A travel guide generation method and system based on keywords and user portraits

Abstract

This invention discloses a method and system for generating travel guides based on keywords and user profiles, belonging to the field of artificial intelligence technology. The method includes the following steps: acquiring multi-dimensional user data and constructing a multi-dimensional user profile; extracting travel intentions and entity keywords; retrieving and filtering matching travel information entities based on the structured travel intentions; generating a preliminary travel guide including routes and contingency plans; receiving user feedback on the preliminary travel guide for continuous optimization. This invention solves the problems of existing technologies where travel guides are generic and fail to meet individual needs, information updates are often outdated, and users still need to spend a lot of time manually filtering and making decisions. These problems are addressed by constructing multi-dimensional user profiles to accurately capture user preferences, utilizing real-time updated knowledge graphs to ensure information accuracy, and introducing reinforcement learning mechanisms to achieve adaptive optimization of the recommendation system.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for generating travel guides based on keywords and user profiles. Background Technology

[0002] With the improvement of the national economic level and the change of consumption concepts, tourism has become an important part of people's daily lives. At the same time, the popularization of the Internet has spawned a massive amount of online tourism information and user-generated content, providing travelers with a wealth of data sources for planning their trips.

[0003] Existing travel guide generation systems typically generate and recommend guides by manually editing and compiling popular tourist routes or by aggregating user-generated content into general templates.

[0004] In existing technologies, such methods heavily rely on static, unstructured data, resulting in low personalization, outdated information, and an inability to dynamically adapt to users' real-time needs. Specifically, recommended travel guides are highly homogenized, failing to deeply match users' unique interests, budgets, and travel styles; users must sift through massive amounts of outdated or conflicting information to make decisions, leading to inefficient planning and severely impacting the travel experience. Summary of the Invention

[0005] This application provides a method and system for generating travel guides based on keywords and user profiles. This solves the problems of high user decision-making costs caused by the severe homogenization of static recommended content and the lag in information updates in the prior art. It realizes the dynamic generation of highly personalized and real-time accurate travel guides, which significantly reduces the user's planning burden and improves the travel experience.

[0006] This application provides a method for generating travel guides based on keywords and user profiles, including: acquiring multidimensional user data and constructing a multidimensional user profile that includes user preferences, budget, and historical behavior; The system receives natural language requests input by users, parses them using a domain-specific deep natural language processing model, and extracts structured travel intentions and entity keywords. The deep natural language processing model is specifically optimized for parsing colloquial and ambiguous travel requests. Based on the structured travel intent, matching travel information entities are retrieved and filtered from a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms. By integrating the multi-dimensional user profile with the matched travel information entities, a preliminary travel guide containing routes and emergency plans is generated through a personalized recommendation algorithm engine. The system receives user feedback on the initial travel guide and uses the feedback data to continuously optimize the personalized recommendation algorithm engine through a reinforcement learning mechanism.

[0007] Furthermore, the steps for constructing a domain-specific deep natural language processing model are as follows: Base model selection: A pre-trained model based on the Transformer architecture was used as the initial model; Domain-specific corpus preparation: Collect publicly available corpora in the travel domain, process them using text cleaning techniques, and construct a dedicated corpus for the travel domain; Domain fine-tuning: Based on transfer learning technology, the pre-trained model is fine-tuned using travel domain corpus. By setting the learning rate and training batch hyperparameters, the model parameters are updated using gradient descent optimization algorithm. Task adaptation: For tasks such as travel intent recognition, entity extraction, and complex demand parsing, a task-specific output layer is added to the fine-tuned model. The model is trained using manually labeled travel domain task data, and the parameters of the task layer are adjusted through cross-validation. Model optimization: The model size is reduced by using model compression methods, the model performance is evaluated through a test set, and iterative optimization is performed until the practical accuracy requirements are met.

[0008] Furthermore, the multi-source travel knowledge graph possesses real-time updates and conflict resolution mechanisms, the specific conflict resolution mechanisms of which include: When N data sources provide conflicting discrete values ​​for the same information entity At that time, by calculating the support score for each value The optimal value is selected using the following calculation method: ; in, It provides value The collection of data sources It is a data source Static credibility, It is a data source Dynamic timeliness weight, It is a balance coefficient that adjusts the static and dynamic weights. The optimal value is determined by verifying historical data. The Calculated using the exponential decay function: ; in, This is the current timestamp. For data source Information update timestamp The time decay constant was ultimately adopted. highest discrete value As authoritative information in a knowledge graph.

[0009] Furthermore, the personalized recommendation algorithm engine calculations include: The knowledge graph is processed using a graph neural network (GNN) model. The multi-dimensional user profile and the structured travel intent are used as query nodes. Features of neighboring travel information entities are aggregated in the graph to generate user... For entities Initial interest embedding .

[0010] Furthermore, the personalized recommendation algorithm engine incorporates context awareness to calculate the user's... For the entity Final personalized preference score : ; in, It is the Sigmoid activation function. It is the embedding vector of the user profile. Entities aggregated through the graph neural network model. Feature embedding, i.e. , It is the embedding vector of the current context. It is a physical entity Context sensitivity vector It is a bias term, and is recommended first. Tall entities.

[0011] Furthermore, a preliminary travel guide is generated, including routes and contingency plans. The generation of contingency plans includes: Real-time monitoring of external emergencies, including weather disasters, traffic control, temporary closure of attractions, or public health events; when an event is detected to affect the original travel route, an emergency response plan generation process is automatically triggered. When a trigger condition is detected, the system automatically retrieves alternative scores with highly personalized preferences from the knowledge graph. Alternative entities They also replanned the route and generated an emergency response plan.

[0012] Furthermore, the personalized recommendation algorithm engine is continuously optimized using the feedback data through a reinforcement learning mechanism, specifically including: The strategy generation process is modeled as a Markov decision process, where: Status: Includes the current user profile, selected trips, and context information; Action: Recommend the next travel destination; Rewards: Users receive instant rewards based on their editing feedback and trip completion rate. .

[0013] Furthermore, instant rewards The calculation method is as follows: ; in, It is the user in Always recommend Feedback actions, It is an indicator function. , These are the weighting coefficients for acceptance and rejection, respectively. It is the user's rating of the final trip completion rate. It is its weight; The goal of the personalized recommendation algorithm engine is to maximize the cumulative expected reward. The reward signal is then used to update the graph neural network model and the personalized preference score. The parameters.

[0014] Furthermore, the receipt of user feedback on the initial travel guide is implemented based on a human-computer collaborative optimized interface: This interface allows users to perform visual editing operations such as dragging, replacing, and deleting AI-generated guides. The calculation is recalculated in real time based on the user's editing actions. And an alternative is provided, wherein the edit operation data is collected as an immediate reward for the reinforcement learning mechanism. Input data.

[0015] This application provides a travel guide generation system based on keywords and user profiles, which implements a method for generating travel guides based on keywords and user profiles, including: The module includes: profile building module, query parsing module, entity retrieval module, strategy generation module, and feedback optimization module. The user profile building module is used to acquire multi-dimensional user data and build a multi-dimensional user profile that includes user preferences, budget, and historical behavior. The query parsing module is used to receive natural language requests input by users, and to parse them using a domain-specific deep natural language processing model to extract structured travel intentions and entity keywords. The deep natural language processing model is specifically optimized for parsing colloquial and ambiguous travel requests. The entity retrieval module is used to retrieve and filter matching travel information entities from a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms, based on the structured travel intent. The travel guide generation module is used to integrate the multi-dimensional user profile with the matched travel information entities, and calculate and generate a preliminary travel guide containing routes and emergency plans through a personalized recommendation algorithm engine; The feedback optimization module is used to receive user feedback on the initial travel guide and to continuously optimize the personalized recommendation algorithm engine using the feedback data through a reinforcement learning mechanism.

[0016] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By constructing multi-dimensional user profiles and parsing users' natural language requests, we can deeply understand users' personalized needs and generate highly customized travel guides, effectively solving the problem of severe homogenization in traditional travel guides and improving the relevance of recommended content. Furthermore, in the personalized recommendation process, the use of a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms can dynamically integrate the latest travel information and automatically filter authoritative data to ensure the timeliness and accuracy of the guides, preventing users from being affected by outdated information. Moreover, in ensuring information accuracy, the personalized recommendation algorithm engine, combined with context awareness and contingency plan generation, can respond in real time to changing user needs and unexpected events, automatically optimizing route planning, reducing users' decision-making time and effort, lowering information filtering costs, and thus comprehensively improving travel planning efficiency and experience satisfaction. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for generating travel guides based on keywords and user profiles, provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a travel guide generation system based on keywords and user profiles, provided in an embodiment of this application. Detailed Implementation

[0018] This application provides a method and system for generating travel guides based on keywords and user profiles. This solves the problems of severe homogenization, poor information timeliness, and high user decision-making costs in the prior art. By constructing multi-dimensional user profiles, accurately analyzing travel intentions, integrating real-time updated knowledge graphs, and continuously optimizing recommendations using reinforcement learning, it achieves the dynamic generation of highly personalized travel guides that are accurate, timely, and adaptive to meet user changes, thereby significantly improving planning efficiency and travel experience.

[0019] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0020] like Figure 1The diagram shown is a flowchart of a travel guide generation method based on keywords and user profiles provided in this application embodiment. The method is applied to a travel guide generation system based on keywords and user profiles. The method includes the following steps: obtaining multi-dimensional user data and constructing a multi-dimensional user profile that includes user preferences, budget, and historical behavior. The system receives natural language requests input by users, parses them using a domain-specific deep natural language processing model, and extracts structured travel intentions and entity keywords. The deep natural language processing model is specifically optimized for parsing colloquial and ambiguous travel requests. Based on the structured travel intent, matching travel information entities are retrieved and filtered from a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms. By integrating the multi-dimensional user profile with the matched travel information entities, a preliminary travel guide containing routes and emergency plans is generated through a personalized recommendation algorithm engine. The system receives user feedback on the initial travel guide and uses the feedback data to continuously optimize the personalized recommendation algorithm engine through a reinforcement learning mechanism.

[0021] Furthermore, the steps for constructing a domain-specific deep natural language processing model are as follows: Base model selection: Use a pre-trained model with the Transformer architecture (such as BERT-base) as the initial model. This model has been trained on a large-scale general corpus and has basic language understanding capabilities. Domain-specific corpus preparation: Collect publicly available corpora in the travel field (including user reviews from travel platforms, official travel guide texts, travel Q&A data, etc.), process them through text cleaning techniques (duplicate removal, noise reduction, and format standardization), and construct a dedicated corpus for the travel field; Domain fine-tuning: Based on transfer learning technology, the pre-trained model is fine-tuned using travel domain corpus. By setting the learning rate (5e-5 to 2e-4) and training batch (16-32) hyperparameters, the gradient descent optimization algorithm is used to update the model parameters, so that the model can adapt to the language features of the travel domain. Task adaptation: For tasks such as travel intent recognition, entity extraction (location, time, budget, etc.), and complex demand parsing, task-specific output layers are added to the fine-tuned model (e.g., a softmax classification layer for intent recognition and a CRF layer for entity extraction). The model is trained using manually labeled travel domain task data (following existing labeling standards such as IOB format), and the task layer parameters are adjusted through cross-validation. Model optimization: The model size is reduced by using model compression methods (such as knowledge distillation). The model performance is evaluated by using a test set (containing diverse travel request samples) (with intent recognition accuracy and entity extraction F1 score as indicators). The model is iteratively optimized until it meets the practical accuracy requirements (accuracy ≥ 90%, F1 score ≥ 85%).

[0022] Furthermore, the multi-source travel knowledge graph possesses real-time updates and conflict resolution mechanisms, the specific conflict resolution mechanisms of which include: When N data sources provide conflicting discrete values ​​for the same information entity (such as the opening hours of a tourist attraction) At that time, by calculating the support score for each value The optimal value is selected using the following calculation method: ; in, It provides value The collection of data sources It is a data source The static credibility is preset based on the data source type (such as official website, authoritative OTA, UGC). It is a data source Dynamic timeliness weight, It is a balance coefficient that adjusts the static and dynamic weights. The optimal value is determined by verifying historical data. The Calculated using the exponential decay function: ; in, This is the current timestamp. For data source Information update timestamp The time decay constant was ultimately adopted. highest discrete value As authoritative information in a knowledge graph.

[0023] Furthermore, the personalized recommendation algorithm engine calculations include: The knowledge graph is processed using a graph neural network (GNN) model. The multi-dimensional user profile and the structured travel intent are used as query nodes. Features of neighboring travel information entities are aggregated in the graph to generate user... For entities Initial interest embedding .

[0024] Furthermore, the personalized recommendation algorithm engine incorporates context awareness to calculate the user's... For the entity Final personalized preference score : ; in, It is the Sigmoid activation function. It is the embedding vector of the user profile. Entities aggregated through the graph neural network model. Feature embedding, i.e. , It is the embedding vector of the current context (weather, time). It is a physical entity The context sensitivity vector represents the degree to which an entity is affected by its context. It is a bias term, and is recommended first. Tall entities.

[0025] Furthermore, a preliminary travel guide is generated, including routes and contingency plans. The generation of contingency plans includes: Real-time monitoring of external emergencies, including weather disasters, traffic control, temporary closure of attractions, or public health events. When an event is detected that affects the original travel route, an emergency response plan generation process is automatically triggered (such as weather changes or temporary closure of attractions). When a trigger condition is detected, the system automatically retrieves alternative scores with highly personalized preferences from the knowledge graph. Alternative entities They also replanned the route and generated an emergency response plan.

[0026] Furthermore, the personalized recommendation algorithm engine is continuously optimized using the feedback data through a reinforcement learning mechanism, specifically including: The strategy generation process is modeled as a Markov decision process (MDP), where: State: Includes the current user profile, selected trips, and context information; Action: Recommend the next travel destination (attraction, restaurant); Rewards: Users receive instant rewards based on their editing feedback (such as deletion, replacement, acceptance) and trip completion rate. .

[0027] Furthermore, instant rewards The calculation method is as follows: ; in, It is the user in Always recommend Feedback actions (acceptance, rejection) It is an indicator function; it returns 1 if the condition is true, and 0 otherwise. , These are the weighting coefficients for acceptance and rejection, respectively. It's a user rating of the final trip's completion level, provided as feedback after the trip ends. It is its weight; The goal of the personalized recommendation algorithm engine (as a policy network) is to maximize the cumulative expected reward. The reward signal is then used to update the graph neural network model and the personalized preference score. The parameters.

[0028] Furthermore, the receipt of user feedback on the initial travel guide is implemented based on a human-computer collaborative optimized interface: This interface allows users to perform visual editing operations such as dragging, replacing, and deleting AI-generated guides. The calculation is recalculated in real time based on the user's editing actions. And an alternative is provided, wherein the edit operation data is collected as an immediate reward for the reinforcement learning mechanism. Input data.

[0029] like Figure 2 The diagram shown is a structural schematic of a travel guide generation system based on keywords and user profiles provided in this application embodiment. The travel guide generation system based on keywords and user profiles provided in this application embodiment includes: a profile building module, a query parsing module, an entity retrieval module, a guide generation module, and a feedback optimization module. The user profile building module is used to acquire multi-dimensional user data and build a multi-dimensional user profile that includes user preferences, budget, and historical behavior. The query parsing module is used to receive natural language requests input by users, and to parse them using a domain-specific deep natural language processing model to extract structured travel intentions and entity keywords. The deep natural language processing model is specifically optimized for parsing colloquial and ambiguous travel requests. The entity retrieval module is used to retrieve and filter matching travel information entities from a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms, based on the structured travel intent. The travel guide generation module is used to integrate the multi-dimensional user profile with the matched travel information entities, and calculate and generate a preliminary travel guide containing routes and emergency plans through a personalized recommendation algorithm engine; The feedback optimization module is used to receive user feedback on the initial travel guide and to continuously optimize the personalized recommendation algorithm engine using the feedback data through a reinforcement learning mechanism.

[0030] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0031] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0032] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0033] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0034] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0035] In conclusion, 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating travel guides based on keywords and user profiles, characterized in that, Includes the following steps: Acquire multidimensional user data to build multidimensional user profiles that include user preferences, budgets, and historical behavior; The system receives natural language requests input by users, parses them using a domain-specific deep natural language processing model, and extracts structured travel intentions and entity keywords. The deep natural language processing model is specifically optimized for parsing colloquial and ambiguous travel requests. Based on the structured travel intent, matching travel information entities are retrieved and filtered from a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms. By integrating the multi-dimensional user profile with the matched travel information entities, a preliminary travel guide containing routes and emergency plans is generated through a personalized recommendation algorithm engine. The system receives user feedback on the initial travel guide and uses the feedback data to continuously optimize the personalized recommendation algorithm engine through a reinforcement learning mechanism.

2. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, The steps for building a domain-specific deep natural language processing model are as follows: Base model selection: A pre-trained model based on the Transformer architecture was used as the initial model; Domain-specific corpus preparation: Collect publicly available corpora in the travel domain, process them using text cleaning techniques, and construct a dedicated corpus for the travel domain; Domain fine-tuning: Based on transfer learning technology, the pre-trained model is fine-tuned using travel domain corpus. By setting the learning rate and training batch hyperparameters, the model parameters are updated using gradient descent optimization algorithm. Task adaptation: For tasks such as travel intent recognition, entity extraction, and complex demand parsing, a task-specific output layer is added to the fine-tuned model. The model is trained using manually labeled travel domain task data, and the parameters of the task layer are adjusted through cross-validation. Model optimization: The model size is reduced by using model compression methods, the model performance is evaluated through a test set, and iterative optimization is performed until the practical accuracy requirements are met.

3. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, A multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms, the specific conflict resolution mechanisms of which include: When N data sources provide conflicting discrete values ​​for the same information entity At that time, by calculating the support score for each value The optimal value is selected using the following calculation method: ； in, It provides value The collection of data sources It is a data source Static credibility, It is a data source Dynamic timeliness weight, It is a balance coefficient that adjusts the static and dynamic weights. The optimal value is determined by verifying historical data. The Calculated using the exponential decay function: ； in, This is the current timestamp. For data source Information update timestamp The time decay constant was ultimately adopted. highest discrete value As authoritative information in a knowledge graph.

4. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, Personalized recommendation algorithm engine calculations include: The knowledge graph is processed using a graph neural network (GNN) model. The multi-dimensional user profile and the structured travel intent are used as query nodes. Features of neighboring travel information entities are aggregated in the graph to generate user... For entities Initial interest embedding .

5. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, The personalized recommendation algorithm engine further incorporates context awareness to calculate the user's... For the entity Final personalized preference score : ； in, It is the Sigmoid activation function. It is the embedding vector of the user profile. Entities aggregated through the graph neural network model. Feature embedding, i.e. , It is the embedding vector of the current context. It is a physical entity Context sensitivity vector It is a bias term, and is recommended first. Tall entities.

6. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, Generate a preliminary travel guide including routes and contingency plans. The generation of contingency plans includes: Real-time monitoring of external emergencies, including weather disasters, traffic control, temporary closure of attractions, or public health events; when an event is detected to affect the original travel route, an emergency response plan generation process is automatically triggered. When a trigger condition is detected, the system automatically retrieves alternative scores with highly personalized preferences from the knowledge graph. Alternative entities They also replanned the route and generated an emergency response plan.

7. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, The personalized recommendation algorithm engine is continuously optimized using the feedback data through a reinforcement learning mechanism, specifically including: The strategy generation process is modeled as a Markov decision process, where: Status: Includes the current user profile, selected trips, and context information; Action: Recommend the next travel destination; Rewards: Users receive instant rewards based on their editing feedback and trip completion rate. .

8. The method for generating travel guides based on keywords and user profiles as described in claim 7, characterized in that, Instant rewards The calculation method is as follows: ； in, It is the user in Always recommend Feedback actions, It is an indicator function. , These are the weighting coefficients for acceptance and rejection, respectively. It is the user's rating of the final trip completion rate. It is its weight; The goal of the personalized recommendation algorithm engine is to maximize the cumulative expected reward. The reward signal is then used to update the graph neural network model and the personalized preference score. The parameters.

9. The method for generating travel guides based on keywords and user profiles as described in claim 1, characterized in that, The process of receiving user feedback on the initial travel guide is implemented based on a human-computer collaborative optimized interface. This interface allows users to perform visual editing operations such as dragging, replacing, and deleting AI-generated guides. The calculation is recalculated in real time based on the user's editing actions. And an alternative is provided, wherein the edit operation data is collected as an immediate reward for the reinforcement learning mechanism. Input data.

10. A travel guide generation system based on keywords and user profiles, used to implement the travel guide generation method based on keywords and user profiles as described in any one of claims 1-9, characterized in that, include: The module includes: profile building module, query parsing module, entity retrieval module, strategy generation module, and feedback optimization module. The user profile building module is used to acquire multi-dimensional user data and build a multi-dimensional user profile that includes user preferences, budget, and historical behavior. The query parsing module is used to receive natural language requests input by users, and to parse them using a domain-specific deep natural language processing model to extract structured travel intentions and entity keywords. The deep natural language processing model is specifically optimized for parsing colloquial and ambiguous travel requests. The entity retrieval module is used to retrieve and filter matching travel information entities from a multi-source travel knowledge graph with real-time updates and conflict resolution mechanisms, based on the structured travel intent. The travel guide generation module is used to integrate the multi-dimensional user profile with the matched travel information entities, and calculate and generate a preliminary travel guide containing routes and emergency plans through a personalized recommendation algorithm engine; The feedback optimization module is used to receive user feedback on the initial travel guide and to continuously optimize the personalized recommendation algorithm engine using the feedback data through a reinforcement learning mechanism.

Images