Model updating method, apparatus, and system

Abstract

Provided are a model updating method, an apparatus, and a system. In the model updating method, a computer device displays a first processing result obtained according to system data of a model processing network system. A user can perform a knowledge data input operation upon seeing the first processing result. The computer device then obtains target knowledge data after the knowledge data undergoes a format conversion. The computer device then responds to a model updating instruction, uses the target knowledge data to update a model, obtains an updated model, and finally, obtains a second processing result according to system data of the updated model processing network system, and displays the second processing result. This enables a user to directly participate in the optimization process of a model, and updates the model of a user in real time to meet changing service requirements and the personalized expectations of the user.

Description

Model updating method, device and system

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office on February 29, 2024, with application number 202410234970.1 and application name “Model Updating Method, Device and System”, all contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of computers, and in particular to a model updating method, device, and system. Background Art

[0003] Models can be used in a variety of fields to perform various data processing tasks. For example, in network operations and maintenance, models can be used to analyze operational data such as network system logs to identify anomalies or failures in the network system, discover the causes of these anomalies or failures, and predict future failures.

[0004] Model updates are necessary to adapt to changing user business needs and improve model quality, such as accuracy. However, currently, model providers must regularly release new models, and users must then opt in to obtain these new models to implement updates. Alternatively, users must regularly communicate with model providers to share their model expectations, and the providers must manually update the models to meet these expectations. This results in a long and cumbersome model update cycle, making it impossible to instantly update models to meet changing business needs and user expectations.

[0005] Summary of the Invention

[0006] The present application provides a model updating method, device and system, which can solve the problem that the model provider performs model updating, resulting in a long and cumbersome operation cycle and the inability to update the model in real time to meet changing business needs and users' personalized expectations.

[0007] In a first aspect, a model updating method is provided, which is executed by a computer device, including displaying a first processing result, the first processing result including a first operation analysis result of the network system obtained by processing the system data of the network system according to the model, and responding to a knowledge data input operation to obtain target knowledge data, which includes data after format conversion of the knowledge data, wherein the knowledge data includes knowledge data input by a user and related to the processing result of the model, and also responding to a model update instruction, using the target knowledge data to update the model to obtain an updated model, and finally processing the system data of the network system according to the updated model to obtain a second processing result and displaying the second processing result, the second processing result including a second operation analysis result of the network system.

[0008] Compared to model updates performed by the model provider, such as the model provider regularly releasing new models or manually updating the user model based on user feedback, the model update method provided in this application allows the computer device running the model to perform model updates locally, eliminating the need for the model provider's long and cumbersome operation cycle, thereby shortening the response time of the model update. Furthermore, the computer device can receive knowledge data input by the user based on business needs and personalized expectations. The computer device can use this knowledge data to intelligently and customized update the model, and the updated model can better meet changing business needs and personalized user expectations.

[0009] In a possible implementation, responding to the knowledge data input operation includes receiving feedback operation data, where the feedback operation data includes a first processing result and a feedback identifier, and the feedback identifier is used to indicate a method for updating the model based on the first processing result.

[0010] By obtaining the user's feedback operation data on the first processing result, the user's feedback identifier for the first processing result, such as positive feedback or negative feedback, can be obtained, and then the user's positive feedback or negative feedback can be used to indicate the model update, so that the user can perform the model update operation intelligently and simply, and the updated model can better meet the user's personalized expectations and business needs.

[0011] In a possible implementation, responding to the knowledge data input operation includes: receiving a user trigger operation on a feedback control displayed on a computer device, and the computer device receiving feedback operation data.

[0012] By obtaining feedback operation data through the user's triggering operation on the feedback control displayed on the computer device, the user operation can be made more efficient and the model update operation can be performed more intelligently.

[0013] In another possible implementation, obtaining target knowledge data in response to a knowledge data input operation includes: performing binary format conversion on feedback operation data to obtain first knowledge data included in the target knowledge data.

[0014] By performing binary conversion on the user's feedback operation data on the first processing result, such as a feedback identifier indicating positive feedback or negative feedback, binary data that can be better recognized and processed by a computer can be obtained, so that the model update operation can be completed more smoothly.

[0015] In another possible implementation, responding to the knowledge data input operation includes: receiving experience text input by the user.

[0016] The experience text can include valuable experience information accumulated by users. By inputting the experience text by users, the existing experience can be fully utilized to update the model, better improve the model effect and better meet business needs, and enable users to perform model update operations intelligently and simply.

[0017] In another possible implementation, in response to the knowledge data input operation, obtaining the target knowledge data includes: performing character string format conversion on the experience text to obtain the second knowledge data included in the target knowledge data.

[0018] By converting the user-entered experience text into a string format, field data organized according to certain specifications that can be better recognized and processed by the computer can be obtained, making the model update operation more smoothly completed.

[0019] In another possible implementation, responding to the knowledge data input operation also includes: receiving a text extraction intention input by a user; converting the experience text into a string format to obtain the second knowledge data included in the target knowledge data, including: extracting text information from the experience text according to the text extraction intention to obtain the target text; converting the target text into a string format to obtain the second knowledge data.

[0020] Users can provide text extraction intentions, such as "alarm operation and maintenance", and then use text extraction intentions and language models to perform targeted information extraction on the experience text to obtain text related to "alarm operation and maintenance". The text format related to "alarm operation and maintenance" is then converted for model update, so that when the model is updated, the model will be specifically adjusted to make the model processing results related to "alarm operation and maintenance" produce the expected changes, achieving more intelligent and targeted satisfaction of users' personalized needs for model updates.

[0021] In another possible implementation, the model update instruction includes the weight of the target knowledge data, and updating the model using the target knowledge data includes: adjusting the parameters of the model using the target knowledge data, wherein the adjustment range of the parameters is related to the weight of the target knowledge data.

[0022] The weights in the model update instructions can be determined based on user operations or set by the user. The parameter adjustment range during model update is determined by the weights, which can make the parameter adjustment during model update better meet the user's expectations, and the updated model can also better meet the user's personalized expectations.

[0023] In another possible implementation, the size of the weight is related to the number of times the user performs feedback operations on the first processing result.

[0024] The weight of the knowledge data is directly determined according to the number of user feedback operations, which enables the user to perform model update operations intelligently and simply.

[0025] In another possible implementation, the computer device further displays a prompt indicating that acquisition of the target knowledge data is complete.

[0026] The computer device displays a data completion prompt, which can prompt the user to perform subsequent model updates more intuitively and efficiently, providing a better user experience.

[0027] In another possible implementation manner, the computer device further displays target knowledge data.

[0028] Displaying target knowledge data on computer devices allows users to intuitively see the data content, helps users ensure data accuracy, and provides a better user experience.

[0029] In a second aspect, a model updating device is provided, which includes a display module, a processing module and an input module, wherein the display module is used to display a first processing result, which includes a first operation analysis result of the network system obtained by processing the system data of the network system according to the model; the input module is used to receive a data input operation; the processing module is used to respond to the knowledge data input operation, and obtain target knowledge data obtained by format conversion of the knowledge data, wherein the knowledge data includes knowledge data input by the user and related to the processing result of the model; the input module is also used to receive a model update instruction; the processing module is also used to respond to the model update instruction, use the target knowledge data to update the model, and obtain an updated model; the processing module is also used to process the system data of the network system according to the updated model to obtain a second processing result, and the second processing result includes a second operation analysis result of the network system; the display module is also used to display the second processing result.

[0030] In a possible implementation, the processing module is further configured to: receive feedback operation data, where the feedback operation data includes a first processing result and a feedback identifier, and the feedback identifier is configured to indicate a method for updating the model based on the first processing result.

[0031] In a possible implementation, the processing module is further configured to: receive a trigger operation of a user on a feedback control displayed on a computer device, and the computer device receives feedback operation data.

[0032] In another possible implementation, the processing module is further configured to: perform binary format conversion on the feedback operation data to obtain first knowledge data included in the target knowledge data.

[0033] In another possible implementation, the processing module is further configured to receive experience text input by a user.

[0034] In another possible implementation, the processing module is further configured to: perform character string format conversion on the experience text to obtain second knowledge data included in the target knowledge data.

[0035] In another possible implementation, the processing module is also used to: receive the text extraction intention input by the user; convert the experience text into a string format to obtain the second knowledge data included in the target knowledge data, including: extracting text information from the experience text according to the text extraction intention to obtain the target text; converting the target text into a string format to obtain the second knowledge data.

[0036] In another possible implementation, the model update instruction includes the weight of the target knowledge data, and the processing module is further configured to adjust the parameters of the model using the target knowledge data, wherein the adjustment range of the parameters is related to the weight of the target knowledge data.

[0037] In another possible implementation, the size of the weight is related to the number of times the user performs feedback operations on the first processing result.

[0038] In another possible implementation, the display module is further configured to display a prompt indicating that acquisition of the target knowledge data is complete.

[0039] In another possible implementation, the display module is further configured to display target knowledge data.

[0040] In a third aspect, a computer device is provided, comprising a processor, a memory, and a display, wherein the memory is used to store instructions, the processor is used to execute the instructions stored in the memory so that the computer device performs the model updating method as described in the first aspect, and the display is used to implement the display steps of the computer device during the execution of the method.

[0041] In a fourth aspect, a computer program product comprising instructions is provided. When the instructions are executed by a computer device, the computer device executes the model updating method as described in the first aspect.

[0042] In a fifth aspect, a computer-readable storage medium is provided, which includes computer program instructions. When the computer program instructions are executed by a computer device, the computer device executes the model updating method as described in the first aspect.

[0043] Based on the implementation methods provided in the above aspects, this application can also be further combined to provide more implementation methods.

[0044] The following description includes more details about the implementation methods provided by the above aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] FIG1 is a schematic diagram of the data processing flow of the model provided in an embodiment of the present application;

[0046] FIG2 is a schematic diagram of a data processing flow of a model updating system provided in an embodiment of the present application;

[0047] FIG3 is a first schematic diagram of an interface of a model updating system provided in an embodiment of the present application;

[0048] FIG4 is a second schematic diagram of an interface of a model updating system provided in an embodiment of the present application;

[0049] FIG5 is a third schematic diagram of an interface of a model updating system provided in an embodiment of the present application;

[0050] FIG6 is a fourth schematic diagram of an interface of a model updating system provided in an embodiment of the present application;

[0051] FIG7 is a fifth schematic diagram of an interface of a model updating system provided in an embodiment of the present application;

[0052] FIG8 is a flow chart of a model updating method according to an embodiment of the present application;

[0053] FIG9 is a second flow chart of the model updating method provided in an embodiment of the present application;

[0054] FIG10 is a third flow chart of the model updating method provided in an embodiment of the present application;

[0055] FIG11 is a schematic diagram of the structure of a model updating device provided in an embodiment of the present application;

[0056] FIG12 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application;

[0057] FIG13 is a schematic diagram of the structure of a computer device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] To facilitate understanding, the terms involved in this application are first introduced.

[0059] Knowledge data refers to data related to knowledge or information, which can include various types of data, such as text, images, audio, video, links, etc., as well as various forms of knowledge, such as facts, concepts, theories, opinions, rules, methods, etc.

[0060] A model is a collection of algorithms and technologies, including neural networks and knowledge graphs. Models can be used to process business data and analyze it to derive business results. For example, in network operations and maintenance, models can be used to analyze operational data such as network system logs to identify anomalies or failures, their causes, and predict future failures.

[0061] The models described in this application may be models used in various technical fields such as power grid operation and maintenance, transportation operation and maintenance, etc. The description mainly uses the models used in network operation and maintenance as examples.

[0062] A network system can refer to a system consisting of network hardware and network software. For example, a system that uses communication equipment and lines to interconnect multiple geographically dispersed and functionally independent computer systems, while using network software to enable resource sharing and information transmission within the network.

[0063] Network operations and maintenance (O&M) refers to the operation and maintenance of network systems, with the goal of ensuring stable operation and efficient management of network services. Network O&M typically involves a range of technical tasks and management responsibilities, such as identifying anomalies or failures, analyzing their causes, and predicting future failures.

[0064] The network operations field faces ever-increasing network scale, complex business demands, and a rapidly changing network environment. Manual management methods are unable to meet these challenges. Therefore, a collection of algorithms and technologies, such as neural networks and knowledge graphs, can be designed specifically for network operations systems. These models, combined with deep data analysis and learning, can improve network operations efficiency in multiple ways. For example, models can monitor network data flows in real time, analyze abnormal patterns, and predict potential failures, thereby improving network availability and stability. Another example is that models can analyze historical data and real-time monitoring results to recommend optimal network configurations, troubleshooting solutions, and performance optimization strategies, providing decision support and enabling operations personnel to more intelligently respond to various scenarios. Furthermore, models can automate network configuration, fault diagnosis, and repair, reducing the workload of operations personnel and improving management efficiency.

[0065] Models are typically designed, developed, and trained by product providers to meet common business requirements. Providers then provide the models to users, who can deploy them in their business systems, such as network systems. Furthermore, model updates are necessary to adapt to changing business needs and improve model quality, such as accuracy.

[0066] One update method is for the model provider to regularly adjust the model offline, such as adjusting the model structure or retraining the model. The updated model is then released, and users can then choose to obtain the new model to update. With this method, the provider typically also provides user training along with the release of the new model to ensure that users understand the new model's features and usage.

[0067] Another update method is for users to regularly communicate and provide feedback to the model provider. For example, users can provide feedback on the business model (such as user experience and expectations for model performance) to the provider through an interactive interface or other channels. Another example is face-to-face communication between users and providers, so that the provider can understand the user's business needs, special scenarios, and expectations for the model. After the provider understands the user's business needs and expectations for the model, it can manually adjust the model used by the user, such as improving the algorithm and adjusting the weights, to improve the model's performance and adaptability to meet the user's business needs and expectations.

[0068] As can be seen, all of the above existing model update methods rely on the provider, resulting in a long and cumbersome model update cycle. This makes it impossible to instantly update the user's model to meet changing business needs and personalized user expectations. Furthermore, these update methods require the provider to perform model maintenance, which consumes a significant amount of the provider's manpower and time. Furthermore, users also need to spend time and resources communicating and coordinating with the provider, increasing the user's operational burden and reducing efficiency.

[0069] In addition, in the first approach, model version updates are performed at a fixed cycle, and the updated model still meets general business needs. This also makes it impossible to adjust the model in real time to meet the specific needs of users, resulting in poor results when users use the model in specific scenarios that differ from general scenarios. In addition, in the second approach, during the communication and feedback process between users and providers, information may be passed layer by layer, resulting in information loss. This also prevents model deficiencies and the target tuning direction from being clearly and accurately fed back to the model tuner. This may lead to misleading the model tuning direction and affect the improvement of model performance. In addition, when model tuners manually adjust the model, there is uncertainty in subjective judgment, the tuning results are unstable, and it is difficult to quickly and accurately determine the optimal direction for model tuning.

[0070] To address the long and cumbersome model update cycle, the inability to instantly update the user-side model to meet changing business needs and user expectations, and the difficulty for users to directly participate in the model tuning process, which limits the user's flexibility in business processes and the accuracy of model tuning, the present application provides a model tuning method and system that provides intelligent human-computer interaction functions, including a computer device that displays a first processing result obtained by processing system data of a network system based on a model. After viewing the first processing result, the user can input knowledge data based on the deficiencies of the processing result or expectations for the model. The computer device then obtains target knowledge data obtained by formatting the knowledge data. The computer device then responds to the model update instruction and uses the target knowledge data to update the model to obtain an updated model. Finally, the computer device processes the system data of the network system based on the updated model to obtain a second processing result, which is also displayed. This allows users to directly participate in the model tuning process and instantly update the user-side model to meet changing business needs and user expectations, thereby improving user flexibility in business processes and the accuracy of model tuning. The entire operation process is also more intelligent, simple and efficient, and the user can intuitively see the changes in the processing results before and after the model update, improving the user experience.

[0071] Figure 1 is a schematic diagram of the data processing flow of the model provided in an embodiment of the present application. As shown in Figure 1, in a network operation and maintenance system, input data (for example, input data including network system logs, network configuration, device parameters, and other system data) can be input into the model 120 for processing to obtain intermediate processing results, and then various downstream business algorithms are used to analyze the intermediate processing results to obtain corresponding business processing results.

[0072] As an example only, model 120 may include various algorithms or models such as neural networks, large language models, knowledge graphs, etc. The intermediate processing results may include analysis results of each log in the input data, such as probability values ​​0.009, 0.01, 0.06...0.2, 0.1, etc. The probability value of a log may represent the probability that the log has an anomaly. The business processing results may include anomaly identification results, root cause analysis results, fault prediction results, repair suggestion results, etc. Among them, the anomaly identification results can be obtained by analyzing the intermediate processing results according to various feasible algorithms such as probability value statistical methods, the root cause analysis results can be obtained by analyzing the intermediate processing results according to various feasible algorithms such as machine learning models, the fault prediction results can be obtained by analyzing the intermediate processing results according to various feasible algorithms such as machine learning models, and the repair suggestion results can be obtained by analyzing the intermediate processing results according to various feasible algorithms such as machine learning models.

[0073] It should be noted that FIG1 is only an embodiment of the data processing flow of a model. In addition to the method shown in FIG1 , the input data can be processed by the model 120 to directly obtain the business processing results, or other feasible methods can be used to process the input data according to the model 120 to obtain the business processing results.

[0074] Figure 2 is a schematic diagram of the data processing flow of the model update system provided in an embodiment of the present application. A provider may provide a model update system to enable local model updates for users. This system may be provided to users in the form of a software product or other feasible product. The model update system may be implemented on a computer device. The computer device may include a single computer device or a computer cluster consisting of multiple computer devices.

[0075] As shown in Figure 2, continuing with the model data processing flow of Figure 1 as an example, through the model update system, a computer device can display the business processing results obtained based on the original model 120. After viewing the business processing results, the user can perform a knowledge data input operation 210 through the system based on their satisfaction with the processing results and their expectations for the processing results, so that the computer device obtains knowledge data related to the model processing results. Furthermore, the computer device can convert the format of the knowledge data to obtain target knowledge data. The target knowledge data can be data that the computer device can recognize and process. Furthermore, after obtaining the target knowledge data, the user can issue a model update command through the system. The computer device responds to the model update command and uses the target knowledge data to update the model 220. After completing the model update operation, the updated model can be used to perform data processing tasks. For example, the updated model can be used to process input data (which can be the original model input data or new input data) to obtain new business processing results. The computer device can also display the new business processing results, allowing the user to intuitively see the changes in the business processing results.

[0076] It should be noted that the model update system may include a processing device and a display device connected to the processing device. The display function and human-computer interaction function of the model update system (such as user data input operations and instruction issuance operations) can be implemented through a human-computer interaction interface, and the human-computer interaction interface can be presented to the user through the display device. The model update system may also include an input device connected to the processing device, such as a mouse, keyboard, or touch screen display capable of user input, and the user can use the input device to input data, issue instructions, etc.

[0077] The following is a detailed introduction to the model update system in conjunction with a schematic diagram of the model update system interface.

[0078] FIG3 is a first schematic diagram of the interface of the model update system provided in an embodiment of the present application. As shown in FIG3 , a first display interface 310 is a human-computer interaction interface of the model update system. The first display interface 310 displays the business processing results obtained based on the original model 120, such as the abnormality identification result of the network system, which is referred to as the first processing result in this application. The abnormality identification result may include information for each log, such as the information of Log 1 and Log 2, and may also include the abnormality identification result corresponding to each log, such as Fault Summary 1 of Log 1 and Fault Summary 2 of Log 2.

[0079] In certain embodiments, in the first display interface 310, a functional module such as the first feedback control-positive feedback 312 and the second feedback control-negative feedback 314 for the user to perform feedback operation on the abnormal identification result may also be included. Wherein, the positive feedback 312 control can be used for when the user triggers the feedback control on the interface (such as clicking the feedback control), and the feedback mark of the user to the abnormal identification result is positive feedback (such as recognition / positive, etc.), and the negative feedback 314 control can be used for when the user clicks the module on the interface, and the feedback mark of the user to the abnormal identification result is negative feedback (such as disapproval / negative, etc.). The user clicks the feedback control and can be realized by a mouse or other feasible methods. In addition, except by the means of the feedback control, a feedback data input box may also be included in the first display interface 310, and the user can also feed back the abnormal identification result by inputting feedback information such as positive feedback / negative feedback or other feasible methods in the feedback data input box.

[0080] In some embodiments, the user can perform feedback operations on the exception identification results corresponding to each log, so that the computer device can obtain the user's feedback operation data on the exception identification results corresponding to each log (including the exception processing result, i.e., the first processing result and the feedback identifier, such as positive feedback / negative feedback).

[0081] After receiving the user's feedback operation data on the first processing result, the computer device can convert the feedback operation data into a format to obtain first knowledge data for subsequent model updates. For more information on the format conversion of the feedback operation data, please refer to Figure 8 and its related description.

[0082] Figure 4 is a second schematic diagram of the interface of the model update system provided in an embodiment of the present application. As shown in Figure 4, a second display interface 410 is a human-computer interaction interface of the model update system. The second display interface 410 may include an input box 412 for experience text, which can be used by the user to enter experience text for network operation and maintenance.

[0083] In some embodiments, the user can manually input experience text through the experience text input box 412, for example, input keywords / text related to network system events, abnormal conditions associated with events, causes associated with abnormal conditions, etc. The computer device can perform format conversion after obtaining the input text to obtain second knowledge data for subsequent model updates. In some embodiments, the experience text input box 412 can provide a text input template, which includes sub-modules with corresponding keywords or texts for network system events, abnormal conditions associated with events, causes associated with abnormal conditions, etc. The user only needs to enter the corresponding category of keywords or text in each sub-module according to the template.

[0084] In some embodiments, the user can upload the experience text document through the experience text input box 412. After obtaining the document, the computer device can extract text information from the experience text and convert the extracted text into a format to obtain second knowledge data for subsequent model updates.

[0085] For more information about format conversion of experience text, please refer to Figure 8 and its related description.

[0086] In some embodiments, the second display interface 410 may further include an input box 414 for text extraction intention, which can be used by the user to input the text extraction intention. The user provides a text extraction intention, such as "alarm operation and maintenance", and then the computer device can perform targeted information extraction on the experience text through the text extraction intention to obtain text related to "alarm operation and maintenance", and then convert the text format related to "alarm operation and maintenance" for model update. For more information about the computer device performing targeted information extraction on the experience text through text extraction intention, please refer to Figure 8 and its related description.

[0087] FIG5 is a third schematic diagram of the interface of the model updating system provided in an embodiment of the present application. As shown in FIG5 , the third display interface 510 is a human-computer interaction interface of the model updating system. After obtaining the target knowledge data through the first display interface 310 and / or the second display interface 410, the third display interface 510 may display a data acquisition completion prompt 512 and may also display the target knowledge data (which may include the first knowledge data and / or the second knowledge data).

[0088] After the user sees the third display interface 510 , he or she knows that the target knowledge data has been acquired and can proceed with subsequent model updates, thus entering the interface shown in FIG. 6 .

[0089] Figure 6 is a fourth interface diagram of the model update system provided in an embodiment of the present application. As shown in Figure 6 , a fourth display interface 610 is a human-computer interaction interface of the model update system. Fourth display interface 610 can be used by a user to issue model update instructions. For example, fourth display interface 610 may include a model update instruction issuance control 612. When a user clicks this control, model update instruction issuance control 612 can be used to issue a model update instruction to a processing device of a computer device.

[0090] In some embodiments, the model update instruction may include a weight of the target knowledge data, which is related to the model parameter adjustment range during the model update. The fourth display interface 610 may also display the weight of each target knowledge data. In some embodiments, the fourth display interface 610 may also be used by the user to modify or set the weight of the target knowledge data. For more details about the weight of the target knowledge data, please refer to Figure 8 and its related description.

[0091] Figure 7 is a fifth interface diagram of the model update system provided in an embodiment of the present application. As shown in Figure 7, fifth display interface 710 is a human-computer interaction interface of the model update system. Fifth display interface 710 can be used to display new business processing results obtained based on the updated model, such as new anomaly identification results of the network system, referred to herein as the second processing result. Because the model has been customized and updated according to user expectations, the second processing result can be more satisfactory to the user, meet user expectations, or satisfy new business needs.

[0092] In some embodiments, the fifth display interface 710 has similar functions to the first display interface 310 , and may also include a feedback control for the user to provide feedback on the abnormality recognition result.

[0093] In some embodiments, after seeing the fifth display interface 710 , the user can also choose whether to perform a new round of model update process according to needs.

[0094] Figure 8 is a flow chart of a model updating method according to an embodiment of the present invention. The model updating method shown in Figure 8 can be implemented by the model updating system described in Figure 2. As shown in Figure 8, the method includes the following steps.

[0095] Step 810: Display the first processing result.

[0096] The first processing result includes a business processing result obtained by processing data according to the current model (eg, original model 120). As an example only, the first processing result includes a first operation analysis result of the network system obtained by processing system data of the network system according to the model.

[0097] Step 820: In response to the knowledge data input operation, obtain target knowledge data.

[0098] The knowledge data includes knowledge data related to the model processing results input by the user, and the knowledge data can be used to guide the model to be adjusted.

[0099] The target knowledge data includes data obtained by format conversion of the knowledge data, and the target knowledge data can be recognized and processed by a computer device.

[0100] In some embodiments, the user hopes that the processing result obtained according to the model can meet his or her own expectations. The user knowledge data input operation may include the user's feedback operation on the first processing result, that is, the knowledge data may include the user's feedback operation data on the first processing result, and the feedback operation data includes the first processing result and a feedback identifier (such as positive feedback / negative feedback). The user's feedback identifier on the first processing result can guide the model adjustment so that the processing result obtained according to the model meets the user's expectations or satisfies the user. For more information about the user's feedback operation on the first processing result, please refer to Figure 3 and its related description.

[0101] In some embodiments, when the knowledge data includes user feedback operation data regarding a first processing result, the computer device may convert the feedback operation data regarding the first processing result into binary format, such as into binary data, to obtain the first knowledge data, which may then serve as the target knowledge data. For example, the computer device may convert positive feedback regarding the first processing result into data "1" and negative feedback regarding the first processing result into data "0."

[0102] In some embodiments, the user hopes that the processing results obtained according to the model can meet some specific business needs, or can be more accurate. The knowledge data input operation can include the experience text entered by the user, that is, the knowledge data can include the experience text entered by the user (for example, a document of the experience text uploaded by the user, or one or more keywords / texts related to the experience directly entered by the user). The experience text can guide the model adjustment so that the processing results obtained according to the model are consistent with the experience knowledge. For more information about the operation of the user entering the experience text, please refer to Figure 4 and its related description.

[0103] In some embodiments, when the knowledge data includes user-entered experience text, the computer device may convert the experience text into a string format, such as into data of a String type, to obtain second knowledge data, which may then serve as the target knowledge data. For example, the computer device may convert the user-entered text into a string consisting of one or more keywords.

[0104] In some embodiments, when the user-input experience text includes documents (such as network operation and maintenance support operation guides, etc.), which may contain a large amount of text content, the computer device can extract text information from the document to obtain the required one or more keywords (referred to as target text in this application), such as keywords related to network operation and maintenance.

[0105] In some embodiments, the user can also input a text extraction intention such as "alarm operation and maintenance", and the computer device can extract text information from the experience text according to the text extraction intention to obtain a text that meets the text extraction intention or is related to the text extraction intention, namely the target text. Taking the example where the text extraction intention includes "alarm operation and maintenance" and the document of the experience text includes a network operation and maintenance support operation guide (including knowledge data on multiple operation and maintenance directions or aspects in network operation and maintenance), the computer device can input both "alarm operation and maintenance" and the network operation and maintenance support operation guide documents into the language model, and perform content mining and text information extraction on the documents through the language model to obtain multiple keywords related to "alarm operation and maintenance" (such as network system events related to alarm operation and maintenance, event-related faults, fault-related causes, etc.).

[0106] In some embodiments, the language model may include various models or algorithms such as Transformer that can implement text analysis and text information extraction.

[0107] In some embodiments, a language model can be trained based on text data related to network system operation and maintenance, so that the language model can extract text information related to network operation and maintenance from text. Language model training methods can include various feasible training methods, such as adjusting language model parameters based on a loss function (e.g., a loss function determined based on the correlation between the text extraction results output by the language model and network operation and maintenance).

[0108] In some embodiments, the language model can also be trained in combination with the text extraction intention so that the language model can extract text related to the text extraction intention in a targeted manner. For example, the text extraction intention and the text sample are input into the language model together, and the language model outputs the text extraction result. The value of the loss function (such as the loss function determined based on the correlation between the text extraction result output by the language model and the text extraction intention) is determined based on whether the text extraction result is associated with the text extraction intention, such as alarm operation and maintenance in network operation and maintenance, and the model parameters of the language model are adjusted to minimize the value of the loss function, thereby enabling the language model to achieve targeted extraction of text related to the text extraction intention.

[0109] In some embodiments, the user can both provide feedback on the first processing result and input experience text, so that the computer device can obtain the first knowledge data and the second knowledge data, and subsequently use the first knowledge data and the second knowledge data together for model updating.

[0110] In some embodiments, the user may only provide feedback on the first processing result, so that the computer device obtains the first knowledge data and subsequently uses the first knowledge data to update the model, such as the model update method shown in Figure 9. In some embodiments, the user may also only input experience text, so that the computer device obtains the second knowledge data and subsequently uses the second knowledge data to update the model, such as the model update method shown in Figure 10.

[0111] In some embodiments, after obtaining the target knowledge data, the computer device may display a data acquisition completion prompt to let the user know that the data acquisition is complete and that the subsequent process, such as step 830, can be continued.

[0112] In some embodiments, the computer device may further display the target knowledge data after obtaining the target knowledge data so as to present it to the user intuitively.

[0113] Step 830: respond to the model update instruction, use the target knowledge data to update the model, and obtain an updated model.

[0114] The model update instruction can be any instruction form that can instruct the computer device to perform a model update operation. The model update instruction can be issued by a user, and the computer device can perform the model update operation after receiving the instruction.

[0115] The model update operation may include adjusting model parameters based on the target knowledge data through various feasible model update algorithms. These include methods such as SFT (supervised fine-tuning), RLHF (reinforcement learning with human feedback), adjusting the knowledge graph by adding entity relationships, and adjusting the language model through prompts.

[0116] As an example, a computer device can perform model updates by the following method: when the model includes a neural network and the target knowledge data includes first knowledge data (such as Binary type data, positive feedback operation data is converted to data "1", and negative feedback operation data is converted to data "0") and second knowledge data (such as String type data), the first knowledge data is converted into a first key-value pair (kv pair), in which the first processing result is used as the key and the feedback data 0 / 1 is used as the value; and the second knowledge data is converted into a second key-value pair, in which the second knowledge data is used as the key and assigned a value of 1. Then, the two key-value pairs are used to fine-tune the parameters of the model through the SFT (supervised fine-tuning) method.

[0117] As another example, a computer device can update a model by the following method: when the model includes a neural network and the target knowledge data includes first knowledge data (such as Binary type data, positive feedback operation data is converted to data "1", and negative feedback operation data is converted to data "0"), the reward model is trained using the first knowledge data, and then under the RLHF (reinforcement learning based on human feedback) framework, the trained reward model is used to fine-tune the parameters of the neural network through the PPO (proximal policy optimization) method.

[0118] As another example, a computer device may perform a model update using the following method: when the model includes a neural network and the target knowledge data includes second knowledge data (such as string type data), the second knowledge data is converted into a second key-value pair, where the second knowledge data is used as a key and a value of 1 is assigned to the second key-value pair. Then, the second key-value pair is used to fine-tune the parameters of the model using an SFT (supervised fine-tuning) method.

[0119] In some embodiments, the model update instruction may further include the weight of the target knowledge data. When the computer device uses the target knowledge data to adjust the parameters of the model, the adjustment range of the parameters is related to the weight of the target knowledge data. For example, when a target knowledge data is used to adjust the model parameters, the weight corresponding to the target knowledge data is 0.5. When the model parameters are adjusted, the value of the model parameters will be adjusted to 0.5 times the original value to be adjusted. For another example, when a target knowledge data is used to adjust the model parameters, the weight corresponding to the target knowledge data is 2. When the model parameters are adjusted, the value of the model parameters will be adjusted to 2 times the original value to be adjusted.

[0120] The weight of the target knowledge data can be determined according to user operations or set by the user.

[0121] As an example, the weight of the first knowledge data can be determined based on the number of feedback operations performed by the user on the first processing result (for example, the number of times the feedback is clicked). The greater the number of operations, the greater the weight. For example, if the number of operations on a processing result is 1, the weight of the first knowledge data obtained based on the feedback operation is 1. For example, if the number of operations on a processing result is 2, the weight of the first knowledge data obtained based on the feedback operation is 2.

[0122] As another example, the weight of the second knowledge data may be set by the user, such as 0.5, 2, etc.

[0123] Step 840: Obtain a second processing result according to the updated model, and display the second processing result.

[0124] The second processing result includes a new business processing result obtained by processing the data according to the updated model. As an example only, the second processing result includes a second operation analysis result of the network system obtained by processing the system data of the network system according to the updated model.

[0125] Figure 9 is a second flow chart of the model updating method provided in an embodiment of the present application. The model updating method shown in Figure 9 can be implemented by the model updating system described in Figure 2. As shown in Figure 9, the method includes the following steps.

[0126] Step 910: Display the first processing result.

[0127] Step 920: In response to the user's feedback operation on the first processing result, obtain first knowledge data.

[0128] Step 930: respond to the model update instruction, use the first knowledge data to update the model, and obtain an updated model.

[0129] Step 940: Obtain a second processing result according to the updated model, and display the second processing result.

[0130] For details of step 910, see step 810 in FIG. 8 and its related contents. For details of step 920, see step 820 in FIG. 8 and its related contents. For details of step 930, see step 830 in FIG. 8 and its related contents. For details of step 940, see step 840 in FIG. 8 and its related contents.

[0131] Figure 10 is a flow chart of the third embodiment of the model updating method provided in the present application. The model updating method shown in Figure 10 can be implemented by the model updating system described in Figure 2. As shown in Figure 10, the method includes the following steps.

[0132] Step 1010: Display the first processing result.

[0133] Step 1020: In response to the user inputting the experience text, obtain the second knowledge data.

[0134] Step 1030: respond to the model update instruction, use the second knowledge data to update the model, and obtain an updated model.

[0135] Step 1040: Obtain a second processing result according to the updated model, and display the second processing result.

[0136] For details of step 1010, see step 810 in FIG. 8 and its related contents. For details of step 1020, see step 820 in FIG. 8 and its related contents. For details of step 1030, see step 830 in FIG. 8 and its related contents. For details of step 1040, see step 840 in FIG. 8 and its related contents.

[0137] The present application also provides a model updating device, as shown in Figure 11, which includes a display module 1110, a processing module 1120, and an input module 1130. The display module 1110, the processing module 1120, and the input module 1130 can be used to execute the method steps in Figures 8, 9, and 10.

[0138] The display module 1110 is used to display a first processing result, which includes a first operation analysis result of the network system obtained by processing the system data of the network system according to the model. The input module 1130 is used to receive a knowledge data input operation. The processing module 1120 is used to respond to the knowledge data input operation and obtain target knowledge data obtained by formatting the knowledge data, wherein the knowledge data includes knowledge data related to the model processing result. The input module 1130 is also used to receive a model update instruction. The processing module 1120 is also used to respond to the model update instruction, update the model using the target knowledge data, and obtain an updated model. In addition, the processing module 1120 is also used to process the system data of the network system according to the updated model to obtain a second processing result, and the second processing result includes a second operation analysis result of the network system. The display module 1110 is also used to display the second processing result.

[0139] Optionally, the processing module 1120 is further configured to: receive feedback operation data, where the feedback operation data includes a first processing result and a feedback identifier, and the feedback identifier is used to indicate a method for updating the model based on the first processing result.

[0140] Optionally, the input module 1130 is further configured to receive a triggering operation of a user on a feedback control displayed on the computer device.

[0141] Optionally, the processing module 1120 is further configured to: perform binary format conversion on the feedback operation data to obtain first knowledge data included in the target knowledge data.

[0142] Optionally, the input module 1130 is further configured to receive experience text input by a user.

[0143] Optionally, the processing module 1120 is further configured to: perform character string format conversion on the experience text to obtain second knowledge data included in the target knowledge data.

[0144] Optionally, the input module 1130 is further used to receive a text extraction intention input by a user.

[0145] Optionally, the processing module 1120 is also used to: convert the experience text into a string format to obtain the second knowledge data included in the target knowledge data, including: extracting text information from the experience text according to the text extraction intention to obtain the target text; converting the target text into a string format to obtain the second knowledge data.

[0146] Optionally, the model update instruction includes the weight of the target knowledge data, and the processing module 1120 is further configured to: use the target knowledge data to adjust the parameters of the model, wherein the adjustment range of the parameters is related to the weight of the target knowledge data.

[0147] Optionally, the size of the weight is related to the number of times the user performs feedback operations on the first processing result.

[0148] Optionally, the display module 1110 is further configured to display a prompt indicating that acquisition of the target knowledge data is complete.

[0149] Optionally, the display module 1110 is also used to display target knowledge data.

[0150] For more details on the method steps implemented by the display module 1110 , the processing module 1120 , and the input module 1130 , please refer to FIG. 2 to FIG. 10 and their related descriptions.

[0151] Optionally, the display module 1110 , the processing module 1120 , and the input module 1130 may each include multiple sub-modules, and the multiple sub-modules may be deployed separately to implement part of the functions of the corresponding modules.

[0152] The apparatus can be implemented by software or hardware. As an example, the implementation of the model updating apparatus 1100 is described below.

[0153] As an example of a software functional unit, a module may include a code running on a computing instance. The computing instance may be at least one of a physical host (computing device), a virtual machine, a container, and other computing devices. Furthermore, the computing device may be one or more. For example, the model updating device may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the application may be distributed in the same region or in different regions. The multiple hosts / virtual machines / containers used to run the code may be distributed in the same AZ or in different AZs, and each AZ includes one data center or multiple data centers with close geographical locations. Generally, a region may include multiple AZs.

[0154] Similarly, the multiple hosts / virtual machines / containers running the code can be distributed within the same VPC or across multiple VPCs. Typically, a VPC is located within a region. Cross-region communication between two VPCs within the same region, or between VPCs in different regions, requires a communication gateway within each VPC to interconnect the VPCs.

[0155] As an example of a hardware functional unit, the model updating device 1100 may include at least one computing device, such as a server. Alternatively, the model updating device may be implemented using an ASIC or a PLD. The PLD may be implemented using a CPLD, FPGA, GAL, or any combination thereof.

[0156] The multiple computing devices included in the model update apparatus 1100 can be distributed in the same region or in different regions. The multiple computing devices included in the model update apparatus can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the YY apparatus can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of servers, ASICs, PLDs, CPLDs, FPGAs, GALs, and other computing devices.

[0157] The present application also provides a computing device 1200. As shown in FIG12 , the computing device 1200 includes a bus 1202, a processor 1204, a memory 1206, and a communication interface 1208. The processor 1204, the memory 1206, and the communication interface 1208 communicate with each other via the bus 1202. The computing device 1200 may be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 1200. The processor 1204 of the computing device 1200 may be connected to a display device 1210 and an input device 1220 via the communication interface 1208.

[0158] Bus 1202 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, among others. Buses may be classified as address buses, data buses, control buses, and the like. For ease of illustration, FIG12 shows a single bus line, but this does not imply a single bus or type of bus. Bus 1202 may include a path for transmitting information between various components of computer device 1200 (e.g., memory 1206, processor 1204, and communication interface 1208).

[0159] The processor 1204 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0160] The memory 1206 may include volatile memory, such as random access memory (RAM). The processor 1204 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0161] Memory 1206 stores executable program code, and processor 1204 executes the executable program code to implement the functions of the aforementioned display module 1110, processing module 1120, and input module 1130, respectively, thereby implementing the model updating method provided in the embodiments of the present application, such as the model updating method provided in Figures 8, 9, and 10. In other words, memory 1206 stores instructions for executing the model updating method provided in the embodiments of the present application.

[0162] The communication interface 1208 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computer device 1200 and other devices or a communication network.

[0163] The display device 1210 may include various display devices capable of realizing a display function, such as a plasma display and a liquid crystal display.

[0164] The input device 1220 may include various input devices such as a keyboard, a mouse, and a touch screen that enables user input.

[0165] The present application also provides a computer device cluster 1300. As shown in FIG13 , the computer device cluster includes at least one computer device 1200. The memory 1206 in one or more computer devices 1200 in the computer device cluster may store the same instructions for executing the model updating method provided in the embodiments of the present application, such as the model updating method provided in FIG8 , FIG9 , and FIG10 .

[0166] In some possible implementations, the memory 1206 of one or more computer devices 1200 in the computer device cluster may also respectively store partial instructions for executing the model updating method provided in the embodiments of the present application, such as instructions for executing partial steps of the model updating method provided in Figures 8, 9, and 10. In other words, the combination of one or more computer devices 1200 can jointly execute instructions for executing the model updating method provided in the embodiments of the present application.

[0167] It should be noted that the memory 1206 in different computer devices 1200 in the computer device cluster can store different instructions, each for executing a portion of the functions of the model updating apparatus. In other words, the instructions stored in the memory 1206 in different computer devices 1200 can implement the functions of one or more of the aforementioned display module 1110, processing module 1120, and input module 1130.

[0168] In some possible implementations, one or more computer devices in the computer device cluster may be connected via a network, which may be a wide area network or a local area network.

[0169] In some possible implementations, the memory 1206 of one or more computer devices 1200 in the computer device cluster may also respectively store partial instructions for executing the model updating method provided in the embodiments of the present application, such as instructions for executing partial steps of the model updating method provided in Figures 8, 9, and 10. In other words, the combination of one or more computer devices 1200 can jointly execute instructions for executing the model updating method provided in the embodiments of the present application.

[0170] The present application also provides a computer program product comprising instructions. The computer program product may be software or a program product comprising instructions that can be run on a computer device or stored in any available medium. When the computer program product is run on at least one computer device, the at least one computer device executes the model update method provided in the present application, such as instructions for executing the model update method provided in Figures 8, 9, and 10.

[0171] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computer device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state drive). The computer-readable storage medium includes instructions that instruct the computer device to execute the model update method provided in the embodiment of the present application, such as instructions for executing the model update method provided in Figures 8, 9, and 10.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the protection scope of the technical solutions of the various embodiments of the present invention.

[0173] The terms "first", "second", "third" and "fourth" in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects rather than to limit a specific order.

[0174] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

Claims

1. A model updating method, characterized in that: The method is executed by a computer device, and includes: displaying a first processing result, wherein the first processing result includes a first operation analysis result of the network system obtained by processing system data of the network system according to the model; In response to the knowledge data input operation, target knowledge data is obtained, wherein the knowledge data includes knowledge data input by the user and related to the processing result of the model, and the target knowledge data includes data after format conversion of the knowledge data; In response to a model update instruction, the model is updated using the target knowledge data to obtain an updated model; The system data of the network system is processed according to the updated model to obtain a second processing result, and the second processing result is displayed. The second processing result includes a second operation analysis result of the network system.

2. The method according to claim 1, characterized in that The response knowledge data input operation includes: Feedback operation data is received, where the feedback operation data includes the first processing result and a feedback identifier, where the feedback identifier is used to indicate a manner of updating the model based on the first processing result.

3. The method according to claim 2, characterized in that The response knowledge data input operation includes: A trigger operation of a user on a feedback control displayed on the computer device is received, and the computer device receives the feedback operation data.

4. The method according to claim 2 or 3, characterized in that The response knowledge data input operation to obtain target knowledge data includes: The feedback operation data is converted into a binary format to obtain the first knowledge data included in the target knowledge data.

5. The method according to any one of claims 1 to 4, characterized in that The response knowledge data input operation includes: Receives user input experience text.

6. The method according to claim 5, characterized in that The response knowledge data input operation to obtain target knowledge data includes: The experience text is converted into a character string format to obtain second knowledge data included in the target knowledge data.

7. The method according to claim 6, characterized in that The response knowledge data input operation further includes: Receive text extraction intention from user input; The converting the experience text into a character string format to obtain the second knowledge data included in the target knowledge data includes: Extracting text information from the experience text according to the text extraction intention to obtain a target text; The target text is converted into a character string format to obtain the second knowledge data.

8. The method according to any one of claims 1 to 7, characterized in that The model updating instruction includes the weight of the target knowledge data, and updating the model using the target knowledge data includes: The target knowledge data is used to adjust the parameters of the model, wherein the adjustment amplitude of the parameters is related to the weight of the target knowledge data.

9. The method according to claim 8, characterized in that The size of the weight is related to the number of times the user performs feedback operations on the first processing result.

10. The method according to any one of claims 1 to 9, characterized in that The method further includes: displaying a prompt indicating that the target knowledge data acquisition is completed.

11. The method according to any one of claims 1 to 10, characterized in that The method further includes displaying the target knowledge data.

12. A model updating device, characterized in that: It includes a display module, a processing module and an input module; The display module is used to display a first processing result, wherein the first processing result includes a first operation analysis result of the network system obtained by processing system data of the network system according to the model; The input module is used to receive data input operations; The processing module is used to respond to the knowledge data input operation to obtain target knowledge data, wherein the knowledge data includes knowledge data related to the model processing result, and the target knowledge data includes data after the format of the knowledge data is converted; The input module is also used to receive model update instructions; The processing module is further configured to respond to a model update instruction and update the model using the target knowledge data to obtain an updated model; The processing module is further configured to process the system data of the network system according to the updated model to obtain a second processing result, wherein the second processing result includes a second operation analysis result of the network system; The display module is further configured to display the second processing result.

13. A computer device, characterized in that: The computer device includes a processor, a memory and a display; The memory is used to store instructions, the processor is used to execute the instructions stored in the memory, so that the computer device performs the method according to any one of claims 1 to 11, and the display is used to implement the display step of the computer device during the execution of the method.

14. A computer program product comprising instructions, characterized in that When the instructions are executed by a computer device, the computer device is caused to perform the method according to any one of claims 1 to 11.

15. A computer-readable storage medium, characterized in that The method comprises computer program instructions, and when the computer program instructions are executed by a computer device, the computer device performs the method according to any one of claims 1 to 11.

Images