Tourism destination recommendation method and system based on big data and computer equipment

Abstract

The invention relates to a travel destination recommendation method and system based on big data and computer equipment, and the method comprises the steps: collecting the original data of travel history, browsing history and social media activities of a user from multiple channels, and constructing a user behavior data warehouse after preprocessing; key features reflecting user travel modes and preferences are extracted from a user behavior data warehouse, user feature vectors are constructed according to the key features, and a candidate destination recommendation list is generated through a mixed recommendation model. And constructing a feature vector according to geographic positions, climate conditions and resource endowment data of the candidate destinations, obtaining an optimal access time period through an ecological influence evaluation model, and matching with a recommendation list to screen out a proper destination. And finally, in combination with multiple preferences of the user, a complete travel plan is generated by using a reinforcement learning algorithm, and the purposes of accurate recommendation, ecological protection and real-time dynamic adjustment are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of tourism big data, and in particular to a tourist destination recommendation method, system and computer equipment based on big data. Background Art

[0002] As people's living standards improve, travel has become a choice of leisure and entertainment for more and more people. However, faced with numerous tourist destinations and complex and diverse travel needs, it is often difficult for tourists to quickly and accurately find tourist destinations that meet their preferences and actual conditions and plan appropriate travel plans.

[0003] Traditional tourist destination recommendation methods are mostly based on simple user surveys, experience recommendations or basic search filtering techniques. These methods have many limitations and cannot make full use of user behavior data to deeply understand their travel patterns and preferences, resulting in stereotyped recommendation results that cannot meet the personalized needs of users. For example, only popular attractions are recommended without considering users' unique travel time, consumption level, accommodation preferences and other factors. In addition, in the recommendation process, the ecological carrying capacity of the destination and the real-time actual supply situation are rarely considered. This may cause tourists to damage the local environment when traveling to ecologically fragile areas, and may also affect the travel experience due to resource shortages (such as full hotels, scenic spots, etc.). With the development of big data technology, it is possible to solve these problems, but at present, mature solutions that deeply apply big data technology to tourist destination recommendations and comprehensively consider multiple factors still need to be improved. Summary of the invention

[0004] The main purpose of the present invention is to provide a tourist destination recommendation method, system and computer equipment based on big data, so as to achieve the purpose of accurately recommending tourist destinations, generating complete travel plans, taking into account ecological protection and real-time dynamic adjustment of travel plans.

[0005] To achieve the above object, the present invention provides a tourist destination recommendation method based on big data, comprising the following steps:

[0006] Collecting raw data of user travel history, browsing history, and social media activities from multiple channels, preprocessing the raw data, and building a user behavior data warehouse;

[0007] Extracting features from the data in the user behavior data warehouse to obtain key features that reflect the user's travel patterns and preferences;

[0008] Constructing a user feature vector based on the extracted key features, and generating a candidate destination recommendation list by adopting a hybrid recommendation model combining a collaborative filtering algorithm and a content-based recommendation algorithm;

[0009] Based on the geographical location, climate conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, construct a destination feature vector, evaluate the ecological carrying capacity of various candidate destinations at different times based on a pre-constructed ecological impact assessment model, and obtain the best visiting period for the destination;

[0010] Matching the candidate destination recommendation list with the optimal visit period, screening and sorting destinations with suitable carrying capacity within the user's intended travel period, and obtaining destination recommendation results;

[0011] Based on the destination recommendation results, combined with the user's time budget, travel mode preferences, accommodation preferences and activity preferences, a reinforcement learning algorithm is used to automatically generate the optimal itinerary route, transportation mode, accommodation and activity arrangements to form a complete user travel plan.

[0012] Furthermore, when collecting user data, a distributed crawler architecture is adopted, and in the process of building a user behavior data warehouse, data encryption technology is used to encrypt and store sensitive data.

[0013] Furthermore, the collaborative filtering algorithm calculates the similarity between users based on an improved cosine similarity algorithm, and the content-based recommendation algorithm uses a deep learning model to deeply mine and match user historical behaviors and preference features. In the content-based recommendation algorithm, the deep learning model uses a structure that combines a convolutional neural network and a long short-term memory network to process the spatiotemporal characteristics in user behavior data.

[0014] Furthermore, the destination feature vectors are classified by using an improved clustering algorithm, wherein the improved clustering algorithm introduces a density peak clustering concept for automatically identifying cluster centers.

[0015] Furthermore, when evaluating the ecological carrying capacity of candidate destinations, the impact of tourists' behavior on the ecological environment is comprehensively considered, and the ecological carrying capacity is evaluated by establishing an ecological impact assessment model.

[0016] Furthermore, in the process of generating user travel plans, knowledge graph technology is used to integrate and associate various types of information about the destination.

[0017] Furthermore, after the step of forming a complete user travel plan, the process further includes:

[0018] Based on the actual supply situation of the real-time destination, the user's travel plan is dynamically adjusted through a real-time optimization algorithm. The real-time optimization algorithm is based on big data real-time data feedback and prediction models to warn and adjust possible resource shortages in advance.

[0019] The present invention also provides a tourist destination recommendation system based on big data, comprising:

[0020] The data acquisition unit is used to collect the original data of user travel history, browsing history and social media activities from multiple channels, pre-process the original data, and build a user behavior data warehouse;

[0021] A feature extraction unit, used to extract features from the data in the user behavior data warehouse to obtain key features reflecting the user's travel patterns and preferences;

[0022] A candidate destination analysis unit, configured to construct a user feature vector based on the extracted key features, and generate a candidate destination recommendation list by adopting a hybrid recommendation model combining a collaborative filtering algorithm and a content-based recommendation algorithm;

[0023] An ecological analysis unit, configured to construct a destination feature vector based on the geographical location, climate conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, evaluate the ecological carrying capacity of various candidate destinations at different times based on a pre-constructed ecological impact assessment model, and obtain the best visiting period for the destination;

[0024] A destination generation unit, used to match the candidate destination recommendation list with the optimal visit period, screen and sort the destinations with suitable carrying capacity within the user's intended travel period, and obtain a destination recommendation result;

[0025] The result output unit is used to automatically generate the optimal itinerary route, transportation method, accommodation and activity arrangement based on the destination recommendation results, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, using a reinforcement learning algorithm to form a complete user travel plan.

[0026] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned tourist destination recommendation method based on big data when executing the computer program.

[0027] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned tourist destination recommendation method based on big data are implemented.

[0028] The tourist destination recommendation method, system and computer device based on big data provided by the present invention have the following beneficial effects:

[0029] The present invention collects and deeply analyzes user data through multiple channels, constructs key features and user feature vectors, and uses a hybrid recommendation model to achieve highly personalized travel destination recommendations, saving user screening time. Through the deep learning model combined with the convolutional neural network and the long short-term memory network, the spatiotemporal characteristics are effectively processed, the user's potential travel preferences are excavated, the clustering algorithm is improved to incorporate the density peak clustering idea, the cluster center is automatically identified, and the destination is scientifically classified, thereby improving the matching accuracy and recommendation efficiency. In addition, by taking into account the impact of tourist behavior when evaluating the ecological carrying capacity of candidate destinations, a model is established to guide sustainable tourism, which helps to protect the ecological environment. The knowledge graph is used to integrate destination information, and a complete travel plan covering itinerary, transportation, accommodation and activities is automatically generated, which improves the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of a tourist destination recommendation method based on big data in one embodiment of the present invention;

[0031] Figure 2 is a structural block diagram of a tourist destination recommendation system based on big data in one embodiment of the present invention;

[0032] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0033] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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 intended to limit the present invention.

[0035] Reference Figure 1 , which is a flow chart of a tourist destination recommendation method based on big data proposed by the present invention, comprising the following steps:

[0036] S1, collecting raw data of user travel history, browsing history and social media activities from multiple channels, pre-processing the raw data, and building a user behavior data warehouse;

[0037] S2, extracting features from the data in the user behavior data warehouse to obtain key features reflecting the user's travel patterns and preferences, wherein the key features include but are not limited to the type of destinations frequently visited by the user, travel time and frequency, length of stay, consumption level, travel mode, accommodation preference, and activity preference;

[0038] S3, constructing a user feature vector based on the extracted key features, and generating a candidate destination recommendation list by adopting a hybrid recommendation model combining a collaborative filtering algorithm and a content-based recommendation algorithm;

[0039] S4, constructing a destination feature vector based on the geographical location, climate conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, evaluating the ecological carrying capacity of various candidate destinations at different times based on a pre-constructed ecological impact assessment model, and obtaining the best visiting period for the destination;

[0040] S5, matching the candidate destination recommendation list with the optimal visit period, screening and sorting the destinations with suitable carrying capacity within the user's intended travel period, and obtaining a destination recommendation result;

[0041] S6, based on the destination recommendation results, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, a reinforcement learning algorithm is used to automatically generate the optimal itinerary route, transportation method, accommodation and activity arrangement to form a complete user travel plan. The reinforcement learning algorithm adopts a multi-agent reinforcement learning framework, and each agent is responsible for optimizing different parts of the itinerary, and the overall plan is carried out through collaboration between agents.

[0042] As described in step S1 above, the original data of user travel history, browsing history and social media activities are collected from multiple channels, the original data are preprocessed, and a user behavior data warehouse is constructed. From multiple different data sources, such as online travel platform records, user personal travel logs, browser browsing history, and travel-related dynamics shared on social media, the original data of user travel history, browsing history and social media activities are widely collected. These data are the basis for subsequent analysis, and multiple channels ensure the comprehensiveness of the data. The collected raw data is preprocessed, which may include data cleaning (removing duplicate, erroneous or incomplete data), data conversion (unifying the data format for subsequent processing), and other operations. Then, the preprocessed data is constructed into a user behavior data warehouse for centralized management and subsequent data mining analysis.

[0043] As described in step S2 above, feature extraction is performed on the data in the user behavior data warehouse to obtain key features that reflect the user's travel mode and preference, including but not limited to the type of destinations frequently visited by users, travel time and frequency, length of stay, consumption level, travel mode, accommodation preference and activity preference. Based on the constructed user behavior data warehouse, specific data mining and analysis techniques are used to extract data features that can reflect the user's travel mode and preference. These key features cover multiple aspects. The types of destinations frequently visited by users (such as coastal cities, historical and cultural cities, etc.) can reflect the user's preference for the nature of the destination; travel time and frequency reflect the user's travel time pattern; length of stay is related to the user's deep experience demand for the destination; consumption level indicates the user's consumption tendency during travel; travel mode (such as airplane, train, etc.), accommodation preference (such as hotel star, homestay type, etc.) and activity preference (such as outdoor sports, cultural visits, etc.) outline the user's travel preference portrait from different dimensions.

[0044] As described in step S3 above, a user feature vector is constructed based on the extracted key features, and a candidate destination recommendation list is generated by adopting a hybrid recommendation model that combines a collaborative filtering algorithm with a content-based recommendation algorithm. The extracted key features are quantified and organized to form a user feature vector for processing and analysis. The user feature vector is a mathematical representation of the user's travel preferences. A candidate destination recommendation list is generated by combining a collaborative filtering algorithm with a content-based recommendation algorithm. Among them, the collaborative filtering algorithm is based on an improved cosine similarity algorithm. By calculating the similarity between different user feature vectors, other users with similar interests to the target user are found, and the destination is recommended with reference to the travel experiences of these similar users. The content-based recommendation algorithm uses a deep learning model to deeply mine the user's historical behavior and preference characteristics, accurately match the destination that meets the user's interests, and the combination of the two algorithms improves the accuracy and diversity of recommendations.

[0045] As described in step S4 above, based on the geographical location, climatic conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, a destination feature vector is constructed, and the ecological carrying capacity of various candidate destinations at different times is evaluated based on the pre-constructed ecological impact assessment model to obtain the best visiting period of the destination. For each candidate destination in the candidate destination recommendation list, a destination feature vector is constructed based on its geographical location (such as longitude and latitude, region, etc.), climatic conditions (such as four-season climate characteristics, annual precipitation, etc.) and resource endowment (such as natural landscape resources, cultural resources, etc.), which is a quantitative representation of the characteristics of the destination. Using the pre-constructed ecological impact assessment model, the ecological carrying capacity of various candidate destinations is evaluated considering different time factors. The ecological carrying capacity reflects the number of tourists and activity intensity that the destination can accommodate without destroying the ecological environment, thereby determining the best visiting period for each destination and providing support for sustainable tourism.

[0046] As described in step S5 above, the candidate destination recommendation list is matched with the optimal visit period, and the destinations with suitable carrying capacity within the user's intended travel period are screened and sorted to obtain the destination recommendation result. The generated candidate destination recommendation list is matched with the optimal visit period obtained through evaluation. The destinations with suitable ecological carrying capacity within the user's intended travel period are screened and sorted according to certain rules (such as the degree of matching with the user's preference, the comprehensive score of the destination, etc.), and finally the destination recommendation result is obtained to ensure that the recommended destination not only meets the user's time willingness, but also has a suitable ecological carrying capacity.

[0047] As described in the above step S6, according to the destination recommendation results, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, a reinforcement learning algorithm is used to automatically generate the optimal itinerary route, transportation, accommodation and activity arrangement to form a complete user travel plan. According to the destination recommendation results, combined with the user's time budget (total time available for travel), travel mode preference, accommodation preference and activity preference information, a reinforcement learning algorithm is used to automatically generate the optimal itinerary route, transportation, accommodation and activity arrangement to form a complete user travel plan. The reinforcement learning algorithm here adopts a multi-agent reinforcement learning framework, and different parts of the travel plan (such as itinerary planning, transportation selection, accommodation arrangement, activity planning, etc.) are respectively assigned to different agents for optimization. Through collaboration between the agents, the optimization of the overall travel plan is completed together to achieve the optimal travel plan arrangement that meets the various needs of users.

[0048] In one embodiment, user data is collected from multiple channels. For example, by cooperating with major online travel platforms, user travel history order data is obtained, including departure place, destination, travel time, accommodation information, consumption amount, etc.; from the user's browsing history on this platform, the user's clicks on different destinations and travel products, browsing time and other information are obtained; through the authorization of the social media platform, the user's travel-related pictures, texts, check-in information, etc. are obtained as social media activity data. A distributed crawler architecture is adopted to distribute the collection tasks to multiple nodes for simultaneous execution to improve data collection efficiency. For example, 10 crawler nodes are set, each node is responsible for collecting data from a specific source or a specific area, and about 100,000 user-related data can be collected every day. The collected raw data has problems such as inconsistent format, missing data, and duplication, and needs to be preprocessed. For data cleaning, duplicate records are removed by writing rules, and for missing values, the mean is filled or predicted based on other relevant data. For example, for missing values ​​of accommodation prices, fill them according to the average price of the same destination and the same type of accommodation. After preprocessing, the data is stored in the user behavior data warehouse, which is managed by a relational database (such as MySQL). When building the user behavior data warehouse, sensitive data (such as user ID number and contact information) is encrypted and stored using the Advanced Encryption Standard (AES) algorithm. In HDFS, strict access rights are set for each user's data file, and only authorized programs can access and decrypt the data.

[0049] Extract key features from the user behavior data warehouse. In this embodiment, take 1,000 user data as an example, and find out through data analysis:

[0050] Destination type preference: about 30% of users often go to coastal cities, 25% prefer historical and cultural cities, and 20% prefer mountain resorts, etc. Travel time and frequency: an average of 3 trips per year, of which 40% of trips are concentrated on holidays, and 30% are in summer vacation, etc. Length of stay: an average of 4 days at the destination, a slightly longer stay in coastal cities, about 5 days, and about 3 days in historical and cultural cities. Consumption level: 40% of users spend 5,000-10,000 yuan per person per trip, and 30% spend 10,000-20,000 yuan per person per trip, etc. Travel mode: 45% of users choose to travel by plane, and 35% choose high-speed rail, etc. Accommodation preference: 50% of users prefer three-star and above hotels, and 30% like homestays. Activity preference: 35% of users are interested in cultural activities (such as visiting museums), and 30% are keen on outdoor activities (such as hiking).

[0051] The key features extracted above are quantified and normalized. For numerical features (such as travel frequency and length of stay), the minimum-maximum normalization method is used to map them to the interval [0,1]. For proportional features (such as the preference ratio of destination type and the ratio of consumption level range), their original values ​​are maintained. Then all features are arranged in a certain order to form a user feature vector. For example, a user's feature vector may be represented as [preference ratio of coastal cities, preference ratio of historical and cultural cities, ..., ratio of air travel, ratio of train travel, ..., ratio of low consumption level, ratio of medium consumption level, ...], and the length is the sum of the number of all key features.

[0052] The traditional cosine similarity algorithm is used to calculate the cosine value of the angle between two vectors to measure their similarity. In this embodiment, it is improved by considering the importance weights of different features in the user feature vector. First, the weight of each key feature is determined by the analytic hierarchy process (AHP). For example, for young users, travel mode and activity preference may have higher weights; for middle-aged and elderly users, destination type and consumption level may have higher weights. Suppose the feature vectors of user A and user B are and The improved cosine similarity calculation formula is:

[0053]

[0054] where w i is the weight of the i-th feature, a i and b i are the i-th eigenvalues ​​of user A and user B, respectively. By calculating the improved cosine similarity between all users, we find a set of similar users whose similarity to the target user is higher than a set threshold (such as 0.7). For each target user, we can find about 800 similar users on average.

[0055] The structure combines convolutional neural network (CNN) and long short-term memory network (LSTM). The CNN layer is used to extract local features from user behavior data, such as capturing changes in user behavior patterns in the short term through convolution kernels of different sizes. The LSTM layer is used to process time series data, capture long-term dependencies of user behavior, and process spatiotemporal features in user behavior data. The user's historical behavior data (including travel history, browsing history, etc.) is divided into training set and test set in chronological order, with a ratio of 8:2. The data is preprocessed, text data is converted into word vector representation (such as using Word2Vec or GloVe), and numerical data is normalized. The Adam optimizer is used to train the model with the cross entropy loss function as the objective function. After multiple rounds of training, the accuracy of the model on the test set reached more than 85%. The feature vector of the target user is input into the trained deep learning model, and the model outputs the destination score that matches the user's historical behavior and preference characteristics. Combined with the recommendation results of similar users obtained by the collaborative filtering algorithm, a recommendation list of candidate destinations is comprehensively generated, with an average of 25 candidate destinations generated for each user.

[0056] The longitude and latitude of the destination are used as the basic geographic location features, and the distance between the destination and the main city center, as well as the geographic coordinate range of the region, are calculated to form a vector containing multiple geographic location related indicators. For example, for a certain destination, its geographic location feature vector can be expressed as [longitude, latitude, distance to the nearest large city, minimum longitude in the region, maximum longitude in the region, minimum latitude in the region, maximum latitude in the region]. Collect historical meteorological data of the destination, including indicators such as annual average temperature, annual precipitation, and monthly average sunshine duration. After normalizing these indicators, they form a climate condition feature vector. For example, the climate condition feature vector of a certain destination is [normalized value of annual average temperature, normalized value of annual precipitation, normalized value of monthly average sunshine duration]. Count the number of scenic spots, the number of world heritage sites, the types of special food, the number of cultural venues, etc. in the destination as quantitative indicators of resource endowment. After standardizing these indicators, they form a resource endowment feature vector. For example, the resource endowment feature vector of a destination is [standardized value of the number of scenic spots, standardized value of the number of world heritage sites, standardized value of the types of special food, and standardized value of the number of cultural venues]. The idea of ​​density peak clustering is introduced to cluster the above three types of feature vectors. First, the distance between each destination feature vector and all other vectors is calculated (using Euclidean distance) to obtain a distance matrix. Then, the local density of each vector (such as using a Gaussian kernel function to calculate the local density) and the minimum distance to the vector with a higher density are calculated based on the distance matrix. By drawing a decision diagram, the cluster center is automatically identified and divided into different categories such as seaside resort type, historical and cultural type, and natural scenery type to form a complete destination feature vector. In the recommendation process, based on the user's preference characteristics, priority is given to recommending from destinations of the corresponding category to improve the accuracy of the recommendation.

[0057] Based on the principle of system dynamics, an ecological impact assessment model is established. The model considers multiple factors, including the number of tourists, the intensity of tourist activities, the self-repair capacity of the ecosystem, and the fragility of the local ecological environment. The variables in the model are interconnected through causal relationships and feedback loops. For example, an increase in the number of tourists will lead to increased pressure on the ecosystem, which in turn affects the ecological carrying capacity, while the self-repair capacity of the ecosystem will alleviate the ecological pressure. The parameter values ​​in the model are determined by collecting historical ecological monitoring data, tourist activity data, and local ecological and environmental research reports. For example, based on the historical data of a mountainous area, it is determined that for every additional 1,000 tourists, the ecosystem pressure index increases by 0.1; the self-repair capacity coefficient of the ecosystem in the mountainous area is 0.05 (indicating the natural recovery ratio of the ecosystem pressure every day), etc. The predicted number of tourists and the intensity of tourist activities at different times (such as monthly and quarterly) are input into the model to simulate and calculate the ecological carrying capacity of the destination at different times. For example, after model calculation, the ecological carrying capacity of a coastal city is 8,000 people per day in the peak tourist season (July-August), and 3,000 people per day in the off-season (November-February of the following year), thus obtaining the best visiting time for the destination.

[0058] Match each destination in the candidate destination recommendation list with the user's intended travel period, and filter out destinations within that period. Then, compare the ecological carrying capacity of the filtered destinations with the user scale (assuming that the number of single users or family travelers is the user scale), and remove destinations whose ecological carrying capacity cannot meet the user scale. For the remaining destinations, sort them according to the degree of match between the destination and user preferences (such as the similarity between the destination type and the user's destination type preference, the matching degree between resource endowment and user activity preference, etc.), and the comprehensive score of the destination (considering factors such as the quality of attractions and word-of-mouth evaluation). The sorting algorithm uses a weighted sorting method. For example, the weight of the degree of match between the destination and user preferences is 0.6, and the weight of the comprehensive score is 0.4. The final score of each destination is calculated and sorted to obtain the destination recommendation results. On average, each user can get 6-10 recommended destinations.

[0059] A multi-agent reinforcement learning framework is adopted to define four agents: itinerary planning agent, transportation selection agent, accommodation arrangement agent and activity planning agent. The state of each agent consists of the user's preference information (such as time budget, travel mode preference, accommodation preference, activity preference), relevant information of the destination (such as scenic spot distribution, transportation hub location, accommodation facility distribution, etc.) and the partial completion of the current travel plan. For example, the state of the itinerary planning agent can be expressed as [the remaining days of the user's time budget, the user's preferred travel mode, the number and distribution of scenic spots at the destination, the planned number of days and scenic spots]. The action space of the itinerary planning agent includes selecting the next scenic spot to visit, determining the stay time at each scenic spot, etc.; the action space of the transportation selection agent includes selecting different modes of transportation (plane, train, bus, etc.) and specific flights; the action space of the accommodation arrangement agent includes selecting different types of accommodation (star hotels, homestays, etc.) and specific accommodation locations; the action space of the activity planning agent includes selecting different activities (cultural activities, outdoor activities, etc.) and determining the activity time. A reward function is designed for each agent to encourage the generation of the optimal travel plan. For example, the reward function of the itinerary planning agent can consider factors such as the compactness of the itinerary (avoiding time waste) and the relevance of attractions (matching with user interests); the reward function of the transportation selection agent can consider factors such as transportation costs and travel time; the reward function of the accommodation arrangement agent can consider factors such as accommodation costs and the distance to attractions and transportation hubs; the reward function of the event planning agent can consider factors such as the matching degree of the event with the user's interests and the feasibility of the event. The total reward function is the weighted sum of the reward functions of each agent, and the weights of the reward functions of each agent are determined through multiple experiments.

[0060] The multi-agent system is trained using the proximal policy optimization algorithm (PPO). During the training process, each agent selects an action based on the current state, observes the reward and new state of the environment feedback after executing the action, and continuously adjusts the strategy to maximize the cumulative reward. After a large number of training steps (such as 10,000 trainings), the agent learns to generate the optimal travel plan. The destination recommendation results and user preference information are input into the trained multi-agent reinforcement learning system, and each agent collaborates to generate the optimal itinerary route, transportation method, accommodation, and activity arrangements to form a complete user travel plan. For example, for a user who plans a 7-day trip and prefers air travel, four-star hotels, and cultural activities, the travel plan generated by the system may be: take a plane to the destination on the first day and stay in a four-star hotel; visit local museums and historical sites on the second day; take the high-speed rail to the surrounding historical and cultural cities on the third day, visit local cultural sites, etc., and plan the daily itinerary, transportation, and accommodation arrangements in detail.

[0061] In the process of generating user travel plans, the Neo4j graph database is used to build a destination knowledge graph. Information on attractions, food, accommodation, transportation, cultural activities, etc. of the destination is collected from various data sources (such as travel websites and encyclopedias). Each information entity is used as a node, and the relationship between them (such as the relationship between attractions and the city where they are located, the relationship between food and attractions, etc.) is used as an edge to build a knowledge graph. Relevant information about the destination is queried through the knowledge graph to provide users with more comprehensive and accurate travel information. For example, when recommending a certain attraction to a user, the food, accommodation and transportation information around the attraction can be queried at the same time to provide users with one-stop travel planning suggestions.

[0062] After generating the user's travel plan, the real-time data interface is continuously established with the hotel reservation system, scenic spot ticketing system, transportation operation department, etc. to obtain the real-time supply situation of the destination, such as the real-time hotel room status, the real-time passenger flow of the scenic spot, the real-time changes in transportation schedules, and other information. Based on the real-time data feedback and prediction model of big data, the dynamic programming algorithm is used to optimize the user's travel plan in real time. For example, when it is predicted that a scenic spot may be full of tourists on the date the user plans to go, the algorithm first evaluates the importance of the scenic spot to the entire travel plan. If the importance is low, other similar attractions nearby are directly recommended; if the importance is high, the algorithm tries to adjust the user's travel time or optimize the itinerary order to avoid peak hours. At the same time, for situations such as full hotel rooms or changes in transportation schedules, the algorithm will also adjust the accommodation arrangements or transportation methods accordingly to ensure the smooth implementation of the travel plan.

[0063] Reference Figure 2 , is a structural block diagram of a tourist destination recommendation system based on big data in one embodiment of the present invention, including:

[0064] The data acquisition unit is used to collect the original data of user travel history, browsing history and social media activities from multiple channels, pre-process the original data, and build a user behavior data warehouse;

[0065] A feature extraction unit, used to extract features from the data in the user behavior data warehouse to obtain key features reflecting the user's travel patterns and preferences;

[0066] A candidate destination analysis unit, configured to construct a user feature vector based on the extracted key features, and generate a candidate destination recommendation list by adopting a hybrid recommendation model combining a collaborative filtering algorithm and a content-based recommendation algorithm;

[0067] An ecological analysis unit, configured to construct a destination feature vector based on the geographical location, climate conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, evaluate the ecological carrying capacity of various candidate destinations at different times based on a pre-constructed ecological impact assessment model, and obtain the best visiting period for the destination;

[0068] A destination generation unit, used to match the candidate destination recommendation list with the optimal visit period, screen and sort the destinations with suitable carrying capacity within the user's intended travel period, and obtain a destination recommendation result;

[0069] The result output unit is used to automatically generate the optimal itinerary route, transportation method, accommodation and activity arrangement based on the destination recommendation results, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, using a reinforcement learning algorithm to form a complete user travel plan.

[0070] For the specific implementation of each unit in the above device example, please refer to the above method embodiment, which will not be repeated here.

[0071] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0072] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0073] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0074] In summary, the original data of user travel history, browsing history and social media activities are collected from multiple channels, the original data are preprocessed, and a user behavior data warehouse is constructed; the data in the user behavior data warehouse are feature extracted to obtain key features that reflect the user's travel mode and preference; a user feature vector is constructed based on the extracted key features, and a candidate destination recommendation list is generated by a hybrid recommendation model that combines a collaborative filtering algorithm with a content-based recommendation algorithm; a destination feature vector is constructed based on the geographical location, climatic conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, and the ecological carrying capacity of various candidate destinations at different times is evaluated based on a pre-constructed ecological impact assessment model to obtain the best visiting period for the destination; the candidate destination recommendation list is matched with the best visiting period, and destinations with suitable carrying capacity within the user's intended travel period are screened and sorted to obtain a destination recommendation result; according to the destination recommendation result, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, a reinforcement learning algorithm is used to automatically generate the optimal itinerary route, transportation mode, accommodation and activity arrangement to form a complete user travel plan, so as to achieve the purpose of accurately recommending tourist destinations, generating a complete travel plan, and taking into account ecological protection and real-time dynamic adjustment of the travel plan.

[0075] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0076] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0077] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A tourist destination recommendation method based on big data, characterized in that: The following steps are involved: Collecting raw data of user travel history, browsing history, and social media activities from multiple channels, preprocessing the raw data, and building a user behavior data warehouse; Extracting features from the data in the user behavior data warehouse to obtain key features reflecting the user's travel patterns and preferences, including but not limited to the types of destinations frequently visited by the user, travel time and frequency, length of stay, consumption level, travel mode, accommodation preference, and activity preference; Constructing a user feature vector based on the extracted key features, and generating a candidate destination recommendation list by adopting a hybrid recommendation model combining a collaborative filtering algorithm and a content-based recommendation algorithm; Based on the geographical location, climate conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, construct a destination feature vector, evaluate the ecological carrying capacity of various candidate destinations at different times based on a pre-constructed ecological impact assessment model, and obtain the best visiting period for the destination; Matching the candidate destination recommendation list with the optimal visit period, screening and sorting destinations with suitable carrying capacity within the user's intended travel period, and obtaining destination recommendation results; According to the destination recommendation results, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, a reinforcement learning algorithm is used to automatically generate the optimal itinerary route, transportation method, accommodation and activity arrangement to form a complete user travel plan. The reinforcement learning algorithm adopts a multi-agent reinforcement learning framework, and each agent is responsible for optimizing different parts of the itinerary, and the overall plan is carried out through collaboration between agents.

2. The tourist destination recommendation method based on big data according to claim 1, characterized in that: When collecting user data, a distributed crawler architecture is adopted, and in the process of building a user behavior data warehouse, data encryption technology is used to encrypt and store sensitive data.

3. The tourist destination recommendation method based on big data according to claim 1, characterized in that: The collaborative filtering algorithm calculates the similarity between users based on the improved cosine similarity algorithm. The content-based recommendation algorithm uses a deep learning model to deeply mine and match user historical behaviors and preference features. In the content-based recommendation algorithm, the deep learning model uses a structure that combines a convolutional neural network and a long short-term memory network to process the spatiotemporal features in user behavior data.

4. The tourist destination recommendation method based on big data according to claim 1, characterized in that: The destination feature vectors are classified by using an improved clustering algorithm, wherein the improved clustering algorithm introduces a density peak clustering idea for automatically identifying cluster centers.

5. The tourist destination recommendation method based on big data according to claim 1, characterized in that: When evaluating the ecological carrying capacity of candidate destinations, the impact of tourists' behavior on the ecological environment is comprehensively considered, and the ecological carrying capacity is evaluated by establishing an ecological impact assessment model.

6. The tourist destination recommendation method based on big data according to claim 1, characterized in that: In the process of generating user travel plans, knowledge graph technology is used to integrate and associate various types of information about the destination.

7. The tourist destination recommendation method based on big data according to claim 1, characterized in that: After the step of forming a complete user travel plan, the following steps are also included: Based on the actual supply situation of the real-time destination, the user's travel plan is dynamically adjusted through a real-time optimization algorithm. The real-time optimization algorithm is based on big data real-time data feedback and prediction models to warn and adjust possible resource shortages in advance.

8. A tourist destination recommendation system based on big data, characterized in that: include: The data acquisition unit is used to collect the original data of user travel history, browsing history and social media activities from multiple channels, pre-process the original data, and build a user behavior data warehouse; A feature extraction unit, used to extract features from the data in the user behavior data warehouse to obtain key features reflecting the user's travel patterns and preferences; A candidate destination analysis unit, configured to construct a user feature vector based on the extracted key features, and generate a candidate destination recommendation list by adopting a hybrid recommendation model combining a collaborative filtering algorithm and a content-based recommendation algorithm; An ecological analysis unit, configured to construct a destination feature vector based on the geographical location, climate conditions and resource endowment of the candidate destinations in the candidate destination recommendation list, evaluate the ecological carrying capacity of various candidate destinations at different times based on a pre-constructed ecological impact assessment model, and obtain the best visiting period for the destination; A destination generation unit, used to match the candidate destination recommendation list with the optimal visit period, screen and sort the destinations with suitable carrying capacity within the user's intended travel period, and obtain a destination recommendation result; The result output unit is used to automatically generate the optimal itinerary route, transportation method, accommodation and activity arrangement based on the destination recommendation results, combined with the user's time budget, travel mode preference, accommodation preference and activity preference, using a reinforcement learning algorithm to form a complete user travel plan.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the tourist destination recommendation method based on big data described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the tourist destination recommendation method based on big data described in any one of claims 1 to 7 are implemented.

Images