Real-time interaction method and system based on tourism intelligent big model and storage medium

Abstract

The application provides a real-time interaction method and system based on a tourism intelligent large model and a storage medium. The real-time interaction method based on the tourism intelligent large model combines a knowledge graph and retrieval enhancement to generate a new retrieval enhancement system based on the knowledge graph, solves the problems of refusal to answer and model hallucination caused by insufficient knowledge of the large model and untimely knowledge update, and solves the technical problem of low retrieval accuracy caused by imprecise input questions in a conventional RAG algorithm.

Description

Technical Field

[0001] This invention relates to the field of intelligent tourism technology, and in particular to a real-time interactive method, system, and storage medium based on a large-scale intelligent tourism model. Background Technology

[0002] Existing hardware devices can be equipped with intelligent tourism models. However, these models suffer from insufficient knowledge and outdated knowledge, leading to issues such as refusing to answer questions and fabricating answers (model illusion). Furthermore, existing models generally employ the conventional RAG algorithm, which, due to the non-standard nature of human-input questions, results in inaccurate retrieval of answers, requiring improvement.

[0003] Therefore, it is necessary to provide a new real-time interactive method, system, and storage medium based on a tourism intelligent big data model to solve the above-mentioned technical problems. Summary of the Invention

[0004] The main objective of this invention is to provide a real-time interactive method based on a large intelligent tourism model, which aims to solve the technical problems of tourism intelligent models in related technologies, such as refusal to answer, model illusion, and low accuracy of retrieved answers.

[0005] To achieve the above objectives, the present invention provides a real-time interactive method based on a large-scale intelligent tourism model, comprising the following steps:

[0006] Full-domain data collection;

[0007] The entire data is first segmented using a vector model algorithm to obtain text paragraphs and construct a vector database based on paragraph knowledge.

[0008] The entire domain data is processed in a second block using a tourism intelligent big data model to construct a graph database based on knowledge graphs.

[0009] The system acquires user question data and performs retrieval enhancement and generation processing on the question data to obtain expanded questions, i.e., more complete and accurate question descriptions.

[0010] The expansion problem is subjected to a first retrieval process, which involves retrieving the vector database to obtain relevant paragraph knowledge, and organizing the relevant paragraph knowledge to obtain a first context based on vector retrieval.

[0011] The expanded problem is subjected to a second retrieval process to identify entities and relationships in the problem, and relevant subgraph data is obtained in the graph database based on the entities and relationships.

[0012] The subgraph data is transformed and processed to obtain a second context based on knowledge graph retrieval by organizing the attribute content of nodes and edges in the subgraph data;

[0013] Based on the expanded question, a search is performed on a third-party search platform to obtain third-party contextual data;

[0014] By combining the first context data, the second context data, and the third context data, the final context data with a preset sorting is obtained.

[0015] Based on the final context data and combined with the prompting template, the final prompting data is obtained and input into the output interface of the intelligent tourism model to obtain the final answer data.

[0016] Preferably, it further includes the following:

[0017] The answer data is then updated and added to both the vector database and the graph database.

[0018] Preferably, the step of obtaining a second context based on knowledge graph retrieval by organizing the attribute content of nodes and edges in the subgraph obtained by the second retrieval specifically includes the following:

[0019] The initial importance weights of nodes are calculated by computing entity similarity, relation similarity, and direct semantic similarity between the input question and the nodes, and then the node weights are updated by a random walk algorithm.

[0020] The content of all subgraph data is organized in descending order based on the updated weights to construct the context.

[0021] Preferably, the specific formula for calculating entity similarity is as follows:

[0022]

[0023] Where i and j represent entity indexes, p i This represents the probability of identifying entity i based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the entity.

[0024] Preferably, the specific formula for calculating the relationship similarity is as follows:

[0025]

[0026] Where s and t represent the indexes of the relation, and q s This represents the probability of identifying relation s based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the relation.

[0027] Preferably, the specific formula for calculating the direct semantic similarity between the input question and the node is as follows:

[0028]

[0029] Where i represents the entity index, p i This represents the probability of identifying entity i based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the entity or the input.

[0030] Preferably, the three similarity measures mentioned above are combined to calculate the initial weights of nodes in the subgraph data. The specific fusion formula is as follows:

[0031]

[0032] Where i represents the entity index, N i pre E represents the set of nodes related to node i in the subgraph data. i pre This represents the set of relationships related to node i in the subgraph data.

[0033] Preferably, the specific formula of the random walk algorithm is as follows:

[0034] W t =αRWt -1 +(1-α)S

[0035] Where S is an N*1 vector, R is an N*N vector, N is the number of nodes in the subgraph data, R is represented by a matrix of the relational network of the subgraph data, W is an N*1 vector, which is the vector for updating weights during the iteration process, and its initial vector is S; t represents the number of iterations, i.e., the number of steps, and α is the restart probability, a small value between 0 and 1.

[0036] To address the aforementioned technical problems, the present invention also provides a real-time interactive system based on a tourism intelligent big data model, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, it implements the steps of the real-time interactive method based on the tourism intelligent big data model.

[0037] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the real-time interactive method based on a tourism intelligent big data model.

[0038] The real-time interactive method based on a large intelligent tourism model provided by this invention combines knowledge graphs and retrieval enhancement generation to form a new knowledge graph-based retrieval enhancement system. This system solves the problems of insufficient knowledge in the large model and untimely knowledge updates, which lead to rejection of answers and fabrication (model illusion), as well as the technical problem of low retrieval accuracy in conventional RAG algorithms due to imprecise input questions. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating a preferred embodiment of the real-time interactive method based on a large intelligent tourism model provided by the present invention.

[0040] Figure 2 This invention provides an architecture diagram of a system based on a large intelligent tourism model.

[0041] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0042] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0043] This invention provides a real-time interactive method based on a large intelligent tourism model.

[0044] Please see Figure 1 To achieve the above objectives, in one embodiment of the present invention, the real-time interaction method based on a tourism intelligent big data model includes the following steps:

[0045] S1, Global Data Acquisition;

[0046] Specifically, the comprehensive data includes network-collected data, user interaction data, and offline IoT device and sensor data.

[0047] The data collected online includes information from various tourism websites, publicly available knowledge data from the Internet, and travel guides and travelogues from various content platforms.

[0048] The user interaction data includes real-time operation logs of users on the product side, user profiles, and real-time user behavior characteristic data.

[0049] The offline IoT device and sensor data includes the collected data of offline turnstiles, scenic area equipment, and security hardware in various destinations or tourist attractions.

[0050] S2, The entire data is processed by a vector model algorithm to obtain text paragraphs and construct a vector database based on paragraph knowledge;

[0051] For example, the overall data is tourism website information data. The tourism website information data is processed by a vector model algorithm to perform the first block processing, converting text data, image data and video data into text paragraphs, so as to obtain the tourism website's introduction to the tourism route and travel guide of a certain scenic spot, in order to build a vector database based on paragraph knowledge.

[0052] For example, the full-domain data is the user's real-time operation log on the mobile terminal. The real-time operation log is processed by the vector model algorithm to perform the first block processing, and the real-time operation log is converted into text paragraphs to obtain the user's search query records for a certain scenic spot or multiple scenic spots on the mobile terminal, so as to build a vector database based on paragraph knowledge.

[0053] For example, the aforementioned comprehensive data refers to the collected data from the offline turnstiles of tourist attractions. The collected data is processed into first blocks using a vector model algorithm to obtain text segments about tourist travel patterns within the tourist attractions, thereby constructing a vector database based on segment knowledge.

[0054] S3, the entire domain data is processed in a second block using the tourism intelligent big data model to construct a graph database based on knowledge graphs;

[0055] Specifically, a knowledge graph (KG) is a structured representation of information and the relationships between it, presented in the form of nodes and edges. Nodes represent entities, such as people, places, and items, while edges represent the relationships between entities.

[0056] The goal of knowledge graphs is to connect complex information in a structured way to form a vast network, enabling computers to better understand and reason about data.

[0057] The construction of knowledge graphs involves steps such as extracting information from multiple data sources, data cleaning, entity recognition and disambiguation, relation extraction, and the storage and retrieval of the graph structure. Through these steps, knowledge graphs can help build a semantically rich and interconnected information world, providing strong support for intelligent applications.

[0058] This invention extracts triple information from various publicly available and internal structured and unstructured data within the tourism industry. The triple data includes node A, node B, and edge-AB, constructing a private tourism industry knowledge graph. Extracted node types include destination, province, city, POI, product, user, order, store, travel guide, itinerary template, and itinerary route. Notably, for unstructured data, this invention utilizes a large model to extract structured triple information by constructing appropriate prompts.

[0059] Furthermore, this invention employs a real-time data processing flow to perform stream computing processing on data such as real-time user interaction data and real-time data collected from offline IoT devices in scenic areas, thereby updating the knowledge graph online.

[0060] Streaming data processing is a technique for processing large amounts of continuously input data, typically transmitted as a data stream. It is designed to handle unbounded datasets and can process both historical and real-time data. Streaming data processing systems usually offer a degree of fault tolerance and scalability and can handle various data processing tasks, such as filtering, aggregation, and windowing operations.

[0061] For example: Node A: Hong Kong, China (a tourist destination), Node B: Disneyland, edge-AB: Hong Kong, China (a tourist destination) and Disneyland.

[0062] S4, obtain the user's question data, and perform retrieval enhancement generation processing on the question data to obtain expanded questions, that is, more complete and accurate question descriptions;

[0063] Specifically, the AI ​​tourism big data model, combined with knowledge graphs and retrieval enhancement generation, forms a new knowledge graph-based retrieval enhancement system. This system addresses the problems of insufficient knowledge and untimely knowledge updates in the big data model, which lead to rejection of answers and fabrication (model illusion). It also solves the common and unsolvable retrieval accuracy problem in conventional RAG algorithms caused by imprecise input.

[0064] Retrieval-enhanced generation refers to an algorithmic system that enhances the generation quality of large language generation models by constructing context relevant to the input question. By supplementing the model with unknown knowledge, it solves the problem that large models cannot answer or answer incorrectly in practical applications due to insufficient knowledge.

[0065] S5, perform a first retrieval process on the expanded problem, retrieve the relevant paragraph knowledge by searching the vector database, and organize the relevant paragraph knowledge to obtain a first context based on vector retrieval;

[0066] S6, perform a second retrieval process on the expanded problem, identify the entities and relationships in the problem, and obtain relevant subgraph data in the graph database based on the entities and relationships;

[0067] S7, the subgraph data is transformed and processed to obtain a second context based on knowledge graph retrieval by organizing the attribute content of nodes and edges in the subgraph data;

[0068] S8, perform a search on a third-party search platform based on the expanded question to obtain third-party context data;

[0069] S9, combine the first context data, the second context data, and the third context data to obtain the final context data with a preset sorting;

[0070] S10. Based on the final context data and combined with the prompting template, the final prompting data is obtained and input into the output interface of the intelligent tourism model to obtain the final answer data.

[0071] In this embodiment, the preset sorting is the second context data, the first context data, and the third context data.

[0072] The real-time interactive method based on a large intelligent tourism model provided by this invention combines knowledge graphs and retrieval enhancement generation to form a new knowledge graph-based retrieval enhancement system. This system solves the problems of insufficient knowledge in the large model and untimely knowledge updates, which lead to rejection of answers and fabrication (model illusion), as well as the technical problem of low retrieval accuracy in conventional RAG algorithms due to imprecise input questions.

[0073] As a preferred embodiment of the present invention, the real-time interactive method based on the intelligent tourism model further includes:

[0074] The answer data is then updated and added to both the vector database and the graph database. This further deepens the database and helps improve the stability of the final answer data.

[0075] This invention implements a knowledge graph-based retrieval enhancement system for relevant subgraph retrieval. For the retrieved subgraph, this invention calculates the initial importance weight of nodes by calculating entity similarity, relation similarity, and direct semantic similarity between the input question and the nodes. Then, it updates the node weights using a random walk algorithm. Finally, it organizes the content of the subgraph into contextual text based on the importance weights from high to low.

[0076] Step S7 specifically includes the following:

[0077] The initial importance weights of nodes are calculated by computing entity similarity, relation similarity, and direct semantic similarity between the input question and the nodes, and then the node weights are updated by a random walk algorithm.

[0078] The content of all subgraph data is organized in descending order based on the updated weights to construct the context.

[0079] The specific formula for calculating entity similarity is as follows:

[0080]

[0081] Where i and j represent entity indexes, p i This represents the probability of identifying entity i based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the entity.

[0082] The specific formula for calculating the similarity of the relationships is as follows:

[0083]

[0084] Where s and t represent the indexes of the relation, and q s This represents the probability of identifying relation s based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the relation.

[0085] The specific formula for calculating the direct semantic similarity between the input question and the node is as follows:

[0086] Where i represents the entity index, p i This represents the probability of identifying entity i based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the entity or the input.

[0087] By combining the three similarity metrics mentioned above, the initial weights of nodes in the subgraph data are calculated. The specific fusion formula is as follows:

[0088]

[0089] Where i represents the entity index, N i pre E represents the set of nodes related to node i in the subgraph data. i pre This represents the set of relationships related to node i in the subgraph data.

[0090] W t =αRW t-1 +(1-α)S

[0091] Where S is an N*1 vector, R is an N*N vector, N is the number of nodes in the subgraph data, R is the matrix representation of the relational network of the subgraph data, W is an N*1 vector, which is the vector for updating weights during the iteration process, and its initial vector is S; t represents the number of iterations, i.e. the number of steps, and α is the restart probability, which is a small value between 0 and 1.

[0092] The matrix in R is a symmetric matrix, where the values ​​represent the connections between nodes; 0 indicates no connection, and 1 indicates a connection. For example, in R... i j =1 indicates that there is an edge between the i-th node and the j-th node, S i It is an element in S.

[0093] This invention is based on calculating w, organizing the text content of nodes in descending order, and then combining the results of vector database retrieval and online search results to form the final context. Content exceeding the length limit of the generative model is truncated. This method solves the problem of excessively large subgraphs and information overload in knowledge graph retrieval results, achieving highly efficient and high-quality retrieval.

[0094] In summary, this invention, based on a knowledge graph-based enhanced retrieval and generation system, introduces real-time data processing to update the knowledge graph. Combined with prompting engineering and language generation of a large model, it addresses the problems of insufficient model knowledge encountered in practical applications and the inaccuracy of traditional enhanced retrieval and generation systems. Furthermore, the knowledge graph and vector database enable real-time adjustments and feedback based on user needs. In business scenarios, this is mainly reflected in: real-time itinerary changes, real-time adjustments and pricing of travel packages, and real-time feedback and push notifications of destination information.

[0095] To facilitate understanding of the technical solution, the key workflow of the present invention is illustrated below with examples:

[0096] Step 1: User Question: Information on popular tourist attractions in Hong Kong, China.

[0097] Step 2: Issues after rewriting:

[0098] What are some popular tourist attractions in Hong Kong?

[0099] Information about Hong Kong Disneyland;

[0100] Information about Ocean Park, Hong Kong;

[0101] Information about Madame Tussauds Hong Kong.

[0102] Step 3: Search for relevant information:

[0103] 1. Retrieve the vector knowledge base: Obtain the result list doc1, doc2, and doc3, sorted in descending order based on semantic similarity.

[0104] Specifically:

[0105] doc1: Popular attractions in Hong Kong include: Hong Kong Disneyland, Hong Kong Ocean Park, Madame Tussauds Hong Kong, Victoria Peak, etc.

[0106] Document 2: Hong Kong Disneyland is located on Lantau Island in the New Territories of Hong Kong. It officially opened on September 12, 2005, and is the fifth Disneyland in the world and the first in China. The park is divided into seven themed lands: Main Street, U.S.A., Adventureland, Fantasyland, Tomorrowland, Toy Story Land, Grizzly Gulch, and Mystic Manor. Grizzly Gulch and Mystic Manor are unique to Hong Kong Disneyland. The park features themed rides, entertainment shows, interactive experiences, dining services, merchandise shops, and snack kiosks. In addition, the park presents a parade and fireworks display every evening.

[0107] doc3: Ocean Park Hong Kong is located at 180 Wong Chuk Hang Road, Southern District, Hong Kong. Nestled between mountains and the sea, it features an aquarium and a theme park, mainly divided into two parts: the Waterfront Park at the foot of the mountain and the Summit Park on the mountain, connected by a cable car. It comprises eight themed areas, including Asian Animals, Aqua City, Whiskers World, and Ocean World. Ocean Park Hong Kong hosts five major events annually, including the Lunar New Year-themed Lucky Draw, the High-Definition Zoo featuring rare animals, the Summer Water Festival, Halloween-themed Halloween in October, and Christmas-themed Christmas HOHOHO. Many unique attractions are available for participants during these events.

[0108] 2. Retrieve graph databases based on knowledge graphs

[0109] 2.1 Entity (relationship) identified: Hong Kong, China (Popular tourist attraction)

[0110] Hong Kong Disneyland (introduction), Hong Kong Ocean Park (introduction), Madame Tussauds Hong Kong (introduction).

[0111] 2.2 Retrieval Subgraph

[0112] 2.3 Text Conversion: Organize the sorting results according to the method described in the text, and then splice the information.

[0113] For example, sub-image text conversion:

[0114] Hong Kong is a Special Administrative Region of China, consisting of Hong Kong Island, Lantau Island, Kowloon Peninsula and the New Territories (including 261 outlying islands). It has a mountainous terrain and a marine subtropical monsoon climate. Popular attractions in Hong Kong include Hong Kong Disneyland, Ocean Park Hong Kong and Madame Tussauds Hong Kong.

[0115] Introduction to Hong Kong Disneyland: xxx

[0116] Introduction to Ocean Park Hong Kong: xxxx

[0117] Madame Tussauds Hong Kong Introduction: xxxx.

[0118] 3. Online search: Sort and combine results based on search ranking.

[0119] doc1: Hong Kong boasts numerous attractions, including Victoria Harbour, Victoria Peak, Hong Kong Disneyland, Ocean Park, Sky100, Lantau Island, and many more, all of which are extremely popular!

[0120] What are some new attractions in Hong Kong that are worth visiting recently? Below, we've compiled a list of must-see spots in Hong Kong. If you're planning a trip to Hong Kong, follow these tips to plan your independent travel itinerary!

[0121] doc2: Hong Kong and Macau boast numerous attractions, including an international metropolis, a world-class ocean park, the vibrant Victoria Harbour, Victoria Peak, the Avenue of Stars where Hong Kong celebrities gather, the world-famous Hong Kong-Zhuhai-Macau Bridge, a dazzling array of luxury brands, Michelin-starred cuisine, and the Lisboa Hotel, making them a top tourist destination for mainland Chinese tourists.

[0122] Step 4: Combine the three types of text in the following order: knowledge graph database, vector database, and third-party search.

[0123] Step 5: Combine prompts: Please answer the user's questions based on the following reference information.

[0124] Reference information: (assembled content)

[0125] User's question: Information on popular tourist attractions in Hong Kong, China

[0126] Step 6: The model generates the answer: Popular attractions in Hong Kong include:

[0127] Hong Kong Disneyland (xxx);

[0128] Ocean Park Hong Kong, China (xxx);

[0129] Madame Tussauds Hong Kong, China (xxx);

[0130] Victoria Peak xxx;

[0131] Hong Kong Disneyland, etc.

[0132] The present invention also provides a real-time interactive system based on a tourism intelligent big data model.

[0133] Please see Figure 2A real-time interactive system based on a tourism intelligent big data model includes a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, it implements the steps of the real-time interactive method based on the tourism intelligent big data model.

[0134] The present invention also provides a computer-readable storage medium.

[0135] The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the real-time interactive method based on the intelligent tourism model.

[0136] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a computer-readable storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to enter the methods described in the various embodiments of the present invention.

[0137] In the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Xth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, method steps, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0138] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0139] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0140] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A real-time interactive method based on a large-scale intelligent tourism model, characterized in that, Includes the following steps: Collection of data across the entire domain; The entire data is first segmented using a vector model algorithm to obtain text paragraphs and construct a vector database based on paragraph knowledge. The entire domain data is processed in a second block using a tourism intelligent big data model to construct a graph database based on knowledge graphs. The system acquires user question data and performs retrieval enhancement and generation processing on the question data to obtain expanded questions, i.e., more complete and accurate question descriptions. The expansion problem is subjected to a first retrieval process, which involves retrieving the vector database to obtain relevant paragraph knowledge, and organizing the relevant paragraph knowledge to obtain a first context based on vector retrieval. The expanded problem is subjected to a second retrieval process to identify entities and relationships in the problem, and relevant subgraph data is obtained in the graph database based on the entities and relationships. The subgraph data is transformed and processed to obtain a second context based on knowledge graph retrieval by organizing the attribute content of nodes and edges in the subgraph data; Based on the expanded question, a search is performed on a third-party search platform to obtain third-party contextual data; By combining the first context data, the second context data, and the third context data, final context data with a preset sorting is obtained; Based on the final context data and combined with the prompting template, the final prompting data is obtained and input into the output interface of the intelligent tourism model to obtain the final answer data.

2. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 1, characterized in that, It also includes the following: The answer data is then updated and added to both the vector database and the graph database.

3. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 1, characterized in that, The step of transforming the subgraph data to obtain the second context based on knowledge graph retrieval by organizing the attribute content of nodes and edges in the subgraph data specifically includes the following: The initial importance weights of nodes are calculated by computing entity similarity, relation similarity, and direct semantic similarity between the input question and the nodes, and then the node weights are updated by a random walk algorithm. The content of all subgraph data is organized in descending order based on the updated weights to construct the context.

4. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 3, characterized in that, The specific formula for calculating entity similarity is as follows: Where i and j represent entity indexes, p i This represents the probability of identifying entity i based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the entity.

5. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 4, characterized in that, The specific formula for calculating the similarity of the relationships is as follows: Where s and t represent the indexes of the relation, and q s This represents the probability of identifying relation s based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the relation.

6. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 5, characterized in that, The specific formula for calculating the direct semantic similarity between the input question and the node is as follows: Where i represents the entity index, p i This represents the probability of identifying entity i based on the input question; n and k represent the dimension and index of the vector, respectively; e is the vector representation of the entity or the input.

7. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 6, characterized in that, By combining the three similarity metrics mentioned above, the initial weights of nodes in the subgraph data are calculated. The specific fusion formula is as follows: Where i represents the entity index, N i pre E represents the set of nodes related to node i in the subgraph data. i pre This represents the set of relationships related to node i in the subgraph data.

8. The real-time interactive method based on a large-scale intelligent tourism model as described in claim 7, characterized in that, The specific formula for the random walk algorithm is as follows: W t =αRW t-1 +(1-α)S Where S is an N*1 vector, R is an N*N vector, N is the number of nodes in the subgraph data, R is the matrix representation of the relational network of the subgraph data, W is an N*1 vector, which is the vector for updating weights during the iteration process, and its initial vector is S; t represents the number of iterations, i.e. the number of steps, and α is the restart probability, which is a small value between 0 and 1.

9. A real-time interactive system based on a large-scale intelligent tourism model, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, it implements the steps of the real-time interactive method based on the intelligent tourism model as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the real-time interactive method based on the intelligent tourism model as described in any one of claims 1-8.

Images