Intelligent railway question and answer management method and system based on AI technology

Abstract

According to the intelligent railway question and answer management method and system based on the AI technology, literature data and question and answer data in the railway field are collected, and a question and answer DIKW model is constructed according to the literature data and the question and answer data; receiving a question and answer request initiated by a user, and extracting user information to construct a user DIKW model; calling a question and answer DIKW model to process the user DIKW model, obtaining a target answer, and obtaining a virtual question based on the target answer; inputting the virtual question into a question and answer DIKW model, processing the virtual question by the question and answer DIKW model to obtain a virtual answer, and evaluating the association degree between the virtual answer and the target answer; when the association degree of the virtual answer and the target answer is greater than a preset threshold value, inputting a question and answer request initiated by a user into a question and answer DIKW model, obtaining an actual answer, introducing the DIKW model to improve the data processing calculation capability and the answer recognition precision, and meanwhile, carrying out self-inspection on the precision of the question and answer DIKW model before answering, so that the accuracy of the question and answer is improved. And accurate and efficient question and answer services can be provided for professionals in the railway field.

Description

Technical Field

[0001] The present invention relates to the field of intelligent question answering technology, and in particular to an intelligent railway question answering management method and system based on AI technology. Background Art

[0002] Against the backdrop of rapid changes in railway technology and continuous breakthrough innovations, the railway equipment field and system application scope are undergoing large-scale transformation and upgrading, and professional practitioners with profound knowledge and technical experts in the railway field are playing an indispensable and important role.

[0003] From the perspective of information management, the traditional model has exposed many limitations. First, in the face of today's massive data resources, traditional information management methods are stretched to the limit and difficult to process and fully utilize them efficiently. Due to the lack of advanced processing mechanisms, the analysis, integration and effective association of data are difficult to carry out smoothly, resulting in data being like a pile of loose sand and unable to play its due value and synergy. Second, traditional information management has obvious shortcomings in search functions. Its search efficiency is low and its accuracy is difficult to be satisfactory. When practitioners are in urgent need of accurate technical information and solutions, traditional management methods are often unable to quickly respond. Respond to needs quickly and accurately, which directly leads to a lot of time cost in dealing with various types of faults, seriously affecting the normal operation and efficient operation and maintenance of the railway system. Thirdly, traditional information management is also plagued by lagging knowledge updates and poor data security. In today's rapidly developing era, knowledge and technology are changing rapidly, but traditional management models cannot keep up in time, resulting in the information provided being outdated and unable to meet actual needs. At the same time, loopholes in data security also make companies lack solid and reliable data support in business management and decision-making processes, just like moving forward in the wind and rain, full of uncertainty and risk. Summary of the invention

[0004] In view of this, the present invention proposes an intelligent railway question and answer management method and system based on AI technology, which can quickly answer users' questions and provide a convenient channel for learning for professionals in the railway field.

[0005] The technical solution of the present invention is achieved in this way: An intelligent railway question-answering management method based on AI technology includes the following steps: Step S1, collecting literature data and question-answer data in the railway field, and constructing a question-answer DIKW model based on the literature data and question-answer data; Step S2: receiving a question-and-answer request initiated by a user, and extracting user information to construct a user DIKW model; Step S3: calling the question-answering DIKW model to process the user DIKW model and obtain a target answer, and obtaining a virtual question based on the target answer; Step S4: input the virtual question into the question-answer DIKW model, process the virtual question with the question-answer DIKW model to obtain a virtual answer, and evaluate the relevance of the virtual answer to the target answer; Step S5: When the correlation between the virtual answer and the target answer is greater than a preset threshold, the question and answer request initiated by the user is input into the question and answer DIKW model, and the actual answer is obtained.

[0006] Preferably, the specific steps of step S1 include: Step S11, collecting professional literature from academic databases, industry standard libraries, enterprise internal knowledge bases, and technical report libraries in the field of railway communications; Step S12: sorting out questions and corresponding answers from consultation records in actual work, question-and-answer sessions in training courses, and hot topics in industry forums, and forming question-and-answer data; Step S13: Map the professional literature and question-answer data into typed resources, and construct a question-answer DIKW model based on the typed resources.

[0007] Preferably, the specific steps of step S2 include: Step S21: receiving a question-and-answer request initiated by a user and identifying the user's identity; Step S22: extract basic information, historical question records, browsing behavior data, and participation project information based on the user's identity, and form user information; Step S23: Map the user information into typed resources, and construct a user DIKW model based on the typed resources.

[0008] Preferably, the typed resources include data resources, information resources, knowledge resources and wisdom resources, and the data resources, information resources, knowledge resources and wisdom resources can be combined and converted with each other.

[0009] Preferably, the specific steps of step S3 include: Step S31, calling the question-answer DIKW model to send a self-check instruction to the user DIKW model, and the user DIKW model processes the data to obtain the user's needs; Step S32: the question-answering DIKW model searches for corresponding documents or question-answering results according to user needs and forms a target answer; Step S33: pre-process the target answer, remove some irrelevant formatting marks and repeated information, extract key features, and construct a virtual question based on the key features.

[0010] Preferably, the specific steps of step S3 also include: Step S34: when the question-and-answer request initiated by the user is a text input, at least one of synonym replacement, word order adjustment, grammatical error addition, or semantic fuzzification processing is performed on the virtual question; Step S35: When the question-and-answer request initiated by the user is a voice input, the nature of the user's work is determined based on the user information, and at least one of background noise simulation, electromagnetic interference simulation, random packet loss simulation or speed change processing is performed in the virtual question based on the nature of the work.

[0011] Preferably, the specific steps of step S4 include: Step S41, inputting the virtual question into the question-answer DIKW model, and extracting keywords by the question-answer DIKW model; Step S42: the question-answer DIKW model searches for documents or question-answer results corresponding to the keywords, and outputs them as virtual answers; Step S43: Evaluate the relevance between the virtual answer and the target answer.

[0012] Preferably, the specific steps of evaluating the relevance between the virtual answer and the target answer in step S43 are: Step S431, using the cosine similarity algorithm to calculate the similarity value between the virtual answer and the target answer; Step S432: extract keywords from the virtual answer and the target answer respectively, perform matching, and calculate the ratio of the number of matching keywords to the total number of keywords; Step S433: assign different weights to the similarity value and the ratio of the number of matching keywords to the total number of keywords, then sum them up and output the sum result as the relevance.

[0013] Preferably, the threshold preset in step S5 is adjusted in real time according to actual conditions.

[0014] An intelligent railway question-answering management system based on AI technology, comprising: A data collection unit is used to collect literature data and question-answer data in the railway field, and to construct a question-answer DIKW model based on the literature data and question-answer data; A request receiving unit, used to receive a question and answer request initiated by a user, and extract user information to construct a user DIKW model; A model calling unit is used to call the question-answering DIKW model to process the user DIKW model and obtain a target answer, and obtain a virtual question based on the target answer; A relevance evaluation unit is used to input a virtual question into the question-answering DIKW model, obtain a virtual answer through processing by the question-answering DIKW model, and evaluate the relevance of the virtual answer to the target answer; The intelligent question-answering unit is used to input the question-answering request initiated by the user into the question-answering DIKW model and obtain the actual answer when the correlation between the virtual answer and the target answer is greater than a preset threshold; The data collection unit is data-connected to the association evaluation unit, the model calling unit is data-connected to the data collection unit, the request receiving unit and the association evaluation unit respectively, and the intelligent question-and-answer unit is data-connected to the data collection unit, the request receiving unit and the association evaluation unit respectively.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides an intelligent railway question-and-answer management method and system based on AI technology. After collecting literature data and question-and-answer data in the railway field, a question-and-answer DIKW model is constructed. The question-and-answer DIKW model can process the question-and-answer request initiated by the user and identify the actual answer. The actual answer includes the corresponding literature or the result that the user wants to obtain. By introducing the DIKW knowledge graph model, massive data can be stored, converted and identified, thereby improving the data processing capability and efficiency, and the actual answer obtained for the question-and-answer request has a high accuracy rate. Before responding to the question and answer request initiated by the user, the present invention also constructs a user DIKW model for the user information, and then processes the user DIKW model through the question and answer DIKW model, and constructs a target answer and a virtual question. The question and answer DIKW model processes the virtual question, and evaluates the correlation between the processed virtual answer and the target answer to achieve self-inspection of the question and answer DIKW model. When the correlation is high, the DIKW model can be used to identify and process the user's question and answer request to ensure that the actual answer obtained can meet the user's requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a flow chart of an intelligent railway question-answering management method based on AI technology of the present invention; Figure 2 This is a flow chart of step S1 of an intelligent railway question-answering management method based on AI technology of the present invention; Figure 3 This is a flow chart of step S2 of an intelligent railway question-answering management method based on AI technology of the present invention; Figure 4This is a flow chart of step S3 of an intelligent railway question-answering management method based on AI technology of the present invention; Figure 5 This is a flow chart of step S4 of an intelligent railway question-answering management method based on AI technology of the present invention; Figure 6 This is a flow chart of step S43 of an intelligent railway question-answering management method based on AI technology of the present invention; Figure 7 A schematic diagram of an intelligent railway question-and-answer management system based on AI technology according to the present invention; In the figure, 1. Data collection unit; 2. Request receiving unit; 3. Model calling unit; 4. Relevance evaluation unit; 5. Intelligent question and answer unit. DETAILED DESCRIPTION

[0018] In order to better understand the technical content of the present invention, a specific embodiment is provided below, and the present invention is further described in conjunction with the accompanying drawings.

[0019] See also Figures 1 to 6 The present invention provides an intelligent railway question-answering management method based on AI technology, comprising the following steps: Step S1, collecting literature data and question-answer data in the railway field, and constructing a question-answer DIKW model based on the literature data and question-answer data; Step S2: receiving a question-and-answer request initiated by a user, and extracting user information to construct a user DIKW model; Step S3: calling the question-answering DIKW model to process the user DIKW model and obtain a target answer, and obtaining a virtual question based on the target answer; Step S4: input the virtual question into the question-answer DIKW model, process the virtual question with the question-answer DIKW model to obtain a virtual answer, and evaluate the relevance of the virtual answer to the target answer; Step S5: When the correlation between the virtual answer and the target answer is greater than a preset threshold, the question and answer request initiated by the user is input into the question and answer DIKW model, and the actual answer is obtained.

[0020] The present invention provides an intelligent railway question and answer management method based on AI technology, which is applied to document retrieval and smart question and answer in the field of railway communications, and is used for railway professionals to search for corresponding documents or some commonly used technical schemes and solutions at any time, so as to improve their professional knowledge. Unlike traditional archives and electronic archives, the present invention combines AI technology to process and identify massive amounts of data, and has high data processing efficiency. The document and solution searches are targeted searches, and the results found will be more in line with user needs, avoiding users from spending too much time searching for some less relevant data from a bunch of irrelevant content.

[0021] The AI ​​core adopted in the present invention is the DIKW knowledge graph model, in which D stands for Data, I stands for Information, K stands for Knowledge, and W stands for Wisdom. The DIKW model can efficiently integrate, properly store and fully utilize knowledge resources, and can also realize flexible conversion and accurate extraction between data, thereby achieving effective storage and high-speed processing of massive data. First, rich literature data and question-and-answer data in the railway field are collected, and the question-and-answer DIKW model is constructed based on this. The question-and-answer DIKW model can not only properly store the collected literature data and question-and-answer data, but also has the ability to decompose these data word by word in detail, deeply interpret the connotation and meaning behind the text, accurately extract keywords from them, and use this to analyze each A document and each question and answer are comprehensively marked, covering important information such as the field to which they belong, the category to which they belong, the corresponding specific railway equipment and relevant departments. When a question and answer request initiated by a user is received, the question and answer DIKW model conducts rapid and accurate identification and efficient processing based on the question and answer request, thereby accurately locating and obtaining the corresponding document or answer, and then integrating the processed document and answer into an actual answer. This actual answer is the accurate response that can meet the user's needs to the greatest extent. By introducing the DIKW knowledge graph model, the entire system can realize rapid data processing. When facing user questions, it can respond in a very short time, and the answers given can achieve the optimal level in terms of accuracy and fit, meet the user's actual needs to the greatest extent, and provide users with efficient, accurate and valuable service experience.

[0022] In addition, after receiving the question and answer request initiated by the user, the present invention will also build a user DIKW model based on the user information. The user DIKW model contains some browsing data of the user, through which some potential needs of the user can be obtained. Thereafter, the question and answer DIKW model can be called to process the user DIKW model, and a target answer is obtained based on the user information. Then, the target answer is reversed to obtain a virtual question corresponding to the target answer. The virtual question is used to self-check the question and answer DIKW model. After the virtual question is input into the question and answer DIKW model, the question and answer DIKW model can search for the data therein based on the virtual question, and obtain the virtual question corresponding to the virtual question after processing and comparison. The virtual answer is generated, and finally the correlation between the virtual answer and the target answer is evaluated. If the correlation is greater than the preset threshold, it means that the recognition and detection accuracy of the question-answering DIKW model is high, and answers that meet user needs can be obtained. Therefore, the question-answering request initiated by the user can be transmitted to the question-answering DIKW model at this time, and the question-answering DIKW model processes it to obtain the final actual answer. The actual answer contains accurate literature or answers, which can be used for users to learn and understand solutions. By introducing the self-check of the question-answering DIKW model, the accuracy of the question-answering DIKW model can be tested before giving an answer. If the accuracy is low, the question-answering DIKW model can be retrained or remodeled to avoid providing users with wrong literature or answers.

[0023] Preferably, the specific steps of step S1 include: Step S11, collecting professional literature from academic databases, industry standard libraries, enterprise internal knowledge bases, and technical report libraries in the field of railway communications; Step S12: sorting out questions and corresponding answers from consultation records in actual work, question-and-answer sessions in training courses, and hot topics in industry forums, and forming question-and-answer data; Step S13: Map the professional literature and question-answer data into typed resources, and construct a question-answer DIKW model based on the typed resources.

[0024] The professional documents obtained from multiple channels include papers, patents, specification documents, engineering case analysis, etc. These professional documents are converted into plain text format, and irrelevant information such as headers, footers, and figure numbers are removed. Only the core text content is retained for subsequent processing. The obtained question and answer data includes questions and answers. The questions and answers are classified and labeled, such as according to communication technology categories (such as wireless communication, wired communication, signal processing, etc.) and problem types (such as principle consultation, troubleshooting, equipment operation, etc.) to facilitate better targeted learning when building the DIKW model. After obtaining professional documents and question and answer data, the document data and question and answer data are mapped into typed resources, and the question and answer DIKW model can be built based on the typed resources.

[0025] Preferably, the specific steps of step S2 include: Step S21: receiving a question-and-answer request initiated by a user and identifying the user's identity; Step S22: extract basic information, historical question records, browsing behavior data, and participation project information based on the user's identity, and form user information; Step S23: Map the user information into typed resources, and construct a user DIKW model based on the typed resources.

[0026] After the Q&A DIKW model is built, it can be used for backup. When a user initiates a Q&A request, the Q&A request will not be sent to the Q&A DIKW model for literature or answer search at the first time. First, the identity of the user who initiated the Q&A request is identified, and then the basic information and some behavior data are extracted based on the user's identity. The basic information includes the user's name, department, position, etc. For example, when the user is an engineer in the railway communication department, his daily work may focus more on technology research and development and troubleshooting; the historical question record includes all the questions raised by the user in the system in the past, and determines the subject area of ​​his concern. For example, it is found that the user has asked questions about railway communication signal interference many times, which indicates that he has a continuous knowledge demand in this field; browsing behavior data refers to the type of literature, subject, and residence time of the user browsed by the system. For example, the user often browses about 5G Technical reports on railway communication applications, with a long stay time, indicate that this is their potential interest direction; and project participation information includes the railway communication projects in which the user is involved. For example, if they are participating in a high-speed railway communication network upgrade project, then related network architecture optimization, equipment compatibility and other issues may be their potential demand points; after unifying the collected information into user information and mapping it into typed resources, a user DIKW model can be constructed based on the typed resources.

[0027] Preferably, the typed resources include data resources, information resources, knowledge resources and wisdom resources, and the data resources, information resources, knowledge resources and wisdom resources can be combined and converted with each other.

[0028] Data resources, information resources, knowledge resources and wisdom resources can be combined and converted with each other, so that data information is not limited to the literal meaning, but can be extended to dig deeper into the content, and adaptive adjustments can be made quickly when facing the ever-changing internal and external environment.

[0029] Preferably, the specific steps of step S3 include: Step S31, calling the question-answer DIKW model to send a self-check instruction to the user DIKW model, and the user DIKW model processes the data to obtain the user's needs; Step S32: the question-answering DIKW model searches for corresponding documents or question-answering results according to user needs and forms a target answer; Step S33: pre-process the target answer, remove some irrelevant formatting marks and repeated information, extract key features, and construct a virtual question based on the key features; Step S34: when the question-and-answer request initiated by the user is a text input, at least one of synonym replacement, word order adjustment, grammatical error addition, or semantic fuzzification processing is performed on the virtual question; Step S35: When the question-and-answer request initiated by the user is a voice input, the nature of the user's work is determined based on the user information, and at least one of background noise simulation, electromagnetic interference simulation, random packet loss simulation or speed change processing is performed in the virtual question based on the nature of the work.

[0030] After the user DIKW model is built, the question-answer DIKW model needs to be self-checked. The question-answer DIKW model can send self-check instructions to the user DIKW model. After receiving the self-check instructions, the user DIKW model will process its internal data to obtain user needs. The user needs will be sent to the question-answer DIKW model for the question-answer DIKW model to process and search its internal data, so as to find the corresponding literature or question-answer results and form the target answer. Since the target answer is obtained based on user needs, the user needs and the question-answer requests initiated by the user may be very similar in field. Therefore, the target answer obtained based on the user needs can perform the first test on the question-answer DIKW model. After obtaining the target answer, the target answer needs to be preprocessed to remove some noise and extract key features. , which can reduce the subsequent amount of calculations, and obtain the virtual question based on the target answer. In order to further simulate the real questioning scenario, the virtual question needs to be further processed. There are different processing methods based on the different ways of asking virtual questions. When the virtual question is asked in text, the words and sentences of the virtual question can be changed, the order can be changed, etc., to increase the difficulty of identifying the virtual question. When the virtual question is voice input, the nature of the user's work is first judged according to the user information. For example, when the user works in a railway station or a computer room, background noise simulation can be added to the virtual question. If the user is in an environment that is easily disturbed for a long time, electromagnetic interference can be added to the virtual question. If the nature of the user's work does not have too much interference, the virtual question can be subjected to random packet loss simulation or speed change processing, etc., to further increase the difficulty of identifying the virtual question.

[0031] Preferably, the specific steps of step S4 include: Step S41, inputting the virtual question into the question-answer DIKW model, and extracting keywords by the question-answer DIKW model; Step S42: the question-answer DIKW model searches for documents or question-answer results corresponding to the keywords, and outputs them as virtual answers; Step S43: Evaluate the relevance between the virtual answer and the target answer. The specific steps are: Step S431, using the cosine similarity algorithm to calculate the similarity value between the virtual answer and the target answer; Step S432: extract keywords from the virtual answer and the target answer respectively, perform matching, and calculate the ratio of the number of matching keywords to the total number of keywords; Step S433: assign different weights to the similarity value and the ratio of the number of matching keywords to the total number of keywords, then sum them up and output the sum result as the relevance.

[0032] After obtaining the processed virtual question, the virtual question can be input into the question-answer DIKW model, which extracts keywords. The question-answer DIKW model can then search for keywords in all the documents within it or in the question-answer results. After finding the corresponding content, it can be output as a virtual answer. Then, it is necessary to evaluate the correlation between the virtual answer and the target answer. The present invention adopts a combination of similarity evaluation and keyword matching when evaluating the correlation. The similarity value can be calculated using the cosine similarity algorithm. In addition, by matching the keywords in the virtual answer and the target answer, the corresponding ratio can be obtained. Then, the similarity and ratio are assigned values ​​respectively, and the values ​​are unified into a data range and then summed. Finally, the summed result is output as the correlation.

[0033] The correlation degree is compared with the preset threshold. If the correlation degree is greater than the threshold, it means that the recognition accuracy of the question-answering DIKW model is high. At this time, the question-answering request initiated by the user can be input into the question-answering DIKW model to obtain the actual answer.

[0034] Preferably, the threshold preset in step S5 is adjusted in real time according to actual conditions.

[0035] The preset threshold value can be adjusted according to actual conditions to suit different usage scenarios.

[0036] Reference Figure 7 An intelligent railway question-answering management system based on AI technology is shown, comprising: Data collection unit 1, used for collecting literature data and question-answer data in the railway field, and building a question-answer DIKW model based on the literature data and question-answer data; Request receiving unit 2, used to receive the question and answer request initiated by the user, and extract the user information to build the user DIKW model; The model calling unit 3 is used to call the question-answering DIKW model to process the user DIKW model and obtain a target answer, and obtain a virtual question based on the target answer; The relevance evaluation unit 4 is used to input the virtual question into the question-answering DIKW model, obtain the virtual answer through the question-answering DIKW model, and evaluate the relevance between the virtual answer and the target answer; The intelligent question-answering unit 5 is used to input the question-answering request initiated by the user into the question-answering DIKW model and obtain the actual answer when the correlation between the virtual answer and the target answer is greater than a preset threshold; The data collection unit 1 is data-connected to the correlation evaluation unit 4, the model calling unit 3 is data-connected to the data collection unit 1, the request receiving unit 2 and the correlation evaluation unit 4 respectively, and the intelligent question-and-answer unit 5 is data-connected to the data collection unit 1, the request receiving unit 2 and the correlation evaluation unit 4 respectively.

[0037] The data collection unit 1 is used to collect document data and question and answer data, and then build a question and answer DIKW model, while the request receiving unit 2 builds a user DIKW model based on user information. After obtaining the above two DIKW models, the model calling unit 3 calls the question and answer DIKW model in the data collection unit 1 to process the user DIKW model, thereby obtaining the target answer and the virtual question. Finally, the relevance evaluation unit 4 inputs the virtual question into the question and answer DIKW model to obtain the virtual answer and evaluate the relevance between the virtual answer and the target answer. When the relevance is greater than the preset threshold, the intelligent question and answer unit 5 can input the question and answer request initiated by the user into the question and answer DIKW model and obtain the actual answer. The actual answer contains the document required by the user or the solution desired by the user. After self-checking the question and answer DIKW model, its data processing and recognition capabilities can be determined. If the recognition accuracy is low, the question and answer DIKW model needs to be reconstructed or trained to avoid providing inappropriate documents or answers to the user.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent railway question-answering management method based on AI technology, characterized in that: The following steps are involved: Step S1, collecting literature data and question-answer data in the railway field, and constructing a question-answer DIKW model based on the literature data and question-answer data; Step S2: receiving a question-and-answer request initiated by a user, and extracting user information to construct a user DIKW model; Step S3: calling the question-answering DIKW model to process the user DIKW model and obtain a target answer, and obtaining a virtual question based on the target answer; Step S4: input the virtual question into the question-answer DIKW model, process the virtual question with the question-answer DIKW model to obtain a virtual answer, and evaluate the relevance of the virtual answer to the target answer; Step S5: When the correlation between the virtual answer and the target answer is greater than a preset threshold, the question and answer request initiated by the user is input into the question and answer DIKW model, and the actual answer is obtained.

2. According to the AI ​​technology-based intelligent railway question-answering management method of claim 1, it is characterized in that: The specific steps of step S1 include: Step S11, collecting professional literature from academic databases, industry standard libraries, enterprise internal knowledge bases, and technical report libraries in the field of railway communications; Step S12: sorting out questions and corresponding answers from consultation records in actual work, question-and-answer sessions in training courses, and hot topics in industry forums, and forming question-and-answer data; Step S13: Map the professional literature and question-answer data into typed resources, and construct a question-answer DIKW model based on the typed resources.

3. According to the AI ​​technology-based intelligent railway question-answering management method of claim 1, it is characterized in that: The specific steps of step S2 include: Step S21: receiving a question-and-answer request initiated by a user and identifying the user's identity; Step S22: extract basic information, historical question records, browsing behavior data, and participation project information based on the user's identity, and form user information; Step S23: Map the user information into typed resources, and construct a user DIKW model based on the typed resources.

4. According to claim 2 or 3, an intelligent railway question-answering management method based on AI technology is characterized in that: The typed resources include data resources, information resources, knowledge resources and wisdom resources, and the data resources, information resources, knowledge resources and wisdom resources can be combined and converted with each other.

5. According to the AI-based intelligent railway question-answering management method of claim 1, it is characterized in that: The specific steps of step S3 include: Step S31, calling the question-answer DIKW model to send a self-check instruction to the user DIKW model, and the user DIKW model processes the data to obtain the user's needs; Step S32: the question-answering DIKW model searches for corresponding documents or question-answering results according to user needs and forms a target answer; Step S33: pre-process the target answer, remove some irrelevant formatting marks and repeated information, extract key features, and construct a virtual question based on the key features.

6. The intelligent railway question-answering management method based on AI technology according to claim 5 is characterized in that: The specific steps of step S3 also include: Step S34: when the question-and-answer request initiated by the user is a text input, at least one of synonym replacement, word order adjustment, grammatical error addition, or semantic fuzzification processing is performed on the virtual question; Step S35: When the question-and-answer request initiated by the user is a voice input, the nature of the user's work is determined based on the user information, and at least one of background noise simulation, electromagnetic interference simulation, random packet loss simulation or speed change processing is performed in the virtual question based on the nature of the work.

7. The intelligent railway question-answering management method based on AI technology according to claim 1 is characterized in that: The specific steps of step S4 include: Step S41, inputting the virtual question into the question-answer DIKW model, and extracting keywords by the question-answer DIKW model; Step S42: the question-answer DIKW model searches for documents or question-answer results corresponding to the keywords, and outputs them as virtual answers; Step S43: Evaluate the relevance between the virtual answer and the target answer.

8. The intelligent railway question-answering management method based on AI technology according to claim 7 is characterized in that: The specific steps of step S43 for evaluating the relevance between the virtual answer and the target answer are: Step S431, using the cosine similarity algorithm to calculate the similarity value between the virtual answer and the target answer; Step S432: extract keywords from the virtual answer and the target answer respectively, perform matching, and calculate the ratio of the number of matching keywords to the total number of keywords; Step S433: assign different weights to the similarity value and the ratio of the number of matching keywords to the total number of keywords, then sum them up and output the sum result as the relevance.

9. The intelligent railway question-answering management method based on AI technology according to claim 1 is characterized in that: The threshold value preset in step S5 is adjusted in real time according to the actual situation.

10. A system using the intelligent railway question-answering management method based on AI technology as described in any one of claims 1 to 9, characterized in that: include: A data collection unit is used to collect literature data and question-answer data in the railway field, and to construct a question-answer DIKW model based on the literature data and question-answer data; A request receiving unit, used to receive a question and answer request initiated by a user, and extract user information to construct a user DIKW model; A model calling unit is used to call the question-answering DIKW model to process the user DIKW model and obtain a target answer, and obtain a virtual question based on the target answer; A relevance evaluation unit is used to input a virtual question into the question-answering DIKW model, obtain a virtual answer through processing by the question-answering DIKW model, and evaluate the relevance of the virtual answer to the target answer; The intelligent question-answering unit is used to input the question-answering request initiated by the user into the question-answering DIKW model and obtain the actual answer when the correlation between the virtual answer and the target answer is greater than a preset threshold; The data collection unit is data-connected to the association evaluation unit, the model calling unit is data-connected to the data collection unit, the request receiving unit and the association evaluation unit respectively, and the intelligent question-and-answer unit is data-connected to the data collection unit, the request receiving unit and the association evaluation unit respectively.

Images