Personalized tourist attraction recommendation method and system based on knowledge graph

Abstract

The invention is suitable for the technical field of scenic spot recommendation, and provides a personalized tourist scenic spot recommendation method and system based on a knowledge graph, and the method comprises the steps: collecting preset scenic spot user behavior data, scenic spot attribute data and time-space information data; a dynamically updated knowledge graph is constructed, nodes comprise user entities, scenic spot entities, time entities and space entities, and edges comprise interaction relations between users and scenic spots, attribute association relations between scenic spots and time-space constraint relations; the A3C algorithm is collected to dynamically optimize the knowledge graph, the dynamic optimization comprises exploring nodes and edges in the knowledge graph through a plurality of asynchronous threads, and each thread executes operation comprising network generation recommendation strategy, network evaluation strategy value and adjustment of weights of the nodes and the edges in the knowledge graph; dynamically adjusting A3C algorithm parameters according to the regional features of the preset scenic area; and a personalized recommendation scheme adaptive to the user demand is generated based on the optimized knowledge graph and parameter configuration, so that the accuracy of the recommendation scheme is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scenic area recommendation, and particularly relates to a personalized tourist scenic area recommendation method and system based on a knowledge graph. Background Art

[0002] With the rapid development of the cultural and tourism market, personalized tourism recommendation systems have become key tools for enhancing the tourist experience and scenic area revenue. However, some current recommendation systems mostly adopt an isolated data analysis mode, failing to effectively integrate multi-source data such as user behavior data, scenic area attribute data, and spatio-temporal information, resulting in single and scene-inadaptable recommendation results. In addition, the ability to process user feedback in real time is insufficient, and the recommendation strategy lacks a self-optimization mechanism. Especially in complex and changeable tourism scenarios, it is unable to adjust model parameters through continuous learning and cannot adapt to the dynamic changes of real-time data, making the recommendation accuracy very low.

[0003] Therefore, there is an urgent need for a recommendation method that can deeply integrate multi-dimensional data, dynamically optimize the recommendation strategy, and have high scene adaptability to solve the core defects of insufficient personalization and inaccurate and incomplete recommendation solutions in the prior art. Summary of the Invention

[0004] Embodiments of the present application provide a personalized tourist scenic area recommendation method and system based on a knowledge graph, which are used to solve the problems of insufficient personalization and inaccuracy and incompleteness of scenic area recommendation solutions.

[0005] The first aspect of the embodiments of the present application provides a personalized tourist scenic area recommendation method based on a knowledge graph, including:

[0006] Collect user behavior data, scenic area attribute data, and spatio-temporal information data corresponding to preset scenic areas;

[0007] Construct a dynamically updated knowledge graph based on the data. The nodes of the knowledge graph include user entities, scenic area entities, time entities, and space entities, and the edges of the knowledge graph include the interaction relationship between users and scenic areas, the attribute association relationship between scenic areas, and the spatio-temporal constraint relationship;

[0008] Collect the A3C algorithm to dynamically optimize the knowledge graph. The dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. Each Actor thread's operations include the Actor network generating a recommendation strategy, the Critic network evaluating the strategy value, and adjusting the weights of the nodes and edges in the knowledge graph;

[0009] Dynamically adjust the A3C algorithm parameters according to the geographical characteristics of the preset scenic areas. The parameters include the exploration rate, the reward function weight, and the network learning rate;

[0010] Generate a personalized recommendation solution adapted to the user's needs based on the optimized knowledge graph and parameter configuration.

[0011] Furthermore, construct a dynamically updated knowledge graph based on the data. The nodes of the knowledge graph include user entities, scenic spot entities, time entities, and space entities. The edges of the knowledge graph include the interaction relationship between users and scenic spots, the attribute association relationship between scenic spots, and the spatio-temporal constraint relationship, including:

[0012] The user entity includes static attributes and dynamic attributes. The scenic spot entity includes static attributes and dynamic attributes. The time entity includes time granularity and associated attributes. The space entity includes regional division and dynamic attributes.

[0013] Furthermore, construct a dynamically updated knowledge graph based on the data. The nodes of the knowledge graph include user entities, scenic spot entities, time entities, and space entities. The edges of the knowledge graph include the interaction relationship between users and scenic spots, the attribute association relationship between scenic spots, and the spatio-temporal constraint relationship, including:

[0014] The relationship types of the interaction relationship between users and scenic spots include preference relationship, visit relationship, and consumption relationship. The relationship types of the attribute association relationship between scenic spots include type similarity, cultural feature matching, and theme complementarity. The spatio-temporal constraint relationship includes time dependence, geographical proximity, and traffic accessibility.

[0015] Furthermore, use the A3C algorithm to dynamically optimize the knowledge graph. The dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. The operations performed by each Actor thread include the Actor network generating a recommendation strategy, the Critic network evaluating the value of the strategy, and adjusting the weights of the nodes and edges in the knowledge graph, including:

[0016] Convert the knowledge graph into an interactive environment for reinforcement learning, and determine the mapping relationship between state elements, action spaces, and rewards;

[0017] Explore the nodes and edges in the knowledge graph in parallel through multiple asynchronous Actor threads;

[0018] Evaluate the long-term value of actions through the Critic network;

[0019] Optimize the weights of the nodes and edges in the knowledge graph according to the feedback of reinforcement learning.

[0020] Furthermore, the exploration of the nodes and edges in the knowledge graph in parallel through multiple asynchronous Actor threads includes:

[0021] Randomly select a current user state from the knowledge graph;

[0022] Output the action probability distribution through the Actor network, and select an action according to the action probability distribution;

[0023] Execute the action and simulate the user feedback, and calculate the immediate reward;

[0024] Update the user state to the next state, and store the user action trajectory in the thread local buffer.

[0025] Furthermore, the evaluation of the long-term value of the action by the Critic network includes:

[0026] The calculation formula of the advantage function is as follows:

[0027]

[0028] Where: a t is the action selected according to the action probability distribution, S t is the current state of the user, V(S T ; φ i ) is the value evaluated for the current state S t of the user, φ i is the Critic network parameter, γ k and γ T-t are the discount factors corresponding to the k and T-t moments respectively, T is the number of steps at which the trajectory terminates, r t+k is the reward at the t + k moment.

[0029] Furthermore, the optimization of the node and edge weights in the knowledge graph according to the reinforcement learning feedback includes:

[0030] Determine the explicit association weight and implicit association weight of the node and the edge, and update the explicit association weight and implicit association weight according to the time decay and the emergency response.

[0031] Furthermore, the dynamic adjustment of the A3C algorithm parameters according to the geographical characteristics of the preset scenic area, the parameters include the exploration rate, the reward function weight, and the network learning rate, includes:

[0032] The calculation formula of the exploration rate:

[0033]

[0034] Where: ∈ t is the exploration rate at the t moment, ∈ base is the basic exploration rate, N current is the number of tourists in the scenic area monitored in real time, N max is the safety carrying capacity of the scenic area, F season is the seasonal decay factor;

[0035] The calculation formula of the reward function weight:

[0036] R total = w base × R click + w season × R order + w culture × R stay

[0037] Where: R total is the total reward value, w base is the basic weight, R click is the reward value for the user to click on the recommended scenic spot, w season is the seasonal weight adjustment factor, R order is the reward value for the user to place an order for the recommended scenic spot, w culture is the cultural matching weight, R stay is the reward value for the user's stay time exceeding 20% of the average value of the scenic spot;

[0038] The calculation formula of the network learning rate:

[0039]

[0040] Where: α t is the network learning rate at time t, α base is the basic learning rate, N benchmark is the historical average hourly data volume, N current-hour is the real-time statistical user behavior data volume, L critic is the value evaluation error of the current Critic network, L critic-initial is the initial Critic loss.

[0041] Furthermore, the personalized recommendation solution adapted to the user's needs generated based on the optimized knowledge graph and parameter configuration includes:

[0042] Retrieving candidate scenic spots and associated paths that match the user's needs through the optimized knowledge graph;

[0043] Generating a multi-dimensional scenic spot recommendation solution including user satisfaction, scenic spot revenue, and traffic efficiency according to reinforcement learning.

[0044] The second aspect of the embodiments of the present application provides a personalized tourist scenic spot recommendation system based on a knowledge graph, including:

[0045] A data collection unit for collecting user behavior data, scenic spot attribute data, and spatio-temporal information data corresponding to preset scenic spots;

[0046] A knowledge graph update unit, which is used to construct a dynamically updated knowledge graph based on the data. The nodes of the knowledge graph include user entities, scenic spot entities, time entities, and space entities, and the edges of the knowledge graph include the interaction relationship between users and scenic spots, the attribute association relationship between scenic spots, and the spatio-temporal constraint relationship;

[0047] A knowledge graph dynamic optimization unit, which is used to collect the A3C algorithm to perform dynamic optimization on the knowledge graph. The dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. The operations performed by each Actor thread include the Actor network generating a recommendation strategy, the Critic network evaluating the value of the strategy, and adjusting the weights of the nodes and edges in the knowledge graph;

[0048] An A3C algorithm parameter adjustment unit, which is used to dynamically adjust the A3C algorithm parameters according to the regional characteristics of the preset scenic spots. The parameters include the exploration rate, the reward function weight, and the network learning rate;

[0049] A personalized recommendation solution generation unit, which is used to generate a personalized recommendation solution adapted to the user's needs based on the optimized knowledge graph and parameter configuration.

[0050] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0051] The present invention deeply integrates multi-source heterogeneous data of user behavior, scenic spot attributes, and spatio-temporal information to depict the association between user behavior and scenic spots, and can provide a comprehensive data basis for subsequent knowledge graph construction and optimization; uses multi-threaded parallel mining in the A3C algorithm to explore the nodes and edges in the knowledge graph. In each Actor thread, the Actor network generates a recommendation strategy, the Critic network evaluates the value of the strategy, and adjusts the weights of the nodes and edges in the knowledge graph to ensure the dynamics and real-time nature of the recommendation strategy; adjusts the parameters according to regional characteristics to make the recommendation results more in line with the needs of the preset scenic spots; finally, the multi-objective optimization algorithm balances user preferences, scenic spot revenues, and traffic efficiency to generate a Pareto-optimal route, effectively improving the recommendation accuracy and user satisfaction. Description of the Drawings

[0052] Figure 1 It is a schematic flowchart of an embodiment of a personalized tourist scenic spot recommendation method based on a knowledge graph in the present invention. Detailed Embodiments

[0053] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] In this embodiment, the electricity consumption prediction method based on climate change is used to improve the efficiency and accuracy of electricity consumption prediction under climate change. The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal, and no specific limitation is made.

[0055] Embodiment 1

[0056] Please refer to Figure 1 , an embodiment of a personalized tourist scenic area recommendation method based on a knowledge graph in the present invention includes the following steps:

[0057] S11. Collect user behavior data, scenic area attribute data, and spatio-temporal information data corresponding to a preset scenic area;

[0058] In this embodiment, the user behavior data includes the user's historical visit records, ratings, comments, and real-time interaction behaviors; taking a scenic area in Guangxi as an example, through the "One-Click Tour of Guangxi" platform API, the user's encrypted ID, browsing path such as "User A → Guilin Lijiang River → Yangshuo West Street", the stay duration accurate to the second, ratings, comment texts, and real-time interaction behaviors such as clicks, collections, and shares are obtained in real time. The scenic area attribute data includes the scenic area type, geographical location, opening hours, ticket price, and cultural characteristics; the scenic area name, category labels including natural, cultural, theme, geographical location involving longitude and latitude, opening hours, ticket price, capacity limit of the maximum carrying capacity, and cultural characteristics such as Zhuang culture and cross-border tourism labels are extracted from the scenic area management system. The spatio-temporal information includes the user's access time, season, weather, and traffic status; in terms of the time dimension, it includes the user access timestamp accurate to minutes, season labels for off-peak or peak seasons, holiday labels such as National Day and the Third Month Third Song Festival, weather data such as sunny, rainy, temperature, and humidity, etc.; in terms of the space dimension, it includes the real-time traffic status determined by obtaining the congestion index through the map API, the geographical distance between scenic areas calculated based on longitude and latitude, and regional divisions such as Guilin City and Yangshuo County.

[0059] S12. Construct a dynamically updated knowledge graph based on the data. The nodes of the knowledge graph include user entities, scenic area entities, time entities, and space entities, and the edges of the knowledge graph include the interaction relationship between the user and the scenic area, the attribute association relationship between scenic areas, and the spatio-temporal constraint relationship;

[0060] In this embodiment, the above acquired data is cleaned and standardized, wherein the denoising process includes filtering invalid data, such as 0-second stay records caused by user accidental touches; filling missing values, including when the stay duration is missing, using the average stay duration of similar scenic spots to complete; when the scenic spot attributes are missing, calling a third-party database (such as Amap POI) to supplement the information. In addition, privacy protection processing is also required, including user anonymization and sensitive information desensitization, such as using the SHA-256 hash algorithm to encrypt the user ID, generate an irreversible unique identifier, and hide personal sensitive fields such as the user's mobile phone number and email address.

[0061] Then, the preprocessed data is mapped to entities and attributes of the knowledge graph to establish a structured representation. The specific entity types are divided as follows:

[0062] The user entity uses the encrypted user ID as the core identifier, including static attributes and dynamic attributes. The static attributes include registration time, membership level, etc.; the dynamic attributes include preference tags (natural scenery preference 0.8, cultural experience preference 0.6), historical behavior sequence ("User A→Guilin Lijiang→Yangshuo West Street"), etc. The scenic spot entity uses the scenic spot ID as the core identifier, including static attributes and dynamic attributes. The static attributes include category (nature / culture / theme), capacity, cultural feature tags (such as Zhuang culture, Yao festivals), longitude and latitude; dynamic attributes include real-time number of tourists, average rating (dynamically calculated based on user ratings). The time entity uses timestamp or time period as the identifier, including time granularity and associated attributes. The time granularity is accurate to the time node of the hour (such as 2023-10-01T15:00), time period label (such as "National Day holiday" and "rainy season"); associated attributes include weather status (sunny / rainy) and holiday mark (yes / no). Spatial entities use geographic regions or coordinates as core identifiers, including regional divisions and dynamic attributes. Regional divisions include administrative regions (such as Nanning City) and geographic clusters (such as the Nanning-Xixiangtang District tourist belt); dynamic attributes include real-time traffic congestion index (0-10) and public transportation accessibility score.

[0063] The attribute weights are calculated based on the above entity type classification. For example, the user preference weight is calculated based on the product of the user's stay time in the scenic spot and the order conversion rate. The scenic spot popularity weight is calculated based on the product of the recent visit volume and the average rating.

[0064] After determining the entity type, we further determine the entity relationship and calculate the edge weights as follows:

[0065] The relationship types of the interaction between users and scenic spots include preference relationship, visit relationship and consumption relationship. Among them, the preference weight of the preference relationship is determined according to the stay duration of the user in the scenic spot and the order conversion rate, the visit frequency weight is calculated based on the normalized value of the number of times the user visits the same scenic spot, and the consumption weight is determined according to the proportion of the order amount in the total consumption. The relationship types of the attribute association relationship between scenic spots include type similarity, cultural feature matching and theme complementarity. Among them, the type similarity weight calculates the category overlap degree according to the Jaccard coefficient, the cultural matching weight is determined based on the co-occurrence frequency of tags (such as the co-occurrence times of the "Zhuang culture" scenic spot group / the total number of scenic spots), and the theme complementarity weight is obtained through user behavior path analysis (such as often visiting Scenic Spot B after Scenic Spot A). The spatio-temporal constraint relationship includes time dependence, geographical proximity and traffic accessibility. Among them, the time dependence weight is determined according to the behavior frequency within a specific time period (such as the visit volume on weekends is twice that on weekdays), the geographical proximity weight is determined according to the ratio of two scenic spots, and the traffic accessibility weight is calculated based on the real-time traffic time (such as the weight is reduced by 30% during congestion).

[0066] Finally, an initial knowledge graph is constructed based on the determined entity types and the relationships between entities, and it is ensured that the graph reflects the latest data and user behavior changes in real time.

[0067] S13. Use the A3C algorithm to dynamically optimize the knowledge graph. The dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. The operations performed by each Actor thread include the Actor network generating a recommendation strategy, the Critic network evaluating the value of the strategy, and adjusting the weights of the nodes and edges in the knowledge graph.

[0068] S131. Convert the knowledge graph into an interactive environment for reinforcement learning, and determine the mapping relationships of state elements, action spaces and rewards.

[0069] The state elements include user state, time state and space state. The user state is the current user location, historical behavior sequence, and preference tags. The time state is the current time period and weather tags. The space state is the longitude and latitude of the user's current location and the real-time traffic congestion index. Encode these state elements into vectors. The action space includes action types and action encodings. The action type is to recommend candidate scenic spots, and the action encoding is to map the scenic spot ID to a discrete action number.

[0070] S132. Explore the nodes and edges in the knowledge graph in parallel through multiple asynchronous Actor threads.

[0071] Step S132 includes the following:

[0072] 1. Randomly select a current user state from the knowledge graph.

[0073] 2. Output the action probability distribution through the Actor network, and select an action according to the action probability distribution;

[0074] 3. Execute the action and simulate the user feedback, and calculate the immediate reward;

[0075] 4. Update the user state to the next state, and store the user action trajectory in the thread local buffer.

[0076] S133. Evaluate the long-term value of the action through the Critic network;

[0077] Specifically, the output of the Critic network is the value of the current state, and the calculation formula of the advantage function is as follows:

[0078]

[0079] Where: a t is the action selected according to the action probability distribution, S t is the current state of the user, V(S T ; φ i ) is the value evaluated for the current state S t of the user, φ i is the Critic network parameter, γ k and γ T-t are the discount factors corresponding to the k and T-t moments respectively, T is the number of steps at which the trajectory terminates, r t+k is the reward at the t + k moment.

[0080] Calculate the Actor loss by maximizing the expectation of the advantage function, calculate the Critic loss by minimizing the value estimation error, calculate the gradient for each thread, and asynchronously update the global network parameters.

[0081] S134. Optimize the node and edge weights in the knowledge graph according to the reinforcement learning feedback.

[0082] Step S134 includes:

[0083] Determine the explicit association weight and implicit association weight of the node and the edge, and update the explicit association weight and implicit association weight according to the time decay and the emergency response

[0084] Finally, coordinate the exploration results of multiple threads to ensure the stable convergence of the model.

[0085] S14. Dynamically adjust the A3C algorithm parameters according to the regional characteristics of the preset scenic area, and the parameters include the exploration rate, the reward function weight, and the network learning rate;

[0086] Adjust the exploration rate to balance the exploration and exploitation trade-off, adapt to the seasonal traffic changes in the tourism scenario, and the exploration rate calculation formula:

[0087]

[0088] Wherein: ∈ t is the exploration rate at time t, ∈ base is the basic exploration rate, N current is the number of tourists in the scenic area monitored in real time, N max is the safety carrying capacity of the scenic area, F season is the seasonal attenuation factor.

[0089] Adjust the reward function weight to differentially define the reward value according to features such as climate and cultural festivals, and improve the adaptability of the recommendation scenario. The calculation formula of the reward function weight is:

[0090] R total = w base ×R click + w season ×R order + w culture ×R syay

[0091] Wherein: R total is the total reward value, w base is the basic weight, R click is the reward value for the user to click on the recommended scenic area, w season is the seasonal weight adjustment factor, R order is the reward value for the user to place an order for the recommended scenic area, w culture is the cultural matching weight, R stay is the reward value when the user's stay time exceeds 20% of the average value of the scenic area.

[0092] Adjust the network learning rate to dynamically control the parameter update speed according to the real-time data traffic and the model convergence state, and prevent overfitting or underfitting. The calculation formula of the network learning rate is:

[0093]

[0094] Wherein: α t is the network learning rate at time t, α base is the basic learning rate, N benchmark is the historical average hourly data volume, N current-hour is the real-time statistical user behavior data volume, L critic is the value evaluation error of the current Critic network, L critic-initial is the initial Critic loss.

[0095] S15. Generate a personalized recommendation solution that adapts to the user's needs based on the optimized knowledge graph and parameter configuration.

[0096] S151. Retrieve candidate scenic spots and associated paths that match the user's needs through the optimized knowledge graph;

[0097] First, extract the user's historical visit paths, stay duration distributions, and rating preferences at scenic spots; analyze the click, favorite, and share behaviors in the current session; and use an NLP model to extract the sentiment polarity in the review text. Then, analyze the user profile based on the extracted data information, such as membership level, registration time, basic preferences, real-time location, and current intent. For example, it is obtained that user A has a historical preference for natural landscapes, recent reviews show an increased interest in ethnic culture, is currently located in Nanning, and the weather is rainy. Then, retrieve candidate scenic spots and associated paths that match the user's needs through the optimized knowledge graph, excluding candidates that are more than 50 kilometers away from the current scenic spot, preferentially recommending indoor scenic spots during the rainy season, and binding the opening hours.

[0098] S152. Generate a multi-dimensional scenic spot recommendation plan that includes user satisfaction, scenic spot revenue, and traffic efficiency based on reinforcement learning.

[0099] Generate high-value recommendation strategies through reinforcement learning and quantify their long-term benefits. Here, the user satisfaction aspect is determined by maximizing the sum of the preference weights of the recommended scenic spots, the scenic spot revenue aspect is determined by balancing the recommended proportions of popular and niche scenic spots, and the traffic efficiency aspect is determined by minimizing the total traffic time of the route. Construct an objective function with these three aspects as target parameters, and select the route with the highest comprehensive score in the Pareto front as the optimal route. Finally, present the recommendation results in a user-friendly form, support dynamic interaction and feedback, and continuously optimize the recommendation system through user behavior feedback.

[0100] The above optimized knowledge graph is deeply integrated with the A3C algorithm, realizing the full-chain intelligence from data to recommendation, and providing an efficient and accurate technical solution for smart tourism.

[0101] Example 4

[0102] An embodiment of a personalized tourist scenic spot recommendation system based on a knowledge graph in the present invention includes the following steps:

[0103] A data collection unit for collecting user behavior data, scenic spot attribute data, and spatio-temporal information data corresponding to preset scenic spots;

[0104] A knowledge graph update unit for constructing a dynamically updated knowledge graph based on the data. The nodes of the knowledge graph include user entities, scenic spot entities, time entities, and space entities, and the edges of the knowledge graph include the interaction relationship between users and scenic spots, the attribute association relationship between scenic spots, and the spatio-temporal constraint relationship;

[0105] A knowledge graph dynamic optimization unit is used to collect the A3C algorithm to perform dynamic optimization on the knowledge graph. The dynamic optimization includes exploring nodes and edges in the knowledge graph through multiple asynchronous Actor threads. Each Actor thread performs operations including an Actor network generating a recommendation strategy, a Critic network evaluating the strategy value, and adjusting the weights of nodes and edges in the knowledge graph;

[0106] An A3C algorithm parameter adjustment unit is used to dynamically adjust the A3C algorithm parameters according to the geographical characteristics of a preset scenic area. The parameters include an exploration rate, a reward function weight, and a network learning rate;

[0107] A personalized recommendation solution generation unit is used to generate a personalized recommendation solution adapted to user needs based on the optimized knowledge graph and parameter configuration.

[0108] Those of ordinary skill in the art can realize that the units of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0109] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc. In addition, the functional units in each embodiment of the present invention can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0110] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), external hard drives, magnetic disks, or optical discs that can store program codes.

[0111] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A personalized tourist attraction recommendation method based on knowledge graph, characterized in that: include: Collect user behavior data, scenic spot attribute data and time-space information data corresponding to the preset scenic spots; Based on the data, a dynamically updated knowledge graph is constructed, wherein the nodes of the knowledge graph include user entities, scenic spot entities, time entities and space entities, and the edges of the knowledge graph include user-scenic spot interaction relationships, attribute association relationships between scenic spots and time-space constraint relationships; The A3C algorithm is used to dynamically optimize the knowledge graph. The dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. Each Actor thread performs operations including generating recommendation strategies through the Actor network, evaluating the strategy value through the Critic network, and adjusting the weights of nodes and edges in the knowledge graph. Dynamically adjusting the A3C algorithm parameters according to the regional characteristics of the preset scenic spot, the parameters including exploration rate, reward function weight and network learning rate; Generate personalized recommendation solutions that meet user needs based on the optimized knowledge graph and parameter configuration.

2. The personalized tourist attraction recommendation method based on knowledge graph according to claim 1 is characterized in that: The dynamically updated knowledge graph is constructed based on the data, the nodes of the knowledge graph include user entities, scenic spot entities, time entities and space entities, and the edges of the knowledge graph include user-scenic spot interaction relationships, attribute association relationships between scenic spots and time-space constraint relationships, including: The user entity includes static attributes and dynamic attributes, the scenic area entity includes static attributes and dynamic attributes, the time entity includes time granularity and associated attributes, and the space entity includes area division and dynamic attributes.

3. The personalized tourist attraction recommendation method based on knowledge graph according to claim 2 is characterized in that: The dynamically updated knowledge graph is constructed based on the data, the nodes of the knowledge graph include user entities, scenic spot entities, time entities and space entities, and the edges of the knowledge graph include user-scenic spot interaction relationships, attribute association relationships between scenic spots and time-space constraint relationships, including: The relationship types of the interactive relationship between users and scenic spots include preference relationship, visit relationship and consumption relationship; the relationship types of the attribute association relationship between scenic spots include type similarity, cultural feature matching and theme complementarity; the time and space constraint relationship includes time dependence, geographical proximity and traffic accessibility.

4. The personalized tourist attraction recommendation method based on knowledge graph according to claim 1 is characterized in that: The acquisition A3C algorithm dynamically optimizes the knowledge graph, and the dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. Each Actor thread performs operations including generating recommendation strategies by the Actor network, evaluating the strategy value by the Critic network, and adjusting the weights of nodes and edges in the knowledge graph, including: Convert the knowledge graph into an interactive environment for reinforcement learning and determine the mapping relationship between state elements, action space and rewards; Explore nodes and edges in the knowledge graph in parallel through multiple asynchronous Actor threads; Evaluate the long-term value of actions through the Critic network; Optimize node and edge weights in knowledge graphs based on reinforcement learning feedback.

5. The personalized tourist attraction recommendation method based on knowledge graph according to claim 4 is characterized in that: The parallel exploration of nodes and edges in the knowledge graph through multiple asynchronous Actor threads includes: Randomly select a user's current status from the knowledge graph; Outputting the action probability distribution through the Actor network, and selecting an action according to the action probability distribution; Execute actions and simulate user feedback to calculate immediate rewards; Update the user state to the next state and store the user action trajectory in the thread local buffer.

6. The personalized tourist attraction recommendation method based on knowledge graph according to claim 4 is characterized in that: The long-term value of the action evaluated by the Critic network includes: The calculation formula of the advantage function is as follows: Among them: a t is the action selected according to the action probability distribution, S t is the user's current state, V(S T ; φ i ) is the current state S of the user t The value of the assessment, φ i is the Critic network parameter, γ k and γ T-t are the discount factors corresponding to time k and Tt respectively, T is the number of steps for trajectory termination, r t+k is the reward at time t+k.

7. The personalized tourist attraction recommendation method based on knowledge graph according to claim 4 is characterized in that: The optimizing the node and edge weights in the knowledge graph according to the reinforcement learning feedback includes: The explicit association weights and implicit association weights of nodes and edges are determined, and the explicit association weights and implicit association weights are updated according to time decay and emergency response.

8. The personalized tourist attraction recommendation method based on knowledge graph according to claim 1 is characterized in that: The A3C algorithm parameters are dynamically adjusted according to the regional characteristics of the preset scenic spot, the parameters include exploration rate, reward function weight and network learning rate, including: Exploration rate calculation formula: Where: ∈ t is the exploration rate at time t, ∈ base is the basic exploration rate, N current is the number of tourists in the scenic area monitored in real time, N max is the safe carrying capacity of the scenic area, F season is the seasonal attenuation factor; Reward function weight calculation formula: R total =w base ×R click +w season ×R order +w culture ×R stay Where: R total is the total reward value, w base is the basic weight, R click The reward value for users clicking on the recommended scenic spot, w season is the seasonal weight adjustment factor, R order The reward value for recommending scenic spots for users to place orders, w culture is the cultural matching weight, R stay The reward value is when the user's stay time exceeds 20% of the average value of the scenic spot; Network learning rate calculation formula: Where: α t is the network learning rate at time t, α base is the basic learning rate, N benchmark is the historical average hourly data volume, N current-hour The amount of user behavior data counted in real time, L critic is the value assessment error of the current Critic network, L critic-initial is the initial Critic loss.

9. The personalized tourist attraction recommendation method based on knowledge graph according to claim 1 is characterized in that: The method of generating a personalized recommendation solution adapted to user needs based on the optimized knowledge graph and parameter configuration includes: Retrieve candidate scenic spots and related paths that match user needs through the optimized knowledge graph; Based on reinforcement learning, a multi-dimensional scenic spot recommendation plan is generated that includes user satisfaction, scenic spot revenue, and traffic efficiency.

10. A personalized tourist attraction recommendation system based on knowledge graph, characterized in that: The method for recommending personalized tourist attractions based on knowledge graphs according to any one of claims 1 to 9 comprises: A data collection unit, used to collect user behavior data, scenic spot attribute data and time-space information data corresponding to a preset scenic spot; A knowledge graph updating unit, used to construct a dynamically updated knowledge graph based on the data, wherein the nodes of the knowledge graph include user entities, scenic spot entities, time entities and space entities, and the edges of the knowledge graph include user-scenic spot interaction relationships, attribute association relationships between scenic spots and time-space constraint relationships; The knowledge graph dynamic optimization unit is used to collect the A3C algorithm to dynamically optimize the knowledge graph. The dynamic optimization includes exploring the nodes and edges in the knowledge graph through multiple asynchronous Actor threads. Each Actor thread performs operations including generating recommendation strategies through the Actor network, evaluating the strategy value through the Critic network, and adjusting the weights of nodes and edges in the knowledge graph. An A3C algorithm parameter adjustment unit, used to dynamically adjust the A3C algorithm parameters according to the regional characteristics of the preset scenic spot, the parameters including exploration rate, reward function weight and network learning rate; The personalized recommendation scheme generation unit is used to generate a personalized recommendation scheme adapted to user needs based on the optimized knowledge graph and parameter configuration.

Images