated knowledge graph, so as to obtain an adjusted large language model, and analyze the adjusted large language model to obtain improvement suggestions corresponding to the user feedback information. The relevant information includes a first data pair and a second data pair. The first data pair includes the user feedback information and its corresponding function point, and the second data pair includes the key issues associated with the function point and the improvement suggestions corresponding to the key issues.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute the improvement method for user feedback according to any one of claims 1 to 7.

10. An electronic device, characterized in that, Comprising: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include the improvement method for user feedback according to any one of claims 1 to 7.

Images