5GC platform intelligent interaction method and device based on intention recognition and storage medium

Abstract

In order to solve the problem that a traditional 5GC platform interaction mode mainly depends on manual instruction input or selection operation of a user, the invention discloses a 5GC platform intelligent interaction method and device based on intention recognition and a storage medium, and belongs to the technical field related to core network element management in 5G mobile communication. The method comprises the steps of collecting and preprocessing user interaction behavior data; identifying a user intention based on the preprocessed user interaction behavior data; implementing the user intention according to the category of the user intention; and sending the user intention implementation result to the client for presentation. According to the method, the operation process is greatly simplified, the decision making process is accelerated, meanwhile, the operation error rate is reduced, and 5GC platform interaction management is more intelligent and efficient.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of core network element management in 5G mobile communication, and specifically relates to a 5GC platform intelligent interaction method based on intent recognition, a device and a storage medium. BACKGROUND

[0002] 5GC: 5G Core, which is the core network part of the 5G mobile communication network.

[0003] Intent Recognition: is an important task in natural language processing (NLP), which aims to determine the intent or purpose expressed in the user's input sentence. Semantic understanding of user's speech in order to better answer user's questions or provide relevant services.

[0004] The traditional 5GC platform interaction method mainly relies on user manual input instruction or selection operation, which limits the convenience and efficiency of user experience to some extent. Users need to be familiar with complex operation interface and interaction process to complete the required business operation, which brings certain inconvenience to user experience. With the development of artificial intelligence and natural language processing technology, users have higher expectations for more intelligent and natural interaction methods. SUMMARY

[0005] The application provides a 5GC platform intelligent interaction method based on intent recognition, a device and a storage medium to solve the problems in the prior art. The application aims to realize intelligent interaction between 5GC platform and users through automatic recognition of user intent, improve user experience and enhance the interaction convenience of 5GC platform.

[0006] To achieve the above purpose, the application adopts the following technical solutions:

[0007] A 5GC platform intelligent interaction method based on intent recognition, comprising the following steps:

[0008] Collecting user interaction behavior data and preprocessing;

[0009] Recognizing user intent based on preprocessed user interaction behavior data;

[0010] Implementing user intent according to the category of user intent;

[0011] Sending user intent implementation results to the client for presentation.

[0012] To optimize the above technical solutions, the specific measures taken include:

[0013] Further, the preprocessing includes:

[0014] Extracting key information from user interaction behavior data, removing stop words, normalizing sentences, correcting spelling errors, identifying language and structure;

[0015] Deleting elements containing missing values or filling missing values by mean and average; at the same time, special values, outliers and obviously inconsistent values are processed;

[0016] Mean, median, area, histogram and scatter plot analysis of user interaction behavior data features to explore data relationships and discover potential features; numerical and categorical features are processed into numerical form using label encoding or one-hot encoding.

[0017] Further, the user intent recognition based on the preprocessed user interaction behavior data is specifically:

[0018] Extracting features from preprocessed user interaction behavior data to generate feature vectors;

[0019] Using the trained multiple intent recognition models to predict the feature vectors respectively to obtain the intent recognition results;

[0020] Calculating the confidence score of the intent recognition result;

[0021] Combining the confidence score and the decision tree rule to determine the final intent implementation item of each intent recognition model;

[0022] Using voting method and weighted average method to integrate the intent implementation items of multiple intent recognition models.

[0023] Further, the intent recognition model includes decision tree, SVM and random forest; the training method of the intent recognition model is:

[0024] The knowledge graph professional library and the scheme recommendation intent recognition feature library are used as the training set to train the intent recognition model, grid search and cross-validation are used for model comparison and analysis, and the trained multiple intent recognition models are obtained.

[0025] Further, the construction method of the knowledge graph professional library is:

[0026] With the help of expert experience and communication professional knowledge, the knowledge graph is constructed, the entity relationship joint extraction model Casrel is used to realize the extraction of "subject-relation-object" triplets, and the knowledge graph professional library is generated with the help of expert confirmation.

[0027] Further, the construction method of the scheme recommendation intent recognition feature library is:

[0028] Analyze historical data including user operation data, historical logs, historical work orders, operation work orders, menus, and index items, count the proportion of different work orders, retain historical data samples greater than a specified threshold, and label the samples, and balance the proportion of label 0 values and 1 values using a multi-label balancing algorithm;

[0029] Based on the Rakel multi-label classification algorithm and the labeled historical data, the model is trained and iterated to obtain a scheme recommendation intention recognition feature, which is incorporated into a scheme recommendation intention recognition feature library.

[0030] Further, the user intention is classified according to the user intention, and the user intention is specifically:

[0031] If the category of the user intention is a menu opening intention, the menu name is identified from the user interaction sentence, and the 5GC platform is prompted to pop up a menu page;

[0032] If the category of the user intention is an index query intention, a pre-trained vector library and its vectorization tool are loaded, and province, device name, and time slots are identified. When the user input is incomplete, the missing slot information is confirmed according to the pre-set rules, and the slot is filled. A thread pool is used to concurrently execute multiple tasks, including word vector processing, date extraction, province identification, and device information extraction. The results of all tasks are collected, and the extracted task results are further processed, including time range normalization, most similar sentence filtering and scoring. According to the user request type, the corresponding label is mapped, and the related sentences are filtered based on the label. The extracted information is dynamically adjusted according to the user's specific needs, and the user's session history record is updated as necessary. The final processed information is stored in the database for subsequent query or analysis. The return object is constructed according to the processing result. If some key fields are empty or uncertain, the fields that need to be confirmed are marked, and the intention guidance is performed;

[0033] If the category of the user intention is a device query intention, the query intention of the specified device is matched, and the 5GC atomic capability is called to query the manufacturer and the life cycle;

[0034] If the category of the user intention is a root cause intention, the knowledge graph is called, the entity relationship of the user input interaction data is extracted, the knowledge graph root cause is recalled, and the scheme and root cause are output;

[0035] If the category of the user intention is a scheme recommendation intention, the trained model is used to recommend a scheme, the type, task, and execution instruction information are filled in, and the scheme is sorted and labeled;

[0036] If the category of the user's intention is an automatic triggering category intention, the 5GC system listens to the user's page operation event, automatically triggers the related function according to the user operation and the system state, directly returns the user confirmation, and automatically pops up the related prompt item when the system occurs an alarm or the performance, the log file is missing;

[0037] If the category of the user's intention is idle chat and other intentions, a general large model is called for processing.

[0038] Further, the sending of the user intention implementation result to the client for presentation is specifically:

[0039] In a streaming response mode, the intention implementation presentation is performed, the client establishes a connection with the server through the WebSocket protocol to realize bidirectional real-time communication, and the user intention implementation result is pushed to the client in real time.

[0040] The application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the intention recognition-based 5GC platform intelligent interaction method is realized.

[0041] The application also provides a computer readable storage medium storing a computer program, wherein the computer program enables a computer to execute the intention recognition-based 5GC platform intelligent interaction method.

[0042] The application has the following beneficial effects:

[0043] (1) The interaction convenience of the 5GC platform is improved, the user does not need to be familiar with a complex operation interface and an interaction process, and the required business operation can be completed through a natural interaction mode.

[0044] (2) The intelligent level of the 5GC platform is enhanced, the interaction intention of the user can be automatically recognized, and the corresponding business logic can be executed, and the user experience is improved.

[0045] (3) The operation cost and the learning cost of the user are reduced, and the use efficiency of the 5GC platform is improved. DETAILED DESCRIPTION

[0046] Figure 1 The overall framework diagram of the intention recognition-based 5GC platform intelligent interaction method is provided.

[0047] Figure 2 The overall interaction diagram between the client and the server is provided.

[0048] Figure 3 The detailed flowchart of the intention recognition-based 5GC platform intelligent interaction method is provided.

[0049] Figure 4 An example page diagram for a 5GC platform. DETAILED DESCRIPTION

[0050] 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 part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0051] Embodiment one

[0052] The present application proposes a 5GC platform intelligent interaction method based on intent recognition, the overall framework of the method is as shown in Figure 1 , the detailed process is as shown in Figure 3 , including the following steps:

[0053] Collecting user interaction behavior data and preprocessing;

[0054] Recognizing user intent based on the preprocessed user interaction behavior data;

[0055] Implementing user intent according to the category of user intent;

[0056] Sending the user intent implementation result to the client for presentation.

[0057] A. User interaction behavior data collection module.

[0058] 1. Real-time monitoring: Real-time monitoring of user text input and other interaction behaviors on the 5GC platform.

[0059] 2. Data collection: Collect all user interaction behavior data, including but not limited to text input, menu selection, operation instructions, etc.

[0060] B. Preprocessing module.

[0061] 1. Extracting key information: Extracting key information from user interaction data, removing stop words, normalizing sentences, correcting spelling errors, and identifying languages and structures.

[0062] 2. Data cleaning: Deleting elements containing missing values or filling missing values by mean, average, etc.; at the same time, special values, outliers, and obviously inconsistent values are processed.

[0063] 3. Data analysis: Using pandas, matplotlib, factes, etc. to analyze single and double variables such as mean, median, area, histogram, scatter plot, etc. through feature data, explore data relationships, and discover potential features.

[0064] 4. The intention recognition algorithm is processed.

[0065] C. Intention recognition algorithm. Based on user interaction behavior data, using natural language processing, machine learning and other technologies, identify the user's interactive intent. Including menu query and opening, index situation, high-risk instructions, alarm query, device status, etc.

[0066] 1. Historical data collection: Collect historical work orders, historical logs, operation orders, and user operation behavior data.

[0067] 2. Data cleaning, conversion and analysis. Delete elements containing missing values or fill in missing values by mean, average, etc. At the same time, special values, outliers, and obviously inconsistent values are processed. Use pandas, matplotlib, factes and other tools to analyze and explore data relationships by exploring potential features through mean, median, region, histogram, scatter plot, etc. Single and double variable analysis. For numerical and categorical features, use label encoding, one-hot encoding, etc. to process them into numerical form.

[0068] 3. Menu index intention recognition vector library construction. For text features, build a professional dictionary and stop words, and after full-mode word segmentation algorithm processing, use the CBOW and Skip-Gram two architectures of the word2vec word embedding model to learn low-dimensional dense vector representations of text, generate professional word vector library for intention recognition, and realize the intention recognition of simple operations such as menu and page opening. According to the professional word vector library, the vector generation and similarity matching of the user's input interaction data are performed.

[0069] 4. Knowledge graph construction. With the help of expert experience and communication professional knowledge, construct a knowledge graph, based on the entity relationship joint extraction model Casrel, use the communication professional knowledge fine-tuning model to realize the extraction of "subject-relation-object" triplets, and generate a knowledge graph professional library with the help of expert confirmation. Retrieve relevant information from the knowledge graph and perform knowledge fusion. Retrieve root cause analysis features (such as entities and relationships) from the 5GC professional knowledge graph and integrate them into the intention recognition features. Extract entity relationships from the user's input interaction data to recall the root cause of the knowledge graph.

[0070] 5. Scheme recommendation library construction. Collect 5GC professional scheme recommendation features and label them, train and iterate the model based on the Rakel multi-label algorithm, and model and analyze the 31 provinces in China. Integrate the scheme recommendation intention recognition feature library. The specific implementation method is:

[0071] (1) Analyze historical data, take the cut scene, cut content, province, device type data as the main training features, operation before preparation, operation implementation, operation back, operation after inspection, the next day on duty dot scheme prediction.

[0072] (2) Data probability statistics (the same scene in different work order ratio statistics, reserve more than the specified threshold sample)

[0073] (3) Determine the same scene feature needs to be multi-label assigned, determine the label set (such as taking threshold value 0.5, then based on threshold operation before scheme evaluation, system state performance check and configuration file backup for label assignment).

[0074] (4) Multi-label labeling (for example, system state performance check, capacity analysis, change table check, alarm detection, performance detection, account password check, business test preparation, health degree analysis, alarm shielding 8 items are labeled as 1, and the rest of the qualified items are labeled as 0)

[0075] (5) Multi-label balance algorithm

[0076] For the problem of uneven distribution of 0 and 1 for each label, an algorithm is designed to count the proportion of 0 and 1. For the proportion less than the threshold, oversampling is expanded.

[0077] (6) Dot scheme model stacking prediction overall process

[0078] Based on the work order, the proportion of the whole work order is counted, the data is multi-label labeled, and the threshold oversampling algorithm is designed to balance the multi-label classification task imbalance. The LabelPowerset algorithm idea is combined with the decision tree basic classifier for training to obtain M0 and M1 models. Overall, the province is divided by the cut scene, cut content, province, and device type as the features of model M0 for labeling training. The output features of model M0 are combined with the cut scene, cut content, and province as the training features of model M1 to predict the label classification result. At the same time of training, the encoder and parameters are saved.

[0079] 6. Model training. Obtain the knowledge graph professional library and scheme recommendation intention recognition feature library, perform vectorization processing, divide the training set, test set, and validation set, select decision tree, SVM, random forest, etc. for model training, grid search, cross-validation for model comparison and analysis, form intention recognition recommendation model. The intention recognition model is integrated and displayed as:

[0080] (1) Intention recognition process:

[0081] Input data preprocessing: preprocess the interactive data input by the user, including word segmentation, stop word removal, etc.

[0082] Feature extraction: Extract features from pre-processed data to generate vector representation.

[0083] Model prediction: Use the trained intent recognition model to predict the feature vector and obtain the intent recognition result.

[0084] Confidence score: Calculate the confidence score of the model prediction result for subsequent decision-making.

[0085] Decision tree decision: Combine the confidence score and decision tree rules to determine the final intent implementation item.

[0086] (2) Ensemble method:

[0087] Use voting method and weighted average method to integrate the prediction results of multiple models.

[0088] 7、The result of intent implementation, using standard templates, sends messages to the presentation and feedback module.

[0089] D. Intent implementation module.

[0090] 1、Query intent.

[0091] (1) Menu open type: Identify the menu name from the user interaction sentence and inform the 5GC platform to pop up the menu page.

[0092] (2) Index query intent: Construct province, device name, and time slots, and confirm the missing slot information according to the pre-set rules. Specifically, it includes:

[0093] ① Initialization and data preparation:

[0094] First, check the semantics in the request, if the conditions are met, then proceed to the subsequent intent implementation.

[0095] Load the pre-trained vector library and its vectorization tool.

[0096] ② Asynchronous task processing:

[0097] Use thread pool to execute multiple tasks concurrently, including word vector processing, date extraction, province recognition, and device information extraction.

[0098] Collect the results of all tasks and handle possible exceptions.

[0099] ③ Result processing and optimization:

[0100] Further process the extracted information, such as time range normalization, most similar sentence filtering and scoring, etc.

[0101] Map the corresponding label according to the request type, and filter the relevant sentences based on this label.

[0102] ④ Dynamic adjustment and history record update:

[0103] Dynamically adjust the extracted information according to the specific needs of the user, and update the user's session history record if necessary.

[0104] Store the final processed information to the database for subsequent query or analysis.

[0105] ⑤ Return result:

[0106] Construct a return object according to the processing result, and if some key fields are empty or have uncertainty, mark the fields that need to be confirmed and guide the intention.

[0107] (3) Device query intent: match to the query intent of the specified device, call the 5GC atomic ability to query the manufacturer and life cycle.

[0108] (4) Constraint rule: return the menu name directly if the confidence is higher than 99%, and return multiple menu names for user confirmation if the confidence is higher than 40%.

[0109] The specific implementation is: if the identified intent is a query intent, automatically execute the corresponding business logic on the 5GC platform. The query intent mainly consists of three parts: first, menu opening type, for menu correction, completion return, and pop-up menu page for the 5GC platform; second, index query intent, construct province, device name, and time slot. When the user inputs are incomplete, the missing slot information is confirmed according to the pre-set rules; the identification algorithm of the province is designed to return the corresponding province according to the city and province information in the semantics, which covers all provincial administrative regions. The device name slot means to identify the nine-character naming of the device in the sentence or the general device composed of letters and numbers, such as AMF01, and supports multiple devices. The time semantic feature of the sentence is designed with TimeSemExtra time processing algorithm to identify the start and end time. Through a series of text preprocessing and regular matching techniques, combined with context information and specific rules, a variety of rules are involved in the parsing process, including but not limited to: setting specific time units such as year, month, and day, processing ambiguous time (such as "tomorrow" and "the day after tomorrow"), and parsing special time expressions (such as "May 1st"). According to the context, the function of accurately parsing time information from natural language text is realized. It can handle various time table forms, including specific date and time, relative time (such as "tomorrow"), and ambiguous time (such as "afternoon"). At the same time, the algorithm can identify the spatial dimension (nationwide, province, and device) and the time dimension (month, day, hour, and minute) in the semantics. In particular, when the user input and slot matching degree are low, intent guidance is performed, and then the corresponding atomic ability of the 5GC platform is called as a parameter for query, while supporting multiple rounds of dialogue, user device, and province authentication; third, device query intent, matching to the specified device query intent, calling the 5GC atomic ability to query the manufacturer and life cycle. The constraint rules for the above three query intents are: if the confidence is higher than 99%, directly return the menu name, and if the confidence is higher than 40%, return multiple menu names for user confirmation. For explicit intent implementation, user and device authentication is performed.

[0110] 2. Root cause intent. The intent identified as the root cause is called the knowledge graph capability to output the scheme and root cause. The confidence is above 90%, and below 40%, then enter the subsequent intent processing.

[0111] 3. Scheme recommendation intent. The intent identified as the scheme recommendation is predicted using the trained model to fill in the type, task, and execution instruction information, and sorted and labeled to return to the presentation and feedback module. The confidence is above 90% and below 40%, then enter the subsequent intent processing.

[0112] 4. Automatic triggering type. The user performs page operations, and the 5GC system automatically triggers according to the needs, directly returns user confirmation, and automatically pops up related prompt items when the system has alarms or performance, log file missing. The automatic triggering mechanism is:

[0113] Event listening: Listen to user page operation events.

[0114] Conditional judgment: Automatically trigger related functions according to user operations and system state.

[0115] Pop-up prompt: Automatically pop up prompt items when the system has alarms or performance, log file missing.

[0116] 5. Idle chat and other intents, call general large models for bottom processing.

[0117] 6. Historical dialogue record. Record historical dialogue records, dialogue information statistics, and support multiple rounds of dialogue.

[0118] E. Presentation and feedback module.

[0119] 1. Streaming response: Use streaming response to implement intent implementation presentation. Use WebSocket to realize two-way real-time communication. The specific implementation steps are:

[0120] (1) Establish a connection: The client establishes a connection with the server through the WebSocket protocol.

[0121] (2) Send request: The client sends interaction behavior data.

[0122] (3) Receive response: The server receives user interaction behavior data and automatic triggering type data in real time, transfers intent recognition service processing, performs intent distribution, atomic capability call processing intent implementation results, and pushes the results to the client in real time through the WebSocket connection.

[0123] 2. User feedback: Collect user feedback (satisfaction score and improvement suggestions).

[0124] (1) Collect feedback:

[0125] Front-end interface: Design a user feedback interface, including a satisfaction score input box and an improvement suggestion input box.

[0126] Back-end processing: Collect user feedback data through API interface and store it in the database.

[0127] (2) Specific implementation steps:

[0128] Front-end interface design:

[0129] Satisfaction score: Provide star rating or slider rating.

[0130] Improvement suggestion interface: provide an improvement suggestion interface for users to fill in specific suggestions.

[0131] Backend processing:

[0132] API interface: design a RESTful API interface for collecting user feedback data.

[0133] Data storage: store user feedback data into a database and conduct statistical analysis.

[0134] Embodiment two

[0135] The present application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the 5GC platform intelligent interaction method based on intent recognition as described in embodiment one.

[0136] Embodiment three

[0137] The present application provides a computer readable storage medium storing a computer program, wherein the computer program enables a computer to execute the 5GC platform intelligent interaction method based on intent recognition as described in embodiment one.

[0138] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium which can contain or store programs for use by or in conjunction with an instruction execution system, device or apparatus. The computer storage medium can include but is not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, device or apparatus, or any suitable combination of the above. More specific examples of computer storage medium can include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

[0139] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0140] The above are only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical scheme falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled in the art, some improvements and refinements without departing from the principles of the present application shall be considered as the protection scope of the present application.

Claims

1. A 5GC platform intelligent interaction method based on intent recognition, characterized in that, The method comprises the following steps: Collecting user interaction behavior data and preprocessing; Identifying user intent based on preprocessed user interaction behavior data; Implementing user intent according to the category of user intent; Sending user intent implementation results to the client for presentation. 2.The intent recognition based 5GC platform intelligent interaction method according to claim 1, wherein, The preprocessing comprises: Extracting key information from user interaction behavior data, removing stop words, normalizing sentences, correcting spelling errors, identifying language and structure; Deleting elements containing missing values or filling missing values by mean and average; at the same time, special values, outliers and obviously inconsistent values are processed; Analyzing the mean, median, area, histogram and scatter plot of the characteristics of user interaction behavior data, exploring data relationships and discovering potential features; for numerical and categorical features, use label encoding or one-hot encoding to process them into numerical form. 3.The intent recognition based 5GC platform intelligent interaction method of claim 1, wherein, The identification of user intent based on preprocessed user interaction behavior data is specifically: Extracting features from preprocessed user interaction behavior data to generate a feature vector; Using multiple trained intent recognition models to predict the feature vector respectively to obtain intent recognition results; Calculating the confidence score of the intent recognition result; Combining the confidence score and the decision tree rule to determine the final intent implementation item of each intent recognition model; Using voting and weighted average methods to integrate the intent implementation items of multiple intent recognition models. 4.The intent recognition based 5GC platform intelligent interaction method of claim 3, wherein, The intent recognition model comprises a decision tree, an SVM and a random forest; the training method of the intent recognition model is: Using the knowledge graph professional library and the scheme recommendation intent recognition feature library as the training set to train the intent recognition model, using grid search and cross-validation for model comparison and analysis to obtain multiple trained intent recognition models. 5.The intent recognition based 5GC platform intelligent interaction method according to claim 4, characterized in that, The construction method of the knowledge graph professional library is: Constructing a knowledge graph with the help of expert experience and communication professional knowledge, based on the entity relationship joint extraction model Casrel, realizing the extraction of "subject-relation-object" triplets, and generating a knowledge graph professional library with the help of expert confirmation. 6.The intent recognition based 5GC platform intelligent interaction method according to claim 4, wherein, The construction method of the scheme recommendation intent recognition feature library is: Analyzing historical data, including user operation data, historical logs, historical work orders, operation work orders, menus and index items, calculating the proportion of different work orders, retaining historical data samples greater than a specified threshold, and labeling the samples, and using a multi-label balancing algorithm to balance the proportion of label 0 and 1 values; Based on the Rakel multi-label classification algorithm and the labeled historical data, the model is trained and iterated to obtain the scheme recommendation intent recognition features, which are incorporated into the scheme recommendation intent recognition feature library. 7.The intent recognition based 5GC platform intelligent interaction method according to claim 1, wherein, The implementation of user intent according to the category of user intent is specifically: If the category of user intent is a menu opening intent, identify the menu name from the user interaction sentence and inform the 5GC platform to pop up the menu page; If the category of user intent is an index query intent, load the pre-trained vector library and its vectorization tool, identify the province, device name and time slot, and when the user input is incomplete, confirm the missing slot information with the user according to the pre-set rules and fill in the slot. A thread pool is used to execute multiple tasks concurrently, including word vector processing, date extraction, province identification, and device information extraction. The results of all tasks are collected and further processed, including time range normalization, most similar sentence filtering and scoring. According to the user request type, the corresponding label is mapped, and the relevant sentences are filtered based on the label. The extracted information is dynamically adjusted according to the user's specific needs, and the user's session history record is updated if necessary. The final processed information is stored in the database for subsequent query or analysis. According to the processing result, the return object is constructed. If some key fields are empty or uncertain, the fields that need to be confirmed are marked, and the intention guidance is performed. If the category of the user's intention is a device query intention, the specified device query intention is matched, and the 5GC atomic capability is called to query the manufacturer and the life cycle. If the category of the user's intention is a root cause intention, the knowledge graph is called to extract entity relationships from the user's input interaction data, perform knowledge graph root cause recall, and output solutions and root causes. If the category of the user's intention is a solution recommendation intention, the trained model is used to recommend solutions, fill in type, task, and execution instruction information, and sort and label the solutions. If the category of the user's intention is an automatic trigger intention, the 5GC system listens to the user's page operation events, automatically triggers related functions according to the user's operation and system state, directly returns user confirmation, and automatically pops up related prompt items when the system occurs alarm or performance, log file missing. If the category of the user's intention is casual conversation and other intentions, a general large model is called for processing. 8.The intent recognition based 5GC platform intelligent interaction method according to claim 1, wherein, The user intention implementation result is sent to the client for presentation, which is specifically: A streaming response method is used to present the intention implementation. The client establishes a connection with the server through the WebSocket protocol to realize two-way real-time communication, and pushes the user intention implementation result to the client in real time.

9. An electronic device, comprising: It includes: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the 5GC platform intelligent interaction method based on intention recognition according to any one of claims 1-8.

10. A computer readable storage medium storing a computer program, characterized in that, The computer program makes the computer execute the 5GC platform intelligent interaction method based on intention recognition according to any one of claims 1-8.

Images