CN118797072A - Question-and-answer based network topology map generation method, device, medium and product - Google Patents

Abstract

The present invention discloses a method, device, medium and product for generating a network topology map based on question and answer, which automatically asks and answers questions according to the interactive data input by the user to generate dialogue data; the entities and intentions in the automatic dialogue data are respectively mapped to the nodes and connection relationships in the topology map to obtain the structured data of the topology map; an XML document of the topology map is created according to the dom4j parsing package, the structured data of the topology map is added to the XML document to generate an XML file; the XML file is loaded and rendered to obtain the network topology map. The generation of the network topology map of the present application solution does not need to rely on SSH commands, has high compatibility, high flexibility and faster speed.

Description

Question-and-answer based network topology map generation method, device, medium and product

Technical Field

The present invention relates to the field of computer technology, and in particular to a method, device, medium and product for generating a network topology diagram based on question and answer.

Background Art

A network topology diagram is a tool that graphically represents the structure of a network. It shows the devices in the network and how they are connected. The nodes in the network topology diagram represent the devices in the network, such as routers, switches, servers, etc., while the connecting lines represent the physical or logical connections between the devices. A network topology diagram can help you understand and plan the network architecture.

The existing network topology map generation method requires logging into the server through the SSH (Secure Shell) command, checking and recording all FC port information on the server, and then logging into the storage device through SSH to check all FC (Fibre Channel) port information on the storage and network connection relationship information. By comparing the information of the server and the storage device, the connection topology relationship between the server FC port and the storage FC port is found, and the human-computer interaction is completed through the mouse to draw the topology map.

The existing network topology map generation solution relies on SSH commands, which has poor compatibility, poor flexibility and slow speed in the topology map generation process.

Summary of the invention

Compared with the prior art, the present invention proposes a method, device, medium and product for generating a network topology map based on question and answer. The generation of the network topology map does not need to rely on SSH commands, has high compatibility, high flexibility and faster speed.

An embodiment of the present invention provides a method for generating a network topology map based on question and answer, the method comprising:

Automatically answer questions based on the interaction data input by the user to generate conversation data;

Mapping entities and intentions in the automatic dialogue data to nodes and connection relationships in a topological graph, respectively, to obtain topological graph structured data;

Creating an XML document of a topology map according to a dom4j parsing package, adding structured data of the topology map to the XML document, and generating an XML file;

The XML file is loaded and rendered to obtain a network topology diagram.

Preferably, an XML document of a topology map is created according to a dom4j parsing package, and the structured data of the topology map is added to the XML document to generate an XML file, including:

Traverse the nodes and connection relationships in the topology structured data, and create corresponding XML elements according to the dom4j parsing package;

Add the created XML element to the XML document, add nodes and attributes, and obtain an XML file.

Preferably, the method further comprises:

The nodes in the network topology graph are arranged using a preset layout algorithm.

Further, arranging the nodes in the network topology diagram using a preset layout algorithm includes:

Obtaining the number of nodes in the network topology graph;

Determining the radius of the circular layout according to the layout and size of the network topology diagram;

Calculate the angle value of each node in the network topology diagram according to the index value of the node;

Determine the coordinate positions of different nodes according to the angle value and radius of each node;

The coordinate position of each node is applied to the network topology diagram to complete the circular layout.

Preferably, arranging the nodes in the network topology diagram using a preset layout algorithm includes:

Obtain the number of nodes and connection relationships in the network topology diagram;

Create an empty layout result data structure;

Taking the root node of the network topology as the origin, traverse the nodes of each layer, calculate the vertical distance between the child node and the root node as the vertical coordinate; take the preset offset distance as the horizontal distance between each child node and its parent node, and determine the horizontal coordinate of the child node;

The horizontal coordinate and the vertical coordinate of each node are saved as position information in the layout result data structure;

The coordinate position of each node is applied to the network topology diagram to complete the hierarchical layout.

Preferably, mapping entities and intentions in the automatic dialogue data to nodes and connection relationships in a topological graph respectively to generate topological graph structured data includes:

Preprocessing the text in the automatic dialogue data to remove text characters;

Identify entity information in the preprocessed conversation text and map it to nodes in the topology graph;

Using natural language processing technology to identify the conversation intention in the conversation text, and mapping it to a connection relationship between a start node and a target node;

Output topology structured data in JSON format based on nodes and corresponding connection relationships.

Preferably, the topology structured data includes node information, connection information and module information;

The node information includes node unique identification, name, type and level information;

The connection information includes a unique identifier of the connection, a connection name, a connection type, a starting connection node, a target connection node, and other connection information;

The module information includes the current module information and a list of all nodes and connections under the current module.

Preferably, the method further comprises:

Parsing the topological map structured data and converting it into structured data recognizable by a program;

Extracting information from the conversation data, calculating the matching degree of each extracted keyword in the structured data and the conversation information by using a preset keyword matching algorithm, and determining the overall matching degree between the structured data and the conversation information;

Output topological structured data with a high degree of matching.

Preferably, the method further comprises:

On the front end, use API endpoints and components based on the bpmn.js library to define node types, connection types, and interaction behaviors, and create a topology editor interface;

The interactive instructions input by the user through the topology diagram editor interface are identified through event monitoring and callback functions, and the interactive instructions are executed.

Preferably, the method further comprises:

Based on receiving the topology map data and topology request sent by the front end through the API endpoint defined by Spring Boot on the back end, the topology map data is processed using the controller and service layer of Spring Boot, a new XML file is generated, and returned to the front end.

Preferably, the automatic question-answering based on the interactive data input by the user to generate the dialogue data includes:

Perform speech recognition on the interactive data input by the user and convert it into natural language text;

Performing natural language understanding on the natural language text to extract the intent and slot-value pairs of the current text;

Perform dialogue management on the intent and slot value pairs of the current text to obtain response actions;

Performing natural language generation on the response action to obtain a natural language text output;

The interaction data and the natural language text are output as the conversation data.

Preferably, the interactive data input by the user is subjected to speech recognition and converted into natural language text, including:

Preprocessing the interaction data;

Extract features from the preprocessed signal waveform of the interactive data to obtain speech features;

Recognizing the speech features through a pre-trained acoustic model to obtain a phoneme sequence;

The phoneme sequence is converted into a word sequence through a dictionary, and the order of the word sequence is adjusted through a pre-trained language model to obtain a recognition result as the natural language text.

Preferably, the natural language text is subjected to natural language understanding to extract the intent and slot-value pairs of the current text, including:

The natural language text is input into a pre-trained text recognition model, and the intent and slot-value pairs of the current text are determined according to the output results.

Preferably, the text recognition model training process includes:

Collect conversation data and perform preprocessing;

Define intent categories based on the application scenarios and requirements of the preprocessed conversation data;

Label the intent of each conversation data and the slot value information in the conversation data;

Add special characters to the annotated conversation data and convert the standardized conversation data into a preset format;

Divide the converted conversation data into training set, validation set and test set according to the preset division ratio;

The preset BERT model is trained using a cross entropy loss function according to the divided training set and the test set, and the BERT model is optimized by using Adam as an optimizer;

The trained model is verified through the verification set, and the hyperparameters of the BERT model are adjusted according to the verified loss value until the model loss value of the verification set reaches the threshold requirement, and the text recognition model is output.

Preferably, the dialog management of the intent and slot value pairs of the current text to obtain a response action includes:

Determine the current conversation state based on the pre-stored conversation history and the intent and slot value pairs of the current text;

Matches the preset response action according to the current session state.

Furthermore, determining the current session state according to the pre-stored conversation history and the intent and slot-value pairs of the current text includes:

In task-based human-computer dialogue, the intention and slot value pairs of the text extracted from each latest round of dialogue are added to the corresponding slots to track the dialogue status and update the current dialogue status.

Preferably, the method further comprises:

When the response action is an inquiry action, questions are asked to the user based on the current dialog state, and the user is guided to answer the key slot value;

When the response action is an answer action, matching the answer text in the knowledge base based on the user's question;

When the response action is a drawing action, a corresponding network topology map is drawn according to a preset topology map template and slot value pair.

Preferably, the generating the response action in natural language to obtain a natural language text output includes:

Determine the text of the information to be communicated according to the response action;

Organize the text structure sequence in the message being conveyed;

Aggregate text information that constitutes the text structure sequence into text sentences;

Adding conjunctions to the text information of the aggregated text sentences to form natural language;

According to the domain to which the natural language belongs, an extended vocabulary is determined, a natural language sentence is constructed, and a natural language text output is obtained.

The embodiment of the present invention further provides a network topology diagram generating device based on question and answer, the device comprising:

Intelligent dialogue module, used to automatically answer questions based on the interaction data input by the user and generate dialogue data;

A data mapping module, used to map entities and intentions in the automatic dialogue data into nodes and connection relationships in a topological graph, respectively, to obtain topological graph structured data;

A creation module is used to create an XML document of a topology map according to a dom4j parsing package, add the structured data of the topology map to the XML document, and generate an XML file;

The loading module is used to load and render the XML file to obtain a network topology diagram.

Preferably, the creation module is used to:

Traverse the nodes and connection relationships in the topology structured data, and create corresponding XML elements according to the dom4j parsing package;

Add the created XML element to the XML document, add nodes and attributes, and obtain an XML file.

Preferably, the device further comprises a layout module, configured to:

The nodes in the network topology graph are arranged using a preset layout algorithm.

Preferably, the layout module is specifically used for:

Obtaining the number of nodes in the network topology graph;

Determining the radius of the circular layout according to the layout and size of the network topology diagram;

Calculate the angle value of each node in the network topology diagram according to the index value of the node;

Determine the coordinate positions of different nodes according to the angle value and radius of each node;

The coordinate position of each node is applied to the network topology diagram to complete the circular layout.

Preferably, the layout module is specifically used for:

Obtain the number of nodes and connection relationships in the network topology diagram;

Create an empty layout result data structure;

Taking the root node of the network topology as the origin, traverse the nodes of each layer, calculate the vertical distance between the child node and the root node as the vertical coordinate; take the preset offset distance as the horizontal distance between each child node and its parent node, and determine the horizontal coordinate of the child node;

The horizontal coordinate and the vertical coordinate of each node are saved as position information in the layout result data structure;

The coordinate position of each node is applied to the network topology diagram to complete the hierarchical layout.

Preferably, the data mapping module is used to:

Preprocessing the text in the automatic dialogue data to remove text characters;

Identify entity information in the preprocessed conversation text and map it to nodes in the topology graph;

Using natural language processing technology to identify the conversation intention in the conversation text, and mapping it to a connection relationship between a start node and a target node;

Output topology structured data in JSON format based on nodes and corresponding connection relationships.

Preferably, the topology structured data includes node information, connection information and module information;

The node information includes node unique identification, name, type and level information;

The connection information includes a unique identifier of the connection, a connection name, a connection type, a starting connection node, a target connection node, and other connection information;

The module information includes the current module information and a list of all nodes and connections under the current module.

Preferably, the device further comprises an auxiliary module, specifically configured to:

Parsing the topological map structured data and converting it into structured data recognizable by a program;

Extracting information from the conversation data, calculating the matching degree of each extracted keyword in the structured data and the conversation information by using a preset keyword matching algorithm, and determining the overall matching degree between the structured data and the conversation information;

Output topological structured data with a high degree of matching.

Preferably, the device further comprises a front-end module, specifically used for:

On the front end, use API endpoints and components based on the bpmn.js library to define node types, connection types, and interaction behaviors, and create a topology editor interface;

The interactive instructions input by the user through the topology diagram editor interface are identified through event monitoring and callback functions, and the interactive instructions are executed.

Preferably, the device further comprises a back-end module, specifically configured to:

Based on receiving the topology map data and topology request sent by the front end through the API endpoint defined by Spring Boot on the back end, the topology map data is processed using the controller and service layer of Spring Boot, a new XML file is generated, and returned to the front end.

Preferably, the intelligent dialogue module is specifically used for:

Perform speech recognition on the interactive data input by the user and convert it into natural language text;

Performing natural language understanding on the natural language text to extract the intent and slot-value pairs of the current text;

Perform dialogue management on the intent and slot value pairs of the current text to obtain response actions;

Performing natural language generation on the response action to obtain a natural language text output;

The interaction data and the natural language text are output as the conversation data.

Preferably, the intelligent dialogue module is specifically used for:

Preprocessing the interaction data;

Extract features from the preprocessed signal waveform of the interactive data to obtain speech features;

Recognizing the speech features through a pre-trained acoustic model to obtain a phoneme sequence;

The phoneme sequence is converted into a word sequence through a dictionary, and the order of the word sequence is adjusted through a pre-trained language model to obtain a recognition result as the natural language text.

Preferably, the intelligent dialogue module is specifically used for:

The natural language text is input into a pre-trained text recognition model, and the intent and slot-value pairs of the current text are determined according to the output results.

Preferably, the intelligent dialogue module is specifically used for:

Collect conversation data and perform preprocessing;

Define intent categories based on the application scenarios and requirements of the preprocessed conversation data;

Label the intent of each conversation data and the slot value information in the conversation data;

Add special characters to the annotated conversation data and convert the standardized conversation data into a preset format;

Divide the converted conversation data into training set, validation set and test set according to the preset division ratio;

The preset BERT model is trained using a cross entropy loss function according to the divided training set and the test set, and the BERT model is optimized by using Adam as an optimizer;

The trained model is verified through the verification set, and the hyperparameters of the BERT model are adjusted according to the verified loss value until the model loss value of the verification set reaches the threshold requirement, and the text recognition model is output.

Preferably, the intelligent dialogue module is specifically used for:

Determine the current conversation state based on the pre-stored conversation history and the intent and slot value pairs of the current text;

Matches the preset response action according to the current session state.

Preferably, the intelligent dialogue module is specifically used for:

In task-based human-computer dialogue, the intention and slot value pairs of the text extracted from each latest round of dialogue are added to the corresponding slots to track the dialogue status and update the current dialogue status.

Preferably, the intelligent dialogue module is further used for:

When the response action is an inquiry action, questions are asked to the user based on the current dialog state, and the user is guided to answer the key slot value;

When the response action is an answer action, matching the answer text in the knowledge base based on the user's question;

When the response action is a drawing action, a corresponding network topology map is drawn according to a preset topology map template and slot value pair.

Preferably, the intelligent dialogue module is specifically used for:

Determine the text of the information to be communicated according to the response action;

Organize the text structure sequence in the message being conveyed;

Aggregate text information that constitutes the text structure sequence into text sentences;

Adding conjunctions to the text information of the aggregated text sentences to form natural language;

According to the domain to which the natural language belongs, an extended vocabulary is determined, a natural language sentence is constructed, and a natural language text output is obtained.

An embodiment of the present invention also provides a question-and-answer based network topology map generation device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a question-and-answer based network topology map generation method as described in any of the above embodiments.

An embodiment of the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a question-and-answer based network topology map generation method as described in any of the above embodiments.

An embodiment of the present invention further provides a computer program product, including a computer program/instruction, which implements the steps of the method described in any of the above embodiments when executed by a processor.

Compared with the prior art, the present invention provides a method, device, medium and product for generating a network topology map based on question and answer, which automatically answers questions according to the interactive data input by the user to generate conversation data; the entities and intentions in the automatic conversation data are respectively mapped to nodes and connection relationships in the topology map to obtain structured data of the topology map; an XML document of the topology map is created according to the dom4j parsing package, and the structured data of the topology map is added to the XML document to generate an XML file; the XML file is loaded and rendered to obtain the network topology map. The generation of the network topology map of the present application solution does not need to rely on SSH commands, and has high compatibility, high flexibility and faster speed.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a flow chart of a method for generating a network topology map based on question-answering provided by an embodiment of the present invention;

2 is another schematic diagram of a flow chart of a method for generating a network topology map based on question-answering provided by an embodiment of the present invention;

3 is a schematic diagram of a flow chart of an automatic speech recognition process provided by an embodiment of the present invention;

FIG4 is a schematic diagram of a flow chart of a feature extraction process provided by an embodiment of the present invention;

5 is another schematic diagram of a flow chart of an automatic speech recognition process provided by an embodiment of the present invention;

FIG6 is a schematic diagram of a training process of a text recognition model provided by an embodiment of the present invention;

FIG7 is a schematic diagram of the principle of a text recognition model training process provided by an embodiment of the present invention;

8 is a flow chart of a dialog management process according to an embodiment of the present invention;

9 is another schematic diagram of a flow chart of a method for generating a network topology map based on question and answer provided in an embodiment of the present invention;

10 is a schematic diagram of a process of outputting a natural language text according to an embodiment of the present invention;

11 is a schematic diagram of the structure of a network topology diagram generating device based on question-answering provided in an embodiment of the present invention;

FIG. 12 is another schematic diagram of the structure of a network topology map generating device based on question and answer provided in an embodiment of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Referring to FIG. 1 , it is a flow chart of a method for generating a network topology map based on question and answer provided in an embodiment of the present invention, wherein the method comprises steps S1 to S4;

S1, automatically answer questions based on the interaction data input by the user to generate dialogue data;

S2, mapping entities and intentions in the automatic dialogue data into nodes and connection relationships in a topological graph, respectively, to obtain topological graph structured data;

S3, creating an XML document of a topology map according to a dom4j parsing package, adding the structured data of the topology map to the XML document, and generating an XML file;

S4, loading and rendering the XML file to obtain a network topology diagram.

In the specific implementation of this embodiment, the application scheme automatically generates a network topology diagram based on question and answer, that is, after the user inputs the interaction data, the interaction data input by the user is automatically answered through intelligent question and answer, the question and answer results are output, and the dialogue data is generated according to the user voice interaction process.

It should be noted that the interaction data includes dialogue data input by the user into the device through natural language, voice and other input methods for human-computer interactive question and answer.

By using technologies such as natural language processing and intent recognition, entities and intents in the automatic conversation data are identified, and the entities and intents in the conversation content are mapped to nodes and relationships in a topological graph to generate topological graph structured data;

Use dom4j to create an empty topology map XML document, add the nodes and connection relationships in the topology map structured data to the XML document, gradually construct the nodes and relationships of the topology map XML file, and generate an XML file;

Save the generated topology map XML document as a file. The front end will use the API provided by bpmn.js to load and render the XML file, parse the XML file into a topology map model, render the parsed topology map model to the page, and display it to the user, completing the generation of the network topology map.

After the required topology map requirements are input through natural language dialogue, voice dialogue and other interactive methods, the application solution automatically extracts and collects data related to the topology map, such as the connection relationship between nodes, device information and other useful information, processes and analyzes the data to determine the connectivity between the nodes, and automatically designs a suitable topology map structure based on the data analysis results, eliminating the complicated steps of manually drawing topology maps in the context of increasingly complex network structures today, thereby improving work efficiency.

In another embodiment provided by the present invention, the step S3 includes:

Use dom4j to create an empty topology map XML document object. According to the topology map structured data, traverse the device and connection information. At the same time, use dom4j to create corresponding XML elements and add them to the topology map XML document, such as Element, Attribute, etc., to create XML elements and attributes, so as to gradually build the nodes and relationships of the topology map XML file, add nodes and attributes, form a hierarchical structure, and obtain an XML document.

In another embodiment of the present invention, when generating the XML file, the method further includes: using a preset layout algorithm to arrange the nodes in the network topology map, and arranging the nodes in the topology map according to a set format.

The preset layout algorithms include a hierarchical layout algorithm and a circular layout algorithm. According to the preset algorithms, the nodes of the network topology diagram are arranged to better display the relationship and structure between the nodes.

In another embodiment provided by the present invention, a circular layout algorithm may be used to arrange the nodes in the network topology diagram, specifically comprising the following steps:

Get the number of nodes in the topology graph, assuming it is n.

Calculate the radius r of the circle and determine the appropriate radius value based on the size of the topology map and layout requirements.

For each node, the angle value angle of each node on the circle of the network topology graph is calculated according to the index value of the node. angle = (2*π*i)/n, where i represents the index of the node, from 0 to n-1.

The coordinate position of each node on the circle is calculated according to the angle value and radius of each node, that is, the coordinates (x, y) of the node in the plane coordinate system are calculated, x = r*cos(angle), y = r*sin(angle).

Apply the coordinate position of each node to the nodes in the topology map to complete the circular layout.

Arrange the nodes in the topology diagram in a circular form to better display the relationship and structure between the nodes.

In another embodiment provided by the present invention, a hierarchical layout algorithm may be used to arrange the nodes in the network topology diagram, specifically comprising the following steps:

Get the relationship between the number of device nodes and cables in the topology diagram.

Create an empty layout result data structure to store the position information of each node.

Find the root node (node without a parent node) in the topology graph and use it as the node of the first layer. For the root node, set its position to the initial position, that is, as the origin position.

Traverse the nodes of each layer: For each node in the current layer, calculate its vertical distance from the root node as the y coordinate. Calculate the position of its parent node and offset it based on the parent node to ensure the hierarchical relationship of the nodes. Use the preset offset distance as the horizontal distance between each child node and its parent node to determine the horizontal coordinate of the child node;

Save the position information of each node into the layout result data structure.

Apply the layout results to the nodes in the topology graph to complete the hierarchical layout.

Using the hierarchical layout algorithm, the nodes are laid out according to the hierarchical structure to show the parent-child relationship and hierarchical structure between the nodes.

In another embodiment provided by the present invention, the above step S2 specifically includes:

The text preprocessing is performed on the conversation content in the automatic conversation data, including removing stop words, punctuation marks and special characters.

Identify entity information in the conversation, such as devices, cables, modules, etc., and use the identified entity information as nodes or elements in the topology map.

The natural language processing technology is used to identify the conversation intention in the conversation text, map it to the connection relationship between the starting node and the target node, and determine the different business processes in the topology map.

Output topology structured data in JSON format based on nodes and corresponding connection relationships.

The present application scheme defines the device node types and cable connection types in the topology map, such as the starting connection node, the target connection node, etc., and maps them to nodes and relationships in the topology map according to the entities and intentions in the conversation content, and outputs the topology map structured data in JSON format for constructing the network topology map.

In another embodiment provided by the present invention, the topology structured data includes the following types: node information, connection information and module information;

Node information includes the unique identifier, name, type, and hierarchical information of the device node (the hierarchical position of the device node in the topology map).

Connection information includes the unique connection identifier, connection name, connection type, starting connection node, target connection node, and other connection information.

Module information: Contains the current module information and a list of all nodes and connections under the module.

By collecting various information related to the topology map and converting it into the structured data format above for storage, it is convenient for subsequent processing and analysis.

In another embodiment provided by the present invention, the present application also assists in obtaining structured data through a dialogue template, specifically assists in obtaining topology structured data:

Parse the pre-stored topology structured data and convert it into a data structure that can be recognized by the program. After extracting the dialogue information input by the user, calculate the matching degree of each keyword in the topology structured data and the dialogue information through the keyword matching algorithm. This module comprehensively considers the matching degree of all keywords, calculates the overall matching degree between the topology data and the dialogue information, and finally outputs the topology structured information with a high matching degree, realizes the filtering of the topology structured data, and improves the quality of the topology structured data.

In another embodiment provided by the present invention, the present application solution also uses the front-end Bpmn.js library to construct a network topology map, specifically:

Integrate the bpmn.js library into the front-end application, use the bpmn.js API and components to define node types, connection types, and interaction behaviors; create a topology editor interface by introducing related dependencies and resource files.

The interactive instructions input by the user through the topology map editor interface are identified through event monitoring and callback functions, and the interactive instructions are executed to realize the editing, saving and exporting functions of the topology map.

This embodiment combines the method of automatically parsing the logical network topology diagram into structured data with Web front-end technology to provide an overall solution for improving the work quality and efficiency of network planning and design. It applies Web technology and can be used through mainstream Web browsers. It has strong versatility and uses the React front-end framework to perform canvas interaction. The drag-and-drop operation has high flexibility and ease of use. A centralized unified maintenance management platform architecture is built based on online Web applications, laying a platform foundation for unified network construction business management and data standardization.

In another implementation provided by the present invention, the present application solution also uses Spring Boot in the backend to construct a network topology diagram, specifically:

The backend uses Spring Boot to configure related dependencies and routes. Define API endpoints to receive topology data and requests sent by the frontend. Use Spring Boot's controller and service layer to process frontend requests and generate XML data for the topology map, generate a new XML file, and return it to the frontend.

The XML file is constructed by using the front-end Bpmn.js library, the back-end Spring Boot framework and the DOM4J library to realize the function of automatically generating the topology map through intelligent dialogue. The intelligent dialogue template selected by the user is converted into the structured data of the topology map or the structured data of the topology map is automatically generated through intelligent dialogue.

In another embodiment of the present invention, when the present invention performs automatic dialogue processing, refer to FIG2 , which is another flowchart of the method for generating a network topology diagram based on question and answer provided by an embodiment of the present invention.

Generating dialogue data specifically includes the following steps, namely automatic speech recognition (ASR), natural language understanding (NLU), dialogue management (DM) and natural language generation (NLG).

Among them, the role of automatic speech recognition ASR is to perform speech recognition on the interactive data input by the user and convert it into natural language text.

The role of natural language understanding (NLU) is to convert natural language into a field that can be processed by machines, perform natural language understanding on the natural language text, and extract the intent and slot-value pairs of the current text.

The role of dialogue management (DM) is to determine the action or strategy that the device needs to take, that is, to conduct dialogue management on the intent and slot value pairs of the current text to obtain response actions.

Natural language generation (NLG) performs natural language generation on the response action to obtain a natural language text output and a system reply, that is, converting the output action into a text result expressed in natural language.

Natural language generation is performed on the response action to obtain a natural language text output.

By conducting multiple rounds of interactions with users, we can extract structured information about the network topology from the user's utterances.

In another embodiment of the present invention, during automatic speech recognition, the following steps are specifically performed:

To realize the intelligent question-answering module, speech recognition technology (Automatic Speech Recognition, ASR) is used to convert user speech into text. Speech recognition technology includes four parts: speech signal preprocessing, feature extraction, pattern matching, and searching for pattern libraries. First, the input speech needs to be preprocessed, and then the speech features are extracted. On this basis, the template required for speech recognition is established.

The construction process of a speech recognition device is generally divided into two parts: training and recognition. The training process is usually completed offline, by performing signal processing and knowledge mining on a large number of pre-collected speech and language databases to obtain the "acoustic model" and "language model" required by the speech recognition device; while the recognition process is usually completed online, automatically recognizing the real-time speech input by the user. The recognition process can usually be divided into two major modules: the "front-end" and the "back-end". The main function of the "front-end" module is to perform endpoint detection (removing redundant silence and non-speech sounds), noise reduction, feature extraction, etc.; the function of the "back-end" module is to use the trained "acoustic model" and "language model" to perform statistical pattern recognition (also known as "decoding") on the feature vector of the user's speech to obtain the text information it contains. In addition, the back-end module also has an "adaptive" feedback module that can self-learn the user's speech, thereby performing necessary "corrections" on the "acoustic model" and "speech model" to further improve the recognition accuracy.

See Figure 3, which is a flow chart of the automatic speech recognition process provided by an embodiment of the present invention. The input of speech recognition is generally a speech signal in the time domain, which is generally represented by a vector in mathematics; the output is text, which is represented by a token. In general, the length of the input speech signal is greater than the length of the output text token.

First, the voice signal is preprocessed. The sound signal is essentially a wave. The actual voice input environment and other influencing factors cause the input voice signal to be incomplete and meaningful. Therefore, the input voice signal needs to be filtered first to cut off the silent part at the beginning and end to reduce the interference to the subsequent steps.

In the intelligent question-answering task scenario, considering that the noise impact in the actual working environment is relatively small, a relatively simple energy-based method is used to remove the silent part, and the silent part is removed by removing the frames whose average energy is lower than 0.01 times the average energy of the entire input speech. This embodiment uses the LTSD (Long-term Spectral Divergence) algorithm to decompose the speech into overlapping frames, and score each frame according to the possibility of speech activity in the frame, and then accumulate the probabilities to extract all intervals with speech activity.

Then feature extraction is performed. Feature extraction is an important part of speech recognition. It analyzes the waveform of the speech signal, extracts parameters that can characterize the characteristics of the speech signal, and converts each frame of the waveform into a multidimensional vector containing sound information to facilitate subsequent pattern matching and recognition.

MFCC (Mel Frequency Cepstral Coefficient) is used as a feature extraction method, which is mainly used for interactive data feature extraction and reducing the operation dimension. Referring to Figure 4, it is a flow chart of the feature extraction process provided by an embodiment of the present invention. First, the speech input is pre-emphasized, framed, and windowed, and a fast Fourier transform, i.e., FFT, is performed. The absolute value or square value of the transformation result is taken, and then Mel filtering is performed. After discrete cosine transform DCT, dynamic features are extracted to obtain feature vectors to complete feature extraction.

Then pattern matching is performed. Pattern matching is the core step of speech recognition. It obtains the best matching result by matching the input speech features with the pre-trained speech recognition model.

This embodiment may selectively use a hidden Markov model to implement pattern matching to process time series data, and use a Viterbi algorithm to search for an optimal path in the established state network.

Determine the matching result through the acoustic model, dictionary and language model, and recognize the speech feature through the pre-trained acoustic model to obtain a phoneme sequence;

The phoneme sequence is converted into a word sequence through a dictionary, and the order of the word sequence is adjusted through a pre-trained language model to obtain a recognition result as the natural language text

Specifically, the acoustic model AM is obtained by training a large amount of interactive data, with feature vectors as input and phoneme information as output. The acoustic model is usually represented by a hidden Markov model (HMM). In speech recognition, the acoustic model uses HMM to describe the transition probability between each phoneme in the speech signal.

A dictionary is a set of all possible phonemes, which is the correspondence between characters or words and phonemes. In speech recognition, the role of the dictionary is to convert the phoneme sequence output by the acoustic model into a specific word sequence. For example, for Chinese speech recognition, the dictionary is the correspondence between pinyin and Chinese characters.

The language model is another important component of speech recognition. It obtains the relationship between each word in the speech signal by analyzing the context information of the speech signal, and adjusts the word sequence output by the acoustic model to make it more in line with the actual language usage habits, thereby improving the accuracy of speech recognition. This embodiment uses conditional random fields (CRF) to build a language model, which obtains the relationship between each word in the speech signal by establishing the conditional probability distribution between each word in the speech signal.

Refer to Figure 5, which is another flow chart of the automatic speech recognition process provided by an embodiment of the present invention. During speech recognition, the user's speech input is recognized to obtain a speech signal, the speech signal is divided into different signal frames, and the signal frames are feature extracted to obtain a feature vector. The acoustic model, dictionary and language model together constitute the recognition engine. The recognition engine receives the input speech features, obtains a phoneme sequence through the acoustic model, and then converts the phoneme sequence into a word sequence through the dictionary, and finally adjusts the word sequence through the language model to obtain the final recognition result. In the intelligent question and answer module where user needs are currently input through voice input.

In another embodiment provided by the present invention, natural language understanding NLU is a key component of human-computer dialogue. The main task of NLU is to understand and parse the user's natural language text output by ASR, and convert unstructured language data into a data form that can be understood by a structured model. The basic goal of NLU is to extract key information from user sentences: user intent and slot value. The former is modeled as an intent recognition task, and the latter is modeled as a slot filling task. The natural language understanding process specifically includes the following steps:

The main goal of intent recognition is to extract specific intent from the user's natural language input, which is usually modeled as a classification task. Slot filling is usually defined as a sequence labeling problem. See Table 1, which is a correspondence table of user utterances, user intents, and slot value pairs.

Table 1 Correspondence table

User Quotes I want to make a network topology diagram for an IT cloud User Intent Draw a network topology diagram Slot-value pairs Field = IT Cloud

The main goal of intent recognition is to extract specific intent from the user's natural language input, which is usually modeled as a classification task. Slot filling is usually defined as a sequence labeling problem.

The natural language text is input into a pre-trained text recognition model, and the intent and slot-value pairs of the current text are determined according to the output results to complete the text recognition.

In another embodiment provided by the present invention, when training the text recognition model, the following steps are specifically performed:

In order to fine-tune BERT, we first need to construct a batch of training data. Based on large-scale human-to-human conversation data, we can semi-automatically construct supervised training data. See Figure 6, which is a schematic diagram of the training process of the text recognition model provided by an embodiment of the present invention.

The data construction process mainly includes six steps:

Data collection,The data mainly comes from two parts. One is a large amount of real conversation data accumulated in an online template way, and the other is a large language model that designs specific prompts to assist in generating conversation data based on given domain knowledge.

Data preprocessing: Clean and preprocess the collected data, which generally includes correcting language errors, standardizing abbreviations and emoticons, removing special characters and noise, etc.

Intent definition: Based on the specific application scenarios and requirements, define a set of user intent categories to be identified, such as "drawing a network topology map", "defining POD type", "modifying device links", etc.

Data labeling: For each conversation data, determine the user's main intention and mark the corresponding category. Furthermore, it is necessary to label the slot value information in the discourse, such as "domain" and "POD type". Data labeling usually requires rigorous executable labeling rules and a professional manual labeling team. With the development of large language models, large models have very strong generalization capabilities in unseen fields and can be used to assist in the extraction of user intentions and slot value information. The manual labeling team only needs to correct the labeling results of the large model, which can greatly improve the labeling efficiency.

BERT preprocessing: Convert the annotated conversation data into an input format that BERT can understand. This usually includes adding special characters (such as '[CLS]' and '[SEP]'), word segmentation, etc.

Divide the data set: randomly divide the labeled data into training set, validation set and test set according to the preset division ratio. Preferably, the ratio can be selected as 8:1:1 for model training and model verification. See Table 2, which is an example table of the data set.

Table 2 Dataset example table

See Figure 7, which is a schematic diagram of the principle of the text recognition model training process provided by an embodiment of the present invention. The BERT model structure is used. The model is a multi-layer bidirectional Transformer encoder designed on the basis of the original Transformer model, and its input is a combination of the word vector representation, position representation and fragment representation of WordPiece. A special classification character ([CLS]) is inserted as the first word, and a special word ([SEP]) is added as the last word. The BERT model uses two strategies to pre-train on large-scale unlabeled text. The BERT model uses the hidden state output by the first special tag [CLS] to predict user intent, and performs softmax on the output of the last layer of all words to predict the sequence label, where: B indicates that the current word is the beginning of an entity; I indicates that the current word is inside an entity, but not at the starting position; O indicates that the current word does not belong to any entity. The BERT model is trained using the cross entropy loss function, using Adam as the optimizer, and all hyperparameters are adjusted according to the loss value on the validation set. The model with the best performance on the validation set is selected as the final model.

Compared with the traditional NLU model based on the RNN architecture, the BERT model based on the Transformer architecture used in this embodiment is faster to train and has better generalization, and has a simpler structure, which is suitable for the construction of NLU models in vertical fields.

In another embodiment provided by the present invention, the main task of the dialogue management DM is to complete the action decision according to the intent and slot value extracted from the NLU module. The dialogue management DM is divided into two sub-modules: dialogue state tracking DST and dialogue strategy optimization DPO. See Figure 8, which is a flowchart of the dialogue management process provided by an embodiment of the present invention.

When managing dialogues, dialogue state tracking (DST) is responsible for determining the current dialogue state based on dialogue history and user input, and dialogue policy optimization (DPO) is responsible for making the most appropriate response action based on the current dialogue state.

In another embodiment provided by the present invention, in a task-based human-computer dialogue, the dialogue state is a series of slots and slot values maintained by the DST. When the dialogue state tracking DST determines the current session state, the following steps are specifically included:

In task-based human-computer dialogue, the intent and slot value pairs of the text extracted from each latest dialogue are added to the corresponding slots;

Complete the update of the conversation status and update the current conversation status.

In another embodiment provided by the present invention, the dialogue process is regarded as a Markov decision process, that is, the factors affecting the current device action decision are only related to the current dialogue state, and have nothing to do with the historical dialogue state, so DST only maintains the current dialogue state. The dialogue state maintained by DST is implemented through an index table. At the beginning of the dialogue, the device initializes the dialogue state according to the domain of the network topology map selected by the user. The initial dialogue state is an index table in which the index key is the slot name of all the parameters required to draw the network topology map of the domain, and the index value is empty. In the index table, some slots are defined as key slots, corresponding to the necessary parameters for completing the drawing of the network topology map. After multiple rounds of dialogue, when all the key slots are filled, the device will draw the corresponding network topology map according to the predetermined topology map template and slot value parameters.

The dialogue strategy optimization DPO is designed using a predetermined rule method and implemented based on a finite state automaton (FSA). During the dialogue process, the DPO executes pre-set response actions based on the current dialogue state maintained by the DST. The actions are divided into three categories: inquiry actions, answer actions, and drawing actions. See Table 3 for a table of the functions of each action.

Table 3 Action function table

Action Function ask Ask users questions based on the current conversation state and guide them to answer key slot values answer Based on the user's question, match the answer in the knowledge base to answer the question Drawing Draw the corresponding network topology diagram according to the predetermined topology diagram template and slot value parameters

In a complete conversation process, the core task of the device is to guide the user to fill in the key slot value through multiple rounds of questions, and the secondary task is to answer the user's questions at any time. In addition, when the user cannot fully describe the required networking architecture, a similar template is automatically recommended for the user to choose. See Figure 9, which is another flow chart of the method for generating a network topology map based on question and answer provided by an embodiment of the present invention.

The state transition within the inquiry state is realized by updating the drawing parameter slot, and the state transition between the inquiry state and the answer state is realized by updating the intention slot. When all key slots are filled, the dialogue state is transferred from the inquiry state to the drawing state. In the three actions corresponding to different states, the implementation method of taking actions is different.

When the dialog state is in the inquiry state, the inquiry action is used. A question is generated based on the next key slot that needs to be filled by the user, and the user is asked to obtain the corresponding slot value. When the user cannot answer accurately, the slot will be filled with a special symbol to mark it as an uncertain state.

When the dialog state is the answer state, this embodiment adopts the answer action. By linking the external knowledge base, a large number of question-answer pairs are stored in the knowledge base. The user's question statement is first converted into a sentence vector, and the similarity is matched with the question vector in the knowledge base, and then the answer corresponding to the question with the highest similarity is returned to the user. The sentence vector and the question vector are respectively input into BERT by inputting the question statement and the question into BERT, and obtaining the vector corresponding to the [CLS] symbol. Since BERT has been pre-trained on a large-scale corpus, after the semantically similar sentences are input into BERT, the distance between the vectors output by their [CLS] symbols in the semantic space is close. Therefore, the cosine distance can be further used to measure the semantic similarity of the two sentence vectors. In addition, in order to save computational overhead, the question vector can be stored in the knowledge base in advance. After answering the user's question, this embodiment will automatically fall back to the previous inquiry state and ask the user again.

When the dialog state is the drawing state, the device adopts the drawing action. If no key slot is in an uncertain state, the device will call the drawing interface to complete the network topology drawing. Otherwise, the device will calculate the similarity between the determined drawing parameters and the template preset parameters, and take the most similar templates and return them to the user for selection.

In another embodiment provided by the present invention, natural language generation (NLG) is a subcategory of natural language processing (NLP), and its main purpose is to automatically convert structured data into human-readable text. The intelligent question-answering module for network topology generation belongs to a task-based dialogue device. The NLG in the task-based dialogue is to generate dialogue replies based on the learned strategies based on NLU (domain classification and intent recognition, slot value filling), DST, and DPL. The general replies include clarifying requirements, guiding users, asking, confirming, and concluding dialogues.

See Figure 10, which is a flow chart of the natural language text output process provided by an embodiment of the present invention. The natural language text output process is divided into 6 steps:

Step 1, Content Determination: Usually the data conveys more information than what is ultimately conveyed, and NLG decides what information should be included in the text being constructed.

Step 2: Text Structuring. After determining what information needs to be conveyed, the NLG device needs to organize the text structure order reasonably.

Step 3: Sentence Aggregation: Not every piece of information needs to be expressed in an independent sentence. The NLG device needs to merge multiple pieces of information into one sentence for coherent expression.

Step 4: grammaticalization. Once the content of each sentence is determined, the NLG device organizes the information into natural language by adding connectives between various information.

Step 5, Referring Expression Generation, the NLG device needs to identify the domain of the content and use the vocabulary of the domain to form a complete sentence;

Step 6, Linguistic Realisation, when all relevant words and phrases have been determined, the NLG device inputs the information into the template to obtain a complete sentence with good structure and accurate expression. See Table 4, which is an example of the implementation process in the current task scenario;

Table 4 Process example table

This embodiment uses a template-based NLG generation method, and the main application scenario is to obtain relevant structured information for generating a network topology map from a user through a dialogue device. The template-based NLG method mainly plays two roles, namely, delivering relevant information for building a network topology map to the user and conducting user inquiries.

In the specific implementation method, the user intention in the dialogue management (DM) module can be classified, and the corresponding templates can be manually designed in advance for the information transmitted by the user and the slots of the inquiry. The slots in the template are generally expressed using "<placeholder>". When the natural language text is actually generated, the slots in the template are replaced with the corresponding values obtained.

Table 5 Dialogue modules

The template-based NLG method has a simple working principle and accurate responses. Before using it, it is necessary to understand the task requirements in advance, set different dialogue scenarios manually, and design corresponding dialogue templates based on each dialogue scenario. Some components of the template are fixed, and the preset slots in the template need to be filled according to the output of the dialogue management (DM) module.

In actual use, it is necessary to maintain the dialog template collection of the current task. When the business scenario is expanded, it can be expanded and maintained by adding new templates to the template collection.

Referring to FIG. 11 , it is a schematic diagram of the structure of a network topology diagram generating device based on question-answering provided by an embodiment of the present invention, wherein the device comprises:

Intelligent dialogue module, used to automatically answer questions based on the interaction data input by the user and generate dialogue data;

A data mapping module, used to map entities and intentions in the automatic dialogue data into nodes and connection relationships in a topological graph, respectively, to obtain topological graph structured data;

A creation module is used to create an XML document of a topology map according to a dom4j parsing package, add the structured data of the topology map to the XML document, and generate an XML file;

The loading module is used to load and render the XML file to obtain a network topology diagram.

It should be noted that the question-and-answer based network topology map generation device provided in this embodiment can execute all the steps and functions of the question-and-answer based network topology map generation method provided in any of the above embodiments, and the specific functions of the device will not be described in detail here.

Refer to Figure 12, which is another structural schematic diagram of a network topology map generation device based on question and answer provided in an embodiment of the present invention. The network topology map generation device based on question and answer includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a network topology map generation program based on question and answer. When the processor executes the computer program, it implements the steps in each of the above-mentioned embodiments of the network topology map generation method based on question and answer, such as steps S1 to S4 shown in Figure 1. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-mentioned device embodiments.

Exemplarily, the computer program can be divided into one or more modules, and the one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the network topology map generation device based on question and answer. For example, the computer program can be divided into several modules, and the specific functions of each module have been described in detail in the network topology map generation method based on question and answer provided in any of the above embodiments, and the specific functions of the device will not be repeated here.

The network topology map generating device based on question and answer can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The network topology map generating device based on question and answer can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the schematic diagram is only an example of a network topology map generating device based on question and answer, and does not constitute a limitation of a network topology map generating device based on question and answer. It can include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the network topology map generating device based on question and answer can also include input and output devices, network access devices, buses, etc.

The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the network topology map generating device based on question and answer, and uses various interfaces and lines to connect the various parts of the entire network topology map generating device based on question and answer.

The memory can be used to store the computer program and/or module, and the processor realizes various functions of the network topology map generating device based on question and answer by running or executing the computer program and/or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

Wherein, if the module integrated in the network topology map generating device based on question and answer is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

An embodiment of the present invention further provides a computer program product, including a computer program/instruction, which implements the steps of the method described in any of the above embodiments when executed by a processor.

It should be noted that the computer program product provided in this embodiment can execute all the steps and functions of the question-and-answer based network topology map generation method provided in any of the above embodiments, and the specific functions of the device are not described in detail here.

It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the protection scope of the present invention.

Claims (22)

1. A method for generating a network topology diagram based on question and answer, characterized in that the method comprises: Automatically answer questions based on the interaction data input by the user to generate conversation data; Mapping entities and intentions in the automatic dialogue data to nodes and connection relationships in a topological graph, respectively, to obtain topological graph structured data; Creating an XML document of a topology map according to a dom4j parsing package, adding structured data of the topology map to the XML document, and generating an XML file; The XML file is loaded and rendered to obtain a network topology diagram. 2. The method for generating a network topology map based on question and answer according to claim 1 is characterized in that an XML document of a topology map is created according to a dom4j parsing package, and the structured data of the topology map is added to the XML document to generate an XML file, comprising: Traverse the nodes and connection relationships in the topology structured data, and create corresponding XML elements according to the dom4j parsing package; Add the created XML element to the XML document, add nodes and attributes, and obtain an XML file. 3. The method for generating a network topology diagram based on question and answer according to claim 1, characterized in that the method further comprises: The nodes in the network topology graph are arranged using a preset layout algorithm. 4. The method for generating a network topology map based on question and answer according to claim 3, wherein the nodes in the network topology map are arranged using a preset layout algorithm, comprising: Obtaining the number of nodes in the network topology graph; Determining the radius of the circular layout according to the layout and size of the network topology diagram; Calculate the angle value of each node in the network topology diagram according to the index value of the node; Determine the coordinate positions of different nodes according to the angle value and radius of each node; The coordinate position of each node is applied to the network topology diagram to complete the circular layout. 5. The method for generating a network topology map based on question and answer according to claim 3, wherein the nodes in the network topology map are arranged using a preset layout algorithm, comprising: Obtain the number of nodes and connection relationships in the network topology diagram; Create an empty layout result data structure; Taking the root node of the network topology as the origin, traverse the nodes of each layer, calculate the vertical distance between the child node and the root node as the ordinate; taking the preset offset distance as the horizontal distance between each child node and its parent node, determine the abscissa of the child node; The horizontal coordinate and the vertical coordinate of each node are saved as position information in the layout result data structure; The coordinate position of each node is applied to the network topology diagram to complete the hierarchical layout. 6. The method for generating a network topology map based on question and answer according to claim 3, characterized in that the entities and intentions in the automatic dialogue data are respectively mapped to nodes and connection relationships in the topology map to generate topology map structured data, including: Preprocessing the text in the automatic dialogue data to remove text characters; Identify entity information in the preprocessed conversation text and map it to nodes in the topology graph; Using natural language processing technology to identify the conversation intention in the conversation text, and mapping it to a connection relationship between a start node and a target node; Output topology structured data in JSON format based on nodes and corresponding connection relationships. 7. The method for generating a network topology map based on question and answer according to claim 1, wherein the topology map structured data includes node information, connection information and module information; The node information includes node unique identification, name, type and level information; The connection information includes a unique identifier of the connection, a connection name, a connection type, a starting connection node, a target connection node, and other connection information; The module information includes the current module information and a list of all nodes and connections under the current module. 8. The method for generating a network topology diagram based on question and answer according to claim 1, characterized in that the method further comprises: Parsing the topological map structured data and converting it into structured data recognizable by a program; Extracting information from the conversation data, calculating the matching degree of each extracted keyword in the structured data and the conversation information by using a preset keyword matching algorithm, and determining the overall matching degree between the structured data and the conversation information; Output topological structured data with a high degree of matching. 9. The method for generating a network topology diagram based on question and answer according to claim 1, characterized in that the method further comprises: On the front end, use API endpoints and components based on the bpmn.js library to define node types, connection types, and interaction behaviors, and create a topology editor interface; The interactive instructions input by the user through the topology diagram editor interface are identified through event monitoring and callback functions, and the interactive instructions are executed. 10. The method for generating a network topology diagram based on question and answer according to claim 1, characterized in that the method further comprises: Based on receiving the topology map data and topology request sent by the front end through the API endpoint defined by Spring Boot on the back end, the topology map data is processed using the controller and service layer of Spring Boot, a new XML file is generated, and returned to the front end. 11. The method for generating a network topology map based on question and answer according to claim 1, characterized in that the step of automatically answering questions based on the interactive data input by the user to generate the dialogue data comprises: Perform speech recognition on the interactive data input by the user and convert it into natural language text; Performing natural language understanding on the natural language text to extract the intent and slot-value pairs of the current text; Perform dialogue management on the intent and slot value pairs of the current text to obtain response actions; Performing natural language generation on the response action to obtain a natural language text output; The interaction data and the natural language text are output as the conversation data. 12. The method for generating a network topology map based on question and answer according to claim 11, characterized in that the interactive data input by the user is subjected to speech recognition and converted into natural language text, comprising: Preprocessing the interaction data; Extract features from the preprocessed signal waveform of the interactive data to obtain speech features; Recognizing the speech features through a pre-trained acoustic model to obtain a phoneme sequence; The phoneme sequence is converted into a word sequence through a dictionary, and the order of the word sequence is adjusted through a pre-trained language model to obtain a recognition result as the natural language text. 13. The method for generating a network topology map based on question-answering according to claim 11, characterized in that the natural language text is subjected to natural language understanding to extract the intent and slot-value pairs of the current text, comprising: The natural language text is input into a pre-trained text recognition model, and the intent and slot-value pairs of the current text are determined according to the output results. 14. The method for generating a network topology map based on question and answer according to claim 13, wherein the text recognition model training process comprises: Collect conversation data and perform preprocessing; Define intent categories based on the application scenarios and requirements of the preprocessed conversation data; Label the intent of each conversation data and the slot value information in the conversation data; Add special characters to the annotated conversation data and convert the standardized conversation data into a preset format; Divide the converted conversation data into training set, validation set and test set according to the preset division ratio; The preset BERT model is trained using a cross entropy loss function according to the divided training set and the test set, and the BERT model is optimized by using Adam as an optimizer; The trained model is verified through the verification set, and the hyperparameters of the BERT model are adjusted according to the verified loss value until the model loss value of the verification set reaches the threshold requirement, and the text recognition model is output. 15. The method for generating a network topology map based on question and answer according to claim 11, wherein the step of performing dialog management on the intent and slot value pairs of the current text to obtain a response action comprises: Determine the current conversation state based on the pre-stored conversation history and the intent and slot value pairs of the current text; Matches the preset response action according to the current session state. 16. The method for generating a network topology map based on question and answer according to claim 15, wherein determining the current session state according to the pre-stored conversation history and the intent and slot value pairs of the current text comprises: In task-based human-computer dialogue, the intention and slot value pairs of the text extracted from each latest round of dialogue are added to the corresponding slots to track the dialogue status and update the current dialogue status. 17. The method for generating a network topology diagram based on question and answer according to claim 15, characterized in that the method further comprises: When the response action is an inquiry action, questions are asked to the user based on the current dialog state, and the user is guided to answer the key slot value; When the response action is an answer action, matching the answer text in the knowledge base based on the user's question; When the response action is a drawing action, a corresponding network topology map is drawn according to a preset topology map template and slot value pair. 18. The method for generating a network topology diagram based on question and answer according to claim 11, wherein the step of generating a natural language for the response action to obtain a natural language text output comprises: Determine the text of the information to be communicated according to the response action; Organize the text structure sequence in the message being conveyed; Aggregate text information that constitutes the text structure sequence into text sentences; Adding conjunctions to the text information of the aggregated text sentences to form natural language; According to the domain to which the natural language belongs, an extended vocabulary is determined, a natural language sentence is constructed, and a natural language text output is obtained. 19. A network topology diagram generation device based on question and answer, characterized in that the device comprises: Intelligent dialogue module, used to automatically answer questions based on the interaction data input by the user and generate dialogue data; A data mapping module, used to map entities and intentions in the automatic dialogue data into nodes and connection relationships in a topological graph, respectively, to obtain topological graph structured data; A creation module is used to create an XML document of a topology map according to a dom4j parsing package, add the structured data of the topology map to the XML document, and generate an XML file; The loading module is used to load and render the XML file to obtain a network topology diagram. 20. A question-and-answer based network topology map generation device, characterized in that it comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the question-and-answer based network topology map generation method as described in any one of claims 1 to 18. 21. A computer-readable storage medium, characterized in that the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the question-and-answer based network topology map generation method as described in any one of claims 1 to 18. 22. A computer program product, comprising a computer program/instruction, characterized in that when the computer program/instruction is executed by a processor, the steps of the method according to any one of claims 1 to 18 are implemented.