Elevator information query method and system

Abstract

The invention discloses an elevator information query method and system, and belongs to elevator system design. According to the system, a natural language problem of a user is converted into structured semantic representation through a semantic input preprocessing and intention recognition module, and intention recognition and entity extraction are carried out on the structured semantic representation through a deep learning model. Based on the structured semantic representation, the system performs efficient query in a knowledge graph related to the elevator through a knowledge graph query module, and returns related nodes and paths matched with user requirements. The multi-modal matching engine further generates candidate answers according to the query result, and performs fusion processing on multi-modal data such as texts and images through a deep learning model so as to provide comprehensive query answers. And the dynamic reply generation module is used for generating a context-related natural language reply in combination with historical dialogue records of the user, and carrying out compliance verification. And the user feedback processing module is used for driving the AI model to optimize by analyzing the user behavior data, so that the continuous improvement of the system performance is realized. The closed loop iterative optimization module further adjusts the AI model and the knowledge graph based on user feedback.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of elevator system design, and in particular to an elevator information query method and system. Background Art

[0002] With the advancement of smart buildings and smart cities, elevators, as crucial vertical transportation equipment within buildings, are gaining increasing attention. Modern elevators not only fulfill basic transportation functions but also address multiple aspects, such as energy conservation, intelligence, and convenience. However, with the increasing diversity and complexity of elevators, users may encounter various inquiries during use, such as information on the elevator's load capacity, speed, operating status, brand and model, and installation requirements. Rapid and accurate access to this information is crucial for elevator operations management and user experience.

[0003] Currently, most elevator information search methods rely on manual queries or static databases. Users rely on manual consultation or elevator manuals to obtain relevant information, resulting in low query efficiency and often lacking personalization and precision. Furthermore, existing query methods struggle to meet users' actual needs for real-time, accurate, and diverse information. In some complex query scenarios, manual queries are not only inefficient but also prone to errors, failing to fully address the diverse user needs and ever-changing elevator information. Summary of the Invention

[0004] The embodiment of the present application provides an elevator information query method and system. The technical solution is as follows.

[0005] According to one aspect of the present application, there is provided an elevator information query system, the system comprising:

[0006] The semantic input preprocessing and intent recognition module receives natural language questions input by users and generates a structured semantic representation containing intent labels, entity lists, and confidence levels through an artificial intelligence model;

[0007] A knowledge graph query module is used to search for corresponding nodes and their relationship paths in the knowledge graph based on the intent labels and entity lists, and return matching subgraph results using a graph neural network optimization algorithm;

[0008] A multimodal matching engine, which generates candidate answers based on the knowledge graph subgraph results, including text answers, confidence levels, and source types, and performs multimodal data fusion through artificial intelligence models;

[0009] A dynamic response generation module is used to generate compliant natural language response text based on the candidate answers and historical conversation records, and optimize the tone and style of the response through artificial intelligence algorithms;

[0010] User feedback processing module, which receives user behavior and drives the optimization of artificial intelligence models through deep learning models or reinforcement learning methods, updating model parameters and knowledge graphs;

[0011] The closed-loop iterative optimization module is used to adjust the artificial intelligence model and knowledge graph based on user feedback, and to iteratively optimize system performance throughout the entire process through machine learning algorithms.

[0012] Optionally, the semantic input preprocessing and intent recognition module includes:

[0013] A natural language processing unit, which uses a deep learning model to convert natural language questions into structured semantic representations;

[0014] Intent recognition unit, used to identify user intent and label the intent through artificial intelligence classification models;

[0015] Entity extraction unit, used to extract and construct entity lists from questions using named entity recognition technology;

[0016] A confidence assessment unit is used to use an artificial intelligence model to evaluate the accuracy of the extracted intent labels and entity lists and generate a confidence value.

[0017] Optionally, the knowledge graph query module includes:

[0018] A knowledge graph storage unit is used to store a knowledge graph containing the elevator's technical specifications, capacity, brand, model, maximum load, operating speed, number of floors, drive mode, door type, usage scenarios, and installation requirements;

[0019] The matching algorithm unit is used to search for matching nodes and relationship paths in the knowledge graph based on the intent label and entity list through the graph neural network, and return the matching results.

[0020] Optionally, the multimodal matching engine includes:

[0021] The text generation unit is used to extract and generate candidate answers in text form from the knowledge graph query results;

[0022] A confidence scoring unit, used to score the confidence of each candidate answer based on the deep learning model;

[0023] A source identification unit is used to identify the source type of each candidate answer, where the source type includes text, image, and audio, wherein the image and audio are processed using computer vision and speech recognition models.

[0024] Optionally, the dynamic reply generation module includes:

[0025] A historical conversation storage unit is used to store the user's historical conversation records;

[0026] A context-aware engine, which uses an LSTM algorithm to generate context-relevant natural language responses based on the context and historical conversations.

[0027] The compliance verification unit is used to check whether the generated responses meet the output requirements based on the artificial intelligence compliance checking model.

[0028] Optionally, the user feedback processing module includes:

[0029] User behavior tracking unit, used to record and analyze user interaction with the system, including click counts and dwell time, and conduct feedback analysis through deep learning models;

[0030] The implicit feedback analysis unit is used to analyze user behavior to generate feedback signals and optimize the artificial intelligence model used based on the signals.

[0031] Optionally, the closed-loop iterative optimization module includes:

[0032] A model optimization unit, which is used to adjust the parameters of the AI ​​model based on user feedback signals and optimize the performance of the AI ​​model using automatic machine learning technology;

[0033] The knowledge updating unit is used to update the elevator information in the knowledge graph based on feedback.

[0034] On the other hand, a method for querying elevator information is provided, characterized in that the method includes:

[0035] Receive natural language questions input by users and generate structured semantic representations;

[0036] Searching for matching nodes and relationship paths in the knowledge graph according to the structured semantic representation, and returning matching subgraph results;

[0037] Generate multiple candidate answers and assign confidence scores to them based on AI algorithms;

[0038] Generate context-sensitive natural language responses based on historical conversation records and perform compliance verification through AI compliance models;

[0039] Optimize AI models and knowledge graphs based on user feedback;

[0040] Optimize the performance of the query system through a closed-loop iterative process, and use algorithms such as reinforcement learning to continuously improve the system's response speed and accuracy.

[0041] Optionally, receiving a natural language question input by a user and generating a structured semantic representation includes:

[0042] User questions generate intent labels and entity lists through intent recognition and entity extraction;

[0043] Confidence evaluation is performed on the generated intent labels and entity lists to generate structured semantic representations.

[0044] Optionally, searching for matching nodes and relationship paths in the knowledge graph according to the structured semantic representation and returning matching subgraph results includes:

[0045] Find nodes and their relationship paths in the knowledge graph that match the intent label and entity list;

[0046] Returns the matching knowledge graph subgraph results. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure is a flowchart of an elevator information query method provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0049] In this document, "plurality" refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0050] Example 1

[0051] An elevator information query system is provided, comprising:

[0052] Semantic input preprocessing and intent recognition module, used to receive natural language questions input by users, and generate structured semantic representations including intent labels, entity lists and confidence levels through artificial intelligence models; knowledge graph query module, used to search for corresponding nodes and their relationship paths in the knowledge graph based on the intent labels and entity lists, and return matching subgraph results using graph neural network optimization algorithms; multimodal matching engine, used to generate candidate answers based on the knowledge graph subgraph results, including text answers, confidence levels and source types, and perform multimodal data fusion through artificial intelligence models; dynamic reply generation module, used to generate compliant natural language reply text based on the candidate answers and historical conversation records, and optimize the tone and style of the reply through artificial intelligence algorithms; user feedback processing module, used to receive user behavior, and drive the optimization of artificial intelligence models through deep learning models or reinforcement learning methods, and update model parameters and knowledge graphs; closed-loop iterative optimization module, used to adjust artificial intelligence models and knowledge graphs based on user feedback, and perform iterative optimization of system performance throughout the entire process through machine learning algorithms.

[0053] This elevator information query system combines artificial intelligence (AI) technology with knowledge graphs to provide an efficient, accurate, and intelligent query method. The core of the system lies in its semantic input preprocessing and intent recognition module, which employs a deep learning model to process natural language questions entered by users. When a user asks a question like "How many people can this elevator carry?", the system uses natural language processing (NLP) technology to segment and identify the intent of the question. It then uses "elevator" as an entity and "load" as an intent label to generate a structured semantic representation. This structured data is then passed to the knowledge graph query module, which uses a graph neural network (GNN) to query nodes and paths related to "load" in the knowledge graph to ensure accurate query results. Based on this, a multimodal matching engine generates candidate answers. These answers not only include text answers but also utilize AI models to fuse multimodal data, combining images, audio, and other information to provide comprehensive query results. For historical queries, the system uses a context-aware engine to generate more contextually appropriate natural language responses. It also uses reinforcement learning algorithms to self-adjust and optimize the model, adjusting query accuracy based on user behavior feedback (such as clicks and dwell time).

[0054] This technology integrates multiple AI technologies into the elevator query system, providing a self-optimizing query method. This system not only responds quickly to user queries but also continuously improves its accuracy and intelligence through real-time feedback, significantly enhancing the user experience. Through the dynamic optimization of AI models, elevator information queries are no longer simply static queries; instead, they become a self-adapting, continuously improving intelligent query process.

[0055] Example 2

[0056] Optionally, the semantic input preprocessing and intent recognition module includes:

[0057] A natural language processing unit, which uses a deep learning model to convert natural language questions into structured semantic representations;

[0058] Intent recognition unit, used to identify user intent and label the intent through artificial intelligence classification models;

[0059] Entity extraction unit, used to extract and construct entity lists from questions using named entity recognition technology;

[0060] A confidence assessment unit is used to use an artificial intelligence model to evaluate the accuracy of the extracted intent labels and entity lists and generate a confidence value.

[0061] In the semantic input preprocessing and intent recognition module, the system first receives the user's natural language input, such as "What is the maximum load of the elevator?" and relies on deep learning models for semantic processing. The natural language processing unit converts the user input into a structured semantic representation through models such as BERT and GPT. In this process, the intent recognition unit uses AI classification models (such as convolutional neural networks, long short-term memory networks, etc.) to identify the intent of the question, thereby extracting the user's query intention, such as "query the elevator load." At the same time, the system extracts a list of entities, such as "elevator" and "load," through named entity recognition technology (NER). In order to ensure the accuracy of recognition, the confidence assessment unit evaluates the extracted labels and entities through a deep neural network to generate a confidence value that reflects the accuracy of the recognition result.

[0062] This module's core technology, through the integration of deep learning and natural language processing, not only accurately recognizes various forms of user input but also improves the system's fault tolerance and recognition accuracy through a confidence assessment mechanism. By deeply understanding question intent and entities, the system effectively interprets user needs and provides accurate input for subsequent knowledge graph queries.

[0063] Example 3

[0064] Optionally, the knowledge graph query module includes:

[0065] A knowledge graph storage unit is used to store a knowledge graph containing the elevator's technical specifications, capacity, brand, model, maximum load, operating speed, number of floors, drive mode, door type, usage scenarios, and installation requirements;

[0066] The matching algorithm unit is used to search for matching nodes and relationship paths in the knowledge graph based on the intent label and entity list through the graph neural network, and return the matching results.

[0067] In the knowledge graph query module, the system uses a graph neural network (GNN) algorithm to deeply mine elevator-related knowledge to ensure the accuracy of user query results. The knowledge graph storage unit integrates elevator data, including detailed information such as elevator technical specifications, capacity, brand, model, maximum load, operating speed, and number of floors, into a graph structure. When a user asks a question such as "What is the maximum load of an elevator?", the system first matches the question using intent tags and entity lists to locate knowledge nodes related to "elevator" and "maximum load." The matching algorithm unit uses a graph neural network to optimize queries against these nodes, using intelligent algorithms to improve the accuracy of relationship path matching between nodes in the graph, ensuring that the most relevant subgraph data is returned.

[0068] By incorporating graph neural networks, this module enables elevator information queries to move beyond the retrieval of single data points to the intelligent analysis of nodes and relationships across the entire graph. Graph neural networks improve query accuracy and efficiency by leveraging the relationships between nodes in the graph, enabling rapid matching of multi-dimensional elevator information.

[0069] Example 4

[0070] Optionally, the multimodal matching engine includes:

[0071] The text generation unit is used to extract and generate candidate answers in text form from the knowledge graph query results;

[0072] A confidence scoring unit, used to score the confidence of each candidate answer based on the deep learning model;

[0073] A source identification unit is used to identify the source type of each candidate answer, where the source type includes text, image, and audio, wherein the image and audio are processed using computer vision and speech recognition models.

[0074] The multimodal matching engine is an important component of this system. It uses AI algorithms to fuse multimodal data such as text, images, and audio to provide richer query results. When users query elevator information, in addition to text descriptions, they may also need to see the appearance of the elevator or hear an audio description. For example, when a user asks "What does the elevator look like?", the system not only extracts a description of the elevator's appearance through the text generation unit, but also extracts an image of the elevator through a computer vision algorithm, and combines it with the text description to generate a multimodal answer containing text and images. The confidence scoring unit scores each candidate answer based on its credibility and ultimately generates a suitable answer. The source identification unit clearly identifies the source type of each candidate answer, including text, image, or audio.

[0075] By integrating multiple data types, the system provides more than just a single text answer; instead, it provides a more comprehensive output, enabling users to gain a multi-dimensional perspective when answering queries. This feature significantly improves the user experience and query accuracy, especially when dealing with complex questions, as multimodal information provides more context and intuitive feedback.

[0076] Example 5

[0077] Optionally, the dynamic reply generation module includes a historical conversation storage unit for storing the user's historical conversation records; a context-aware engine for generating context-related natural language replies based on the context and historical conversations through an LSTM algorithm; and a compliance verification unit for checking whether the generated replies meet the output requirements based on an artificial intelligence compliance check model.

[0078] The dynamic response generation module is based on a context-aware engine and uses algorithms such as LSTM to generate natural language responses that are consistent with the context. After a user asks a question, the system not only relies on the current query content, but also refers to historical conversation records to generate a coherent response. For example, if a user first asks "What is the maximum load of the elevator?" and then asks "What is the rated load of the elevator?", the system can recognize that "rated load" and "maximum load" are different parameters of the same elevator and generate an appropriate response based on the historical conversation. This module also includes a compliance verification unit to ensure that the generated responses comply with relevant laws and regulations. In particular, when it comes to sensitive topics such as security and privacy, the system will automatically verify the compliance of the response content.

[0079] This module ensures the consistency and legality of system responses through context understanding and compliance verification technology, improving the adaptability to user queries and the accuracy of system responses.

[0080] Example 6

[0081] Optionally, the user feedback processing module includes a user behavior tracking unit, which is used to record and analyze the user's interaction behavior with the system, including the number of clicks and the length of stay, and perform feedback analysis through a deep learning model; an implicit feedback analysis unit, which is used to analyze user behavior to generate a feedback signal, and optimize the artificial intelligence model used based on the signal.

[0082] The user feedback processing module uses deep learning and reinforcement learning technologies to analyze user interactions with the system in real time and optimize the system based on user feedback. Whenever a user interacts with a query result, the system records relevant data, such as the number of clicks and duration of stay, and analyzes the user's feedback using a deep learning model. For example, if a user quickly leaves the page after clicking on an answer, the system may infer that the answer does not meet the user's needs and automatically adjust the query strategy or model parameters. The implicit feedback analysis unit generates feedback signals based on user behavior to optimize the AI ​​model's predictive capabilities and the accuracy of query results.

[0083] This module intelligently analyzes user behavior to ensure that the system can adjust query strategies in real time based on feedback, thereby continuously improving the performance and responsiveness of the elevator information query system.

[0084] Example 7

[0085] Optionally, the closed-loop iterative optimization module includes a model optimization unit, which is used to adjust the parameters of the artificial intelligence model according to user feedback signals and optimize the performance of the artificial intelligence model using automatic machine learning technology; and a knowledge updating unit, which is used to update the elevator information in the knowledge graph according to feedback.

[0086] The closed-loop iterative optimization module utilizes reinforcement learning and automated machine learning (AutoML) technologies to optimize the AI ​​model. Whenever the system receives user feedback, the model optimization unit adjusts the AI ​​model's parameters based on the feedback signal and automatically tunes the model using AutoML technology. Furthermore, the knowledge graph update unit automatically updates the elevator information in the graph based on user query data and feedback, ensuring that the system always maintains the latest elevator technical information. Through this closed-loop iterative optimization process, the elevator information query system is able to continuously learn and self-optimize, improving query efficiency and accuracy.

[0087] This module uses an AI-driven self-optimization mechanism to ensure that the system can continuously improve its performance over time and adapt to the query needs of different users.

[0088] Example 8

[0089] An elevator information query method is provided, characterized in that the method comprises:

[0090] Step 201: Receive a natural language question input by a user and generate a structured semantic representation.

[0091] In one possible implementation, user questions generate intent labels and entity lists through intent recognition and entity extraction; confidence evaluation is performed on the generated intent labels and entity lists to generate structured semantic representations.

[0092] This step converts the user's natural language questions into structured semantic representations through natural language processing (NLP) technology. The query question entered by the user (such as "What is the maximum load of the elevator?") is subjected to intent recognition and entity extraction to generate a structured representation with intent labels and entity lists. Intent recognition uses deep learning models (such as BERT, GPT, etc.) to identify the user's query intent (for example, "query the elevator load"). Entity extraction uses named entity recognition (NER) technology to extract key entity information (such as "elevator" and "load") to ensure that important elements in the question are effectively captured. Finally, the confidence assessment module evaluates the accuracy of intent and entities, and generates a confidence value for the query. This structured semantic representation can be accurately passed to downstream modules to ensure that the system can understand the user's true intentions and provide fast and accurate responses.

[0093] By processing complex user queries through deep learning models, the system can deeply understand the syntax and semantics in natural language, going beyond traditional simple rule-based matching.

[0094] A confidence assessment module is added to dynamically evaluate the accuracy of intent and entity extraction to ensure that the system can provide reliable responses in different scenarios.

[0095] Step 202: Search for matching nodes and relationship paths in the knowledge graph according to the structured semantic representation, and return matching subgraph results.

[0096] In one possible implementation, nodes and their relationship paths that match the intent labels and entity lists are searched in the knowledge graph; and matching knowledge graph subgraph results are returned.

[0097] Step 203: Generate multiple candidate answers and assign confidence scores to them based on an artificial intelligence algorithm.

[0098] In one possible implementation, the system generates multiple candidate answers based on the query results of the knowledge graph subgraph. These candidate answers include not only text answers, but also other possible forms of information, such as images or audio. Each candidate answer is scored with a confidence score using a deep learning model to assess its relevance and accuracy to the user's query. For example, for the question "maximum load capacity of an elevator", candidate answers may include technical specifications, images, or videos of the elevator. The system uses multimodal data fusion technology to combine text, images, audio and other information to generate the best candidate answers, and selects the answers that best meet the user's needs based on the confidence scoring mechanism.

[0099] By integrating multimodal data such as text, images, and audio, it provides richer and more intuitive answers than single-text queries. This multimodal matching engine can greatly improve the comprehensiveness of queries and user experience.

[0100] A deep learning model is used to assign a confidence score to each candidate answer, enabling the system to make accurate choices between multiple answers, thereby ensuring the reliability and relevance of the final answer.

[0101] Step 204: Generate context-related natural language responses based on historical conversation records and perform compliance verification through an artificial intelligence compliance model.

[0102] The dynamic response generation module relies on a context-aware engine to generate natural language responses that are consistent with the current state of the conversation. By analyzing historical conversation records and the user's current query, the system generates natural language responses based on algorithms such as LSTM (long short-term memory network) to ensure that they conform to contextual logic and user needs. For example, if the user previously asked about the weight of the elevator, the system will generate a coherent natural language response that conforms to the conversation context based on this historical record and the current question (such as the speed of the elevator). In addition, the compliance verification unit uses an artificial intelligence compliance check model to ensure that the generated responses meet the output requirements, especially when it comes to sensitive topics, to ensure their compliance and accuracy.

[0103] By leveraging models like LSTM to contextualize historical conversations, each response seamlessly connects to previous conversations, improving the system's conversational coherence and intelligence. Automatic compliance verification technology ensures that generated responses meet legal, ethical, and technical standards, making it particularly useful for inquiries involving user privacy or security.

[0104] Step 205: Optimize the artificial intelligence model and knowledge graph based on user feedback.

[0105] The user feedback processing module further optimizes the AI ​​model by recording user interaction behaviors (such as clicks, dwell time, etc.) and user satisfaction with query results. The system analyzes feedback signals through deep learning models or reinforcement learning methods and adjusts the parameters of the AI ​​model based on this feedback. Feedback signals can be explicit (such as ratings, comments) or implicit (such as interaction time, click-through rate, etc.). The system will perform appropriate optimization based on different types of feedback, allowing the model to continuously improve its response accuracy and efficiency. At the same time, user feedback will also drive the update of the knowledge graph, ensuring that the elevator information in the graph remains up to date.

[0106] By combining explicit and implicit feedback, the system analyzes user needs and system performance from multiple perspectives, providing accurate data support for model optimization. Using reinforcement learning algorithms, the system automatically adjusts its model and knowledge graph based on real-time feedback, enabling the elevator information query system to continuously improve itself over time, enhancing query efficiency and accuracy.

[0107] Step 206 , optimize the performance of the query system through a closed-loop iterative process, and continuously improve the system response speed and accuracy using algorithms such as reinforcement learning.

[0108] This method encompasses the entire query process, from receiving a user's question to ultimately returning an answer. First, the system transforms the user's natural language question into a structured semantic representation using intent recognition and entity extraction techniques. Then, based on this structured representation, the system searches for matching nodes in the knowledge graph and returns relevant subgraph results. After generating multiple candidate answers, the system uses a deep learning model to score the confidence of each answer and select the one that best matches the user's query. Finally, the system generates a personalized natural language response based on historical conversations and context, and verifies this response using an AI compliance model.

[0109] Through this method, the system can quickly and accurately provide users with elevator-related information, and continuously optimize the query process based on user feedback, thereby improving the intelligence level of the query system.

[0110] In summary, this invention significantly enhances the intelligence and adaptability of the elevator information query system by integrating multiple advanced artificial intelligence technologies, such as deep learning, reinforcement learning, graph neural networks, and natural language processing. The system not only rapidly parses user questions and extracts relevant data from the knowledge graph, but also enhances query accuracy and user experience through the fusion of multimodal information and dynamic response generation. Through a feedback optimization mechanism based on user behavior, the system adjusts models and query strategies in real time, achieving continuous self-learning and optimization, ensuring that query results consistently meet user needs. This provides an intelligent, efficient, and personalized elevator information query solution.

[0111] The above description is merely an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. An elevator information query system, characterized in that: The system comprises: The semantic input preprocessing and intent recognition module receives natural language questions input by users and generates a structured semantic representation containing intent labels, entity lists, and confidence levels through an artificial intelligence model; A knowledge graph query module is used to search for corresponding nodes and their relationship paths in the knowledge graph based on the intent labels and entity lists, and return matching subgraph results using a graph neural network optimization algorithm; A multimodal matching engine, which generates candidate answers based on the knowledge graph subgraph results, including text answers, confidence levels, and source types, and performs multimodal data fusion through artificial intelligence models; A dynamic response generation module is used to generate compliant natural language response text based on the candidate answers and historical conversation records, and optimize the tone and style of the response through artificial intelligence algorithms; User feedback processing module, which receives user behavior and drives the optimization of artificial intelligence models through deep learning models or reinforcement learning methods, updating model parameters and knowledge graphs; The closed-loop iterative optimization module is used to adjust the artificial intelligence model and knowledge graph based on user feedback, and to iteratively optimize system performance throughout the entire process through machine learning algorithms.

2. The elevator information query system according to claim 1, characterized in that: The semantic input preprocessing and intention recognition module includes: A natural language processing unit, which uses a deep learning model to convert natural language questions into structured semantic representations; Intent recognition unit, used to identify user intent and label the intent through artificial intelligence classification models; Entity extraction unit, used to extract and construct entity lists from questions using named entity recognition technology; A confidence assessment unit is used to use an artificial intelligence model to evaluate the accuracy of the extracted intent labels and entity lists and generate a confidence value.

3. The elevator information query system according to claim 1, characterized in that: The knowledge graph query module includes: A knowledge graph storage unit is used to store a knowledge graph containing the elevator's technical specifications, capacity, brand, model, maximum load, operating speed, number of floors, drive mode, door type, usage scenarios, and installation requirements; The matching algorithm unit is used to search for matching nodes and relationship paths in the knowledge graph based on the intent label and entity list through the graph neural network, and return the matching results.

4. The elevator information query system according to claim 1, characterized in that: The multimodal matching engine comprises: The text generation unit is used to extract and generate candidate answers in text form from the knowledge graph query results; A confidence scoring unit, used to score the confidence of each candidate answer based on the deep learning model; A source identification unit is used to identify the source type of each candidate answer, where the source type includes text, image, and audio, wherein the image and audio are processed using computer vision and speech recognition models.

5. The elevator information query system according to claim 1, characterized in that: The dynamic reply generation module includes: A historical conversation storage unit is used to store the user's historical conversation records; A context-aware engine, which uses an LSTM algorithm to generate context-relevant natural language responses based on the context and historical conversations. The compliance verification unit is used to check whether the generated responses meet the output requirements based on the artificial intelligence compliance checking model.

6. The elevator information query system according to claim 1, characterized in that: The user feedback processing module includes: User behavior tracking unit, used to record and analyze user interaction with the system, including click counts and dwell time, and conduct feedback analysis through deep learning models; The implicit feedback analysis unit is used to analyze user behavior to generate feedback signals and optimize the artificial intelligence model used based on the signals.

7. The elevator information query system according to claim 1, characterized in that: The closed-loop iterative optimization module includes: A model optimization unit, which is used to adjust the parameters of the AI ​​model based on user feedback signals and optimize the performance of the AI ​​model using automatic machine learning technology; The knowledge updating unit is used to update the elevator information in the knowledge graph based on feedback.

8. An elevator information query method, characterized in that: The method comprises: Receive natural language questions input by users and generate structured semantic representations; Searching for matching nodes and relationship paths in the knowledge graph according to the structured semantic representation, and returning matching subgraph results; Generate multiple candidate answers and assign confidence scores to them based on AI algorithms; Generate context-sensitive natural language responses based on historical conversation records and perform compliance verification through AI compliance models; Optimize AI models and knowledge graphs based on user feedback; Optimize the performance of the query system through a closed-loop iterative process, and use algorithms such as reinforcement learning to continuously improve the system's response speed and accuracy.

9. The elevator information query method according to claim 8, characterized in that: The receiving of a natural language question input by a user and generating a structured semantic representation includes: User questions generate intent labels and entity lists through intent recognition and entity extraction; Confidence evaluation is performed on the generated intent labels and entity lists to generate structured semantic representations.

10. The elevator information query method according to claim 8, characterized in that: The searching for matching nodes and relationship paths in the knowledge graph according to the structured semantic representation and returning matching subgraph results includes: Find nodes and their relationship paths in the knowledge graph that match the intent label and entity list; Returns the matching knowledge graph subgraph results.

Images