Festival and holiday travel-related place popularity analysis method based on multi-modal large model

Abstract

The invention relates to the technical field of data analysis, in particular to a holiday travel-related place popularity analysis method based on a multi-modal large model. According to the technical scheme provided by the invention, a multi-modal large model acquisition and analysis mode is adopted, so that information from various data sources, including multi-modal data such as characters, images and videos, can be effectively integrated. By training the model, the travel popularity of travel-related places in holidays and festivals can be predicted more accurately, and more accurate data support and decision reference are provided for related industries. A self-adaptive weight distribution mechanism is introduced, and the weights of different modal data are dynamically adjusted by using an attention mechanism, so that the algorithm can be flexibly adapted under different scenes and conditions, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method for analyzing the popularity of travel-related places during holidays based on a multimodal large model. Background Art

[0002] With the improvement of people's living standards and the increase in demand for leisure tourism, the scale of the tourism market continues to expand. However, during holidays, the demand for urban tourism has increased dramatically, resulting in an uneven distribution of tourists and increasingly prominent service quality. In order to meet the growing demand for the tourism market, the tourism industry urgently needs more effective tools to understand and predict tourism popularity. Among them, the popularization of the Internet and the development of social media have brought us rich multimodal data, including text, images, videos, etc. These multimodal data contain rich information and can be used to analyze tourism popularity. At the same time, the development of big data and artificial intelligence technology has made it possible to process and analyze massive multimodal data. Technologies such as deep learning have made significant progress in image recognition and natural language processing, providing us with powerful tools that can be applied to tourism heat analysis. By understanding and predicting the tourism popularity of holiday-related tourist attractions, tourism practitioners can make better resource allocation and service planning to meet the needs of tourists and improve the travel experience. For example, the opening hours of scenic spots can be adjusted, temporary staff can be added, and traffic routes can be optimized according to the prediction results, so as to achieve the goals of demand prediction and resource optimization. At the same time, an accurate tourism heat analysis system can also provide data support for decision-making in the tourism industry, helping companies to formulate more targeted marketing strategies, develop new tourism products and services, and maintain competitive advantages in the market. Therefore, this technical solution will greatly enhance the decision-making support capabilities of the tourism industry and help the industry continue to develop healthily. Summary of the invention

[0003] The algorithm for heat analysis is constructed by building a multimodal large model. The model can integrate information from multiple data sources, including text non-formatted data, image and video data, and effectively integrate and utilize it when predicting tourism heat. A comprehensive adjustment coefficient is introduced into the heat calculation, including traffic capacity, historical traffic flow and congestion coefficient, so that the prediction is more in line with the actual situation. Through these adjustments, the impact of complex factors in reality on tourism heat can be better reflected.

[0004] The present invention solves the deficiencies of the prior art and provides the following technical solutions: The heat analysis method of holiday travel-related places based on multimodal large model includes the following steps: 1) Data preparation: Collect and organize multiple data sources including transportation service bookings, amusement service bookings, Internet search volume, and weather data; 2) Feature extraction: Extract features from the data, such as transportation service bookings and weather data; 3) Build a multimodal large model: Build a multimodal large model, including multiple hidden layers and output layers, for feature prediction; 4) Training model: Use historical data to train the model and adjust model parameters to predict tourism popularity; 5) Predicting tourism popularity: Use the trained model to predict future tourism popularity.

[0005] The data collection types include formatted data, text data, video data and image data; the data sources include: mobile communication data, Internet open data, video AI analysis platform; mobile communication data includes group company signaling and provincial company signaling, among which group company signaling is pulled regularly through the interface to obtain formatted data such as roaming-in and roaming-out records, travel routes, and length of stay; provincial company signaling is pushed by the provincial company, providing detailed information such as the source of tourists, tourist portraits, and tourist transportation; Internet open data is pulled regularly through the API, providing transportation service reservations, amusement service reservations, real-time weather, and Internet search volume data; the video AI analysis platform pushes video and image data of the entrances and exits and core areas of the cooperative scenic spots in real time, and the pictures are uploaded to MinIO storage, and the acquisition module regularly pulls these pictures for multimodal fusion analysis; feature extraction preprocessing cleans and preprocesses non-formatted text, digital data, images, and video data; formatted data is de-duplicated and de-empty according to fixed time periods and designated area codes to eliminate duplicate data from affecting the processing results; data formats and units are standardized.

[0006] The multimodal fusion and large model construction, feature alignment maps the features of different modes to the same dimensional space; time alignment aligns the data of different modes according to timestamps; spatial alignment spatially aligns the multimodal data of different areas of the scenic area, maps the multimodal data to the specific location of the scenic area, uses geographic information system GIS technology to match the data with the map, and associates the data with the area; latitude alignment is that text data is a high-dimensional word vector representation, and image video data is a low-dimensional feature vector extracted by a convolutional neural network. The high-dimensional data is reduced in dimension, and the high-dimensional text features of the text are reduced in dimension using principal component analysis PCA or t-SNE to match the latitude of image and video features; feature fusion uses splicing, weighted averaging, and attention mechanism to fuse multimodal features; the attention mechanism fusion uses the BERT deep learning model, the image uses the ResNet deep learning model, and the video uses the 3D-CNN deep learning model to obtain their respective prediction results, and the attention mechanism is used to weight the features of different modes.

[0007] Traditional tourism popularity prediction methods often only consider some factors, and the prediction algorithm cannot fully utilize the information from multiple data sources, resulting in inaccurate prediction results. The technical solution proposed in the present invention adopts a multimodal large model collection and analysis method, which can effectively integrate information from multiple data sources, including text, images, videos and other multimodal data. By training this model, the tourism popularity of holiday-related tourist attractions can be more accurately predicted, providing more accurate data support and decision-making references for related industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a thermal analysis flow chart of the present invention. DETAILED DESCRIPTION

[0009] The heat analysis method of holiday travel-related places based on multimodal large model includes the following steps: 1) Data preparation: Collect and organize multiple data sources including transportation service bookings, amusement service bookings, Internet search volume, and weather data; 2) Feature extraction: Extract features from the data, such as transportation service bookings and weather data; 3) Build a multimodal large model: Build a multimodal large model, including multiple hidden layers and output layers, for feature prediction; 4) Training model: Use historical data to train the model and adjust model parameters to predict tourism popularity; 5) Predicting tourism popularity: Use the trained model to predict future tourism popularity.

[0010] The data collection types include formatted data, text data, video data and image data; the data sources include: mobile communication data, Internet open data, video AI analysis platform; mobile communication data includes group company signaling and provincial company signaling, among which group company signaling is pulled regularly through the interface to obtain formatted data such as roaming-in and roaming-out records, travel routes, and length of stay; provincial company signaling is pushed by the provincial company, providing detailed information such as the source of tourists, tourist portraits, and tourist transportation; Internet open data is pulled regularly through the API, providing transportation service reservations, amusement service reservations, real-time weather, and Internet search volume data; the video AI analysis platform pushes video and image data of the entrances and exits and core areas of the cooperative scenic spots in real time, and the pictures are uploaded to MinIO storage, and the acquisition module regularly pulls these pictures for multimodal fusion analysis; feature extraction preprocessing cleans and preprocesses non-formatted text, digital data, images, and video data; formatted data is de-duplicated and de-empty according to fixed time periods and designated area codes to eliminate duplicate data from affecting the processing results; data formats and units are standardized.

[0011] The multimodal fusion and large model construction, feature alignment maps the features of different modes to the same dimensional space; time alignment aligns the data of different modes according to timestamps; spatial alignment spatially aligns the multimodal data of different areas of the scenic area, maps the multimodal data to the specific location of the scenic area, uses geographic information system GIS technology to match the data with the map, and associates the data with the area; latitude alignment is that text data is a high-dimensional word vector representation, and image video data is a low-dimensional feature vector extracted by a convolutional neural network. The high-dimensional data is reduced in dimension, and the high-dimensional text features of the text are reduced in dimension using principal component analysis PCA or t-SNE to match the latitude of image and video features; feature fusion uses splicing, weighted averaging, and attention mechanism to fuse multimodal features; the attention mechanism fusion uses the BERT deep learning model, the image uses the ResNet deep learning model, and the video uses the 3D-CNN deep learning model to obtain their respective prediction results, and the attention mechanism is used to weight the features of different modes.

[0012] The parameters used to predict tourism popularity are as follows: In this formula, Indicates tourism popularity. is the feature vector containing all influencing factors, and are the weight and bias parameters of the model, respectively. It is a deep neural network model with multiple hidden layers, which is used to perform nonlinear mapping on input features to predict tourism popularity. is the error term of the model, i.e., the adjustment coefficient; The neural network model is expressed as: in, is the number of layers of the neural network, The activation function is , and It is The weights and bias parameters of the layer; Use historical data to train the neural network model, and then use the trained model to predict the tourism popularity of new holiday travel destinations.

[0013] The specific process of the embodiment is as follows: S1: Data Collection The data collection module covers multiple types of data, including formatted data, text data, video data, and image data. Mobile communication data: including group company signaling and provincial company signaling. Group company signaling is pulled regularly through the interface to obtain formatted data such as roaming records, travel routes, and length of stay. Provincial company signaling is pushed by the provincial company, providing detailed information such as the source of tourists, tourist portraits, and tourist transportation. Internet open data: pulled regularly through the API, providing data such as transportation service reservations, amusement service reservations, real-time weather, and Internet search volume. Video AI analysis platform: real-time streaming push of video and image data of the entrances and exits and core areas of cooperative scenic spots, pictures are uploaded to MinIO storage, and the collection module regularly pulls these pictures for multimodal fusion analysis.

[0014] Through the integration of multiple data sources, the data collection module can provide comprehensive tourism popularity analysis and lay a solid data foundation for subsequent multimodal fusion and intelligent analysis.

[0015] S2: Data preprocessing The preprocessing module mainly cleans and preprocesses unformatted text, digital data, images, and video data.

[0016] Format data: remove duplicates and blanks according to a fixed time period and specified area code to eliminate the impact of duplicate data on processing results.

[0017] Data cleaning (delete duplicate and erroneous data) and data standardization (unify data formats and units). Use the big data processing framework, Apache Spark, to clean and standardize data.

[0018] Image Scaling: Scale the images to a uniform size (224x224 pixels). Use an image processing library (such as OpenCV, Pillow) to scale the images to 224x224 pixels to ensure the consistency of the input image size for subsequent processing and analysis steps.

[0019] Image enhancement: Perform data enhancement, rotation, flipping, cropping, color adjustment, etc. to increase data diversity. Use the image processing library torchvision to perform a variety of enhancement operations to improve the generalization ability of the model.

[0020] Image normalization: Normalize pixel values ​​to the range [0, 1] or [-1, 1]. Use the image processing library to normalize image pixel values ​​to ensure the consistency of data input.

[0021] Feature extraction: Use ResNet50 to extract image features. Use PyTorch or similar frameworks to load the pre-trained ResNet50 model and extract the feature vector.

[0022] Image preprocessing (resizing, color standardization), feature extraction (using ResNet model). Use the open source image processing library, OpenCV, for image preprocessing, and use the deep learning framework to load the pre-trained model for feature extraction.

[0023] Video frame extraction: Extract frames from the video at fixed intervals (such as 5 frames per second). Use the OpenCV library to read the video and extract frames at specified intervals. Video frame preprocessing: Perform image preprocessing (such as image scaling, normalization, etc.) on each frame. Use Pillow and torchvision libraries to process each frame. Feature extraction: Use LSTM network to extract video features. First use a pre-trained convolutional neural network (such as ResNet50) to extract features for each frame, and then input these features into the LSTM network for sequence feature extraction. Use the open source video processing library, OpenCV, for frame extraction, and use the deep learning framework PyTorch to load the pre-trained model 3D-CNN for feature extraction.

[0024] S3: Multimodal fusion and large model construction Feature alignment: Map features of different modalities to the same dimensional space to ensure that they can be fused. For example, use a neural network to convert feature vectors of different modalities (such as text, images, and audio) to the same dimensional space.

[0025] Temporal alignment: Align data of different modalities according to timestamps, for example, extract video frames, grab image data and related text comments, and ensure that these data are in the same time window according to their timestamps. Spatial alignment: For different areas of the scenic spot (such as entrance, main scenic spot, exit), we need to spatially align the multimodal data of these areas in order to distinguish the tourist behavior in a specific area. We can map the multimodal data to a specific location in the scenic spot, use geographic information system (GIS) technology to match the data with the map, and associate the data with the area. Latitude alignment: Text data is a high-dimensional word vector representation, while image and video data is a low-dimensional feature vector extracted by convolutional neural network, so we need to align the dimension and reduce the high-dimensional data. For example, use principal component analysis (PCA) or t-SNE to reduce the high-dimensional text features of the text to match the latitude of image and video features. Feature fusion: Use different methods (such as splicing, weighted average, attention mechanism, etc.) to fuse multimodal features. Attention mechanism fusion: After each modality data has been processed independently (BERT deep learning model is used for text, ResNet deep learning model is used for image, and 3D-CNN deep learning model is used for video), the respective prediction results are obtained. The attention mechanism is used to weight the features of different modalities and dynamically adjust the contribution of each modality feature during fusion. For example, weights are assigned to the visual features of images, the emotional features of text, and the dynamic features of videos. After the prediction value weights are processed, the final comprehensive prediction value can be obtained. Model training: Use the fused features to train the model for classification or regression tasks. Prediction and evaluation: Use test data for prediction and calculate the evaluation indicators of the model, such as accuracy, precision, recall, F1-score, etc. Map text and image features to a 128-dimensional vector space, fuse and concatenate text and image features, and finally train the fused features for model training and binary classification tasks.

[0026] After analysis format the data based on Table 1: Table 1 Formatted data S4: Data Analysis After the collected data is cleaned, deduplicated, and processed for missing values, the data enters the heat analysis module. The core algorithms involved here are as follows: In this formula, Indicates tourism popularity. is the feature vector containing all influencing factors, and are the weight and bias parameters of the model, respectively. It is a deep neural network model with multiple hidden layers, which is used to perform nonlinear mapping on input features to predict tourism popularity. is the error term of the model, i.e., the adjustment coefficient. Specifically, the neural network model can be expressed as: in, is the number of layers of the neural network, The activation function is , and It is The weights and bias parameters of the layer.

[0027] Use historical data to train this neural network model, and then use the trained model to predict the tourism popularity of new holiday travel destinations.

[0028] First, substitute the data and model parameters into the neural network model: Hidden Layer: Now let's bring real data and model parameters into the calculation.

[0029] Assume the model parameters are as follows: The weight matrix of the first hidden layer and the bias vector (W and b values ​​here are model training simulation values) The weight matrix of the second hidden layer and the bias vector : The bias vector of the output layer : Assume the known data example: Train ticket bookings Airline ticket bookings Hotel reservations Scenic spot ticket reservations Internet search volume temperature Air Quality Index .

[0030] The specific calculation logic is as follows: Compute linear combinations: application Activation function, calculate the linear combination part : The linear combination part of the first hidden layer is obtained: The linear combination is added to the bias vector and applied Activation function: application Activation function: Finish calculate: Will After summing the values, we get value, For related reservation heat, historical heat, and search heat, For adjustment factors, such as road traffic capacity , current traffic , historical traffic , historical traffic weight , current traffic weight The adjustment factor is 0.18.

[0031] Calculated output: , the Y calculated here is the current tourism heat of the tourism-related places. Through the above calculation, the final heat score can be obtained to provide data support for tourism management and planning. A comprehensive adjustment coefficient is introduced in the heat calculation, including traffic capacity, historical traffic flow and congestion coefficient, so that the prediction is more in line with the actual situation. Through these adjustments, the impact of complex factors in reality on tourism heat can be better reflected. Technical relevance proof: The introduction of the comprehensive adjustment coefficient is similar to the weight adjustment in the multi-factor regression model, which uses the combination of historical data and real-time data to make more accurate predictions. This method is widely used in economics and operations research, but it is the first time to be used in tourism heat prediction.

[0032] The technical effects of the present invention are as follows: (1) Fusion of multiple data sources: Traditional tourism heat analysis methods often rely on a single data source, such as historical passenger flow data or Internet search volume. This invention innovatively combines multiple data sources (formatted data, text data, image and video data) together to improve the comprehensiveness and accuracy of heat analysis by integrating information from different data sources.

[0033] (2) Adaptive weight allocation and comprehensive adjustment coefficient: This invention introduces an adaptive weight allocation mechanism. By using the attention mechanism, the weights of different modal data are dynamically adjusted, so that the algorithm can flexibly adapt to different scenarios and conditions, thereby improving the accuracy of prediction. In the heat calculation, comprehensive adjustment coefficients such as traffic capacity, historical traffic flow and congestion coefficient are introduced to make the prediction results more in line with the actual situation. This innovation takes into account the impact of complex factors in reality on tourism heat, making the prediction results more practical and guiding.

[0034] The beneficial effects of the present invention are as follows: (1) Market forecasting and planning: by analyzing the tourism popularity during holidays and non-holidays, it helps tourism operators predict the changing trend of market demand, improve the utilization efficiency of tourism resources, optimize the allocation of human and material resources, reduce the waste of resources in the off-season, and improve the service level in the peak season. Through the multimodal data analysis system, it is possible to identify the changes in tourist preferences and quantity during different seasons and holidays, and predict market demand in advance. For example, the system predicts that the number of tourists in a certain scenic spot will increase significantly during the National Day this year. The operator can increase staff in advance, arrange more tour guides and temporary facilities, and ensure the service quality during the peak period of tourists. (2) Product development and optimization: based on tourism popularity data, identify tourist interests and preferences, develop new tourism products, and optimize existing product lines to increase the market adaptability and competitiveness of tourism products, meet the needs of different types of tourists, and improve tourist satisfaction and repurchase rate. By analyzing tourists' Internet search behavior and booking data, it is found that many tourists show a strong interest in a certain historical site. Based on this, a new in-depth cultural tour product can be developed, including historical explanations and interactive experiences. At the same time, the system reported that the popularity of existing city sightseeing products has declined, so the operator optimized them by adding night sightseeing sessions and new dining experiences. (3) Precision marketing: By analyzing the popularity of different travel-related places, we formulate personalized marketing strategies, carry out precise advertising and event planning, improve the return on marketing investment, attract target customer groups, and enhance brand influence and market share. Using popularity data, travel companies have found that the popularity of short-distance city tours has increased recently, especially surrounding rural vacations. Therefore, the company can place targeted advertisements on social media for young families and self-driving enthusiasts, and launch limited-time discount activities, which greatly improves the marketing effect.

[0035] Effects of the invention: (1) Help tourism operators predict market demand: formulate more reasonable planning and resource allocation strategies. Through accurate popularity forecasting, resources can be reasonably arranged during holidays to avoid resource waste or insufficient supply.

[0036] (2) Develop tourism products that meet market demand based on popularity data: Optimize existing products. According to the popularity data of different regions and attractions, targeted tourism products and services can be launched to enhance market competitiveness.

[0037] (3) Carry out precision marketing for tourist-related places with different popularity: Through popularity analysis, popular attractions and areas can be identified, and targeted marketing strategies can be formulated to attract more tourists.

[0038] (4) Destination image building: By using popularity data, we can identify the attractions and activities preferred by tourists, carry out targeted publicity and promotion, and enhance the popularity and attractiveness of the destination.

[0039] (5) Avoid overdevelopment: Through heat prediction, the number of tourists can be effectively controlled, excessive damage to natural landscapes and cultural heritage can be avoided, and sustainable development can be achieved.

[0040] Possible application scenarios in the future: (1) Smart Tourism City: By predicting and analyzing the popularity of various attractions in the city, it helps city managers optimize transportation routes, arrange public resources reasonably, and improve the tourist experience. For example, if it is predicted that a certain attraction will have a large number of tourists during a holiday, additional public transportation services and security personnel can be arranged in advance.

[0041] (2) Scenic spot operation management: Help scenic spot operators predict tourist flow and optimize resource allocation and service facilities within the scenic spot. For example, according to the popularity prediction results, reasonably arrange catering, rest areas, amusement facilities and other services within the scenic spot to avoid queues and congestion caused by too many tourists.

[0042] (3) Tourism marketing strategy: Use the results of popularity analysis to develop precise marketing strategies to attract more tourists. For example, launch preferential activities and promotion plans for popular attractions and activities, and conduct precision marketing through social media and online platforms.

[0043] (4) Personalized services for tourists: Provide personalized travel recommendations and services. For example, based on tourists’ search records, booking history, and preference data, combined with travel popularity analysis, recommend the most suitable travel destinations, attractions, and activities.

[0044] Glossary Deep learning: Deep learning is a branch of artificial intelligence that learns complex feature representations of data by building deep neural network models. In this system, deep learning models can be used to extract features and learn representations of multimodal data, such as using CNN to process image data and using RNN and Transformer to process text data.

[0045] NLP: Natural Language Processing. (Natural language processing is a branch of artificial intelligence that involves the technology of processing and understanding human language. In this system, NLP technology can be used to perform word segmentation, sentiment analysis, topic modeling, etc. on text data in order to better understand user comments and opinions) Multimodal fusion: Multimodal fusion refers to the technology of integrating multiple types of data from different data sources for joint analysis and processing. In this system, it is necessary to study and develop multimodal fusion methods to integrate multiple types of data such as text, images, and videos, and use multimodal information to improve the accuracy and effectiveness of tourism heat analysis.

[0046] Model training and optimization: Model training and optimization refers to the process of adjusting the parameters of the machine learning model using training data to improve the performance and generalization ability of the model. In this system, it is necessary to design appropriate training strategies and optimization algorithms to train and optimize the constructed multimodal large model to improve its ability to predict tourism popularity.

[0047] Data mining and analysis: Data mining and analysis is the process of discovering hidden patterns and knowledge from large-scale data, involving the use of statistics, machine learning and other methods to analyze and mine data. In this system, the collected tourism-related data needs to be mined and analyzed to discover the laws and trends of tourism popularity.

Claims

1. A heat analysis method for holiday travel sites based on a multimodal large model, characterized by The steps include: 1) Data preparation: Collect and organize multiple data sources including transportation service bookings, amusement service bookings, Internet search volume, and weather data; 2) Feature extraction: Extract features from the data, such as transportation service bookings and weather data; 3) Build a multimodal large model: Build a multimodal large model, including multiple hidden layers and output layers, for feature prediction; 4) Training model: Use historical data to train the model and adjust model parameters to predict tourism popularity; 5) Predicting tourism popularity: Use the trained model to predict future tourism popularity.

2. The method for analyzing the popularity of holiday travel-related places based on a multimodal large model according to claim is characterized in that The data collection types include formatted data, text data, video data and image data; the data sources include: mobile communication data, Internet open data, video AI analysis platform; mobile communication data includes group company signaling and provincial company signaling, among which group company signaling is pulled regularly through the interface to obtain formatted data such as roaming-in and roaming-out records, travel routes, and length of stay; provincial company signaling is pushed by the provincial company, providing detailed information such as the source of tourists, tourist portraits, and tourist transportation; Internet open data is pulled regularly through the API, providing transportation service reservations, amusement service reservations, real-time weather, and Internet search volume data; the video AI analysis platform pushes video and image data of the entrances and exits and core areas of the cooperative scenic spots in real time, and the pictures are uploaded to MinIO storage, and the acquisition module regularly pulls these pictures for multimodal fusion analysis; feature extraction preprocessing cleans and preprocesses non-formatted text, digital data, images, and video data; formatted data is de-duplicated and de-empty according to fixed time periods and designated area codes to eliminate duplicate data from affecting the processing results; data formats and units are standardized.

3. The method for analyzing the popularity of holiday travel-related places based on a multimodal large model according to claim is characterized in that The multimodal fusion and large model construction, feature alignment maps the features of different modalities to the same dimensional space; time alignment aligns the data of different modalities according to timestamps; space Alignment will spatially align the multimodal data of different areas of the scenic area, map the multimodal data to specific locations in the scenic area, use geographic information system (GIS) technology to match the data with the map, and associate the data with the area; Latitude alignment means that text data is represented by high-dimensional word vectors, while image and video data is low-dimensional feature vectors extracted by convolutional neural networks. The high-dimensional data is processed by reducing the dimensionality, and the high-dimensional text features of the text are reduced by principal component analysis PCA or t-SNE to match the dimensionality of the image and video features. Feature fusion uses splicing, weighted averaging, and attention mechanism to fuse multimodal features. The attention mechanism fusion uses the BERT deep learning model, the image uses the ResNet deep learning model, and the video uses the 3D-CNN deep learning model to obtain their respective prediction results, and uses the attention mechanism to weight the features of different modalities.

4. The method for analyzing the popularity of holiday travel-related places based on a multimodal large model according to claim is characterized in that The parameters used to predict tourism popularity are as follows: In this formula, Y represents tourism popularity, is a feature vector containing all influencing factors, W and b are the weight and bias parameters of the model respectively, It is a deep neural network model with multiple hidden layers, which is used to perform nonlinear mapping on input features to predict tourism popularity. is the error term of the model, i.e., the adjustment coefficient; The neural network model is expressed as: Where L is the number of layers in the neural network, The activation function is , and It is The weights and bias parameters of the layer; Use historical data to train the neural network model, and then use the trained model to predict the tourism popularity of new holiday travel destinations.

Images