System for generating publicity directions for tourism-preferred populations based on learning from tourism big data

Abstract

The present invention discloses a system for generating publicity directions for tourism preference groups based on learning of tourism big data, which relates to the technical field of cultural and tourism publicity, and includes: acquiring and integrating unstructured tourism big data from the Internet for multi-modal analysis and extraction to generate cross-modal tourism dynamic interest representations of the Internet; obtaining de-identified tourism digital footprints of the population according to a public platform, and dividing the population according to the radiation range of the tourism industry in the area to be publicized to obtain target populations and non-target populations; analyzing and screening the tourism preferences of the target populations and non-target populations in combination with the tourism industry in the area to be publicized to generate publicity directions for the tourism preferences of the target populations and non-target populations; and updating the publicity directions for the tourism preferences of the target populations and non-target populations by using the cross-modal tourism dynamic interest representations of the Internet to generate optimal publicity directions for the corresponding tourism preference groups. The advantages of the present invention are as follows: realizing targeted user mining and improving the accuracy and coverage of tourism marketing.

Description

Technical Field

[0001] The present invention relates to the technical field of cultural and tourism publicity and promotion, and specifically relates to a system for generating publicity directions for tourism preference groups based on tourism big data learning. Background Art

[0002] Generating publicity directions for tourism preference groups through tourism big data learning refers to analyzing the historical behaviors, interest changes of target and non-target groups, as well as unstructured data (such as social media comments, pictures, videos, etc.), to update in real time and accurately identify the tourism preferences of different user groups, so as to formulate personalized and timely tourism publicity strategies to improve marketing effects and user engagement.

[0003] The mainstream of existing tourism publicity and promotion methods is to establish corresponding population user portraits based on the historical data of users, and perform tourism industry matching according to the population user portraits to achieve publicity and push. Although targeted push can be achieved, due to the serious homogenization of the tourism industry, users will choose nearby similar tourism industries for play, resulting in a poor actual conversion rate of tourism publicity and push for constructing population user portraits. Summary of the Invention

[0004] To solve the above technical problems, a system for generating publicity directions for tourism preference groups based on tourism big data learning is provided. This technical solution solves the problem that the mainstream of the existing tourism publicity and promotion methods is to establish corresponding population user portraits based on the historical data of users, perform tourism industry matching according to the population user portraits to achieve publicity and push. Although targeted push can be achieved, due to the serious homogenization of the tourism industry, users will choose nearby similar tourism industries for play, resulting in a poor actual conversion rate of tourism publicity and push for constructing population user portraits.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A system for generating publicity directions for tourism preference groups based on tourism big data learning, including:

[0007] A tourism trend module, a population division module, a population preference module, and a publicity update module;

[0008] The tourism trend module is used to obtain and extract the integrated unstructured tourism big data on the Internet for multimodal analysis, and generate an Internet cross-modal tourism dynamic interest representation;

[0009] The population division module is used to obtain the desensitized tourism digital footprints of the population according to the public platform, and divide the population according to the radiation range of the tourism industry in the area to be publicized to obtain the target population and the non-target population;

[0010] The population preference module is electrically connected to the population division module. The population preference module is used to analyze the tourism preferences of the target population and the non-target population and fit and screen them with the tourism industry in the area to be promoted, and generate the tourism preference promotion directions of the target population and the non-target population.

[0011] The promotion update module is electrically connected to the tourism trend module and the population preference module. The promotion update module is used to update the tourism preference promotion directions of the target population and the non-target population by using the cross-modal tourism dynamic interest representation on the Internet, and generate the optimal promotion directions for the corresponding tourism preference populations.

[0012] Furthermore, the tourism trend module specifically includes:

[0013] The data division unit integrates unstructured tourism big data based on the Internet, divides it according to each dimension of the data, and obtains unstructured tourism multi-dimensional type data; each dimension of the data includes: text data, video data, and spatio-temporal data.

[0014] The data dimension conversion unit establishes corresponding data encoders according to each dimension of the data, performs vector conversion on the unstructured tourism multi-dimensional type data, and obtains unstructured tourism multi-dimensional type vector data.

[0015] The data unification unit uses multivariate analysis - canonical correlation analysis method to construct a ternary loss function, calculates the maximum similarity between the unstructured tourism multi-dimensional type vector data for cross-modal projection, and obtains unstructured tourism multi-dimensional type unified vector data, and the method is as follows:

[0016] ;

[0017] Among them, is to construct a ternary loss function, is the feature extraction function of the text data of, is the video data of the feature extraction function, is the spatio-temporal data of the feature extraction function, is the similarity metric function, is the weight of the similarity between the text feature and the video feature, is the weight of the similarity between the video feature and the spatio-temporal feature, is the weight of the similarity between the spatio-temporal feature and the text feature;

[0018] The timeliness preference unit performs normalization processing based on unstructured tourism multi-dimensional type unified vector data, trains a Transformer deep learning network, uses video feature segments as query vectors, uses spatio-temporal features as key-value variables, and performs local fusion according to the cross-attention mechanism to generate video-spatio-temporal local fusion features; uses text feature segments as query variables, uses video-spatio-temporal local fusion features as keys, and uses original video features as values, and performs global fusion according to the multi-attention mechanism to generate text-video-spatio-temporal global fusion features. Combining the time-aware attention mechanism, taking the Internet-integrated unstructured tourism big data as input and the Internet cross-modal tourism dynamic interest representation as output, an Internet tourism dynamic timeliness preference model is obtained.

[0019] Further, the tourism trend module specifically includes:

[0020] The attraction estimation unit counts the basic factors that affect the crowd's choice in the desensitized tourism digital footprint of the crowd, and estimates the corresponding comprehensive attraction index for the tourism industry in the area to be promoted; the basic factors that affect the crowd's choice include: the number of scenic spots, tourism ratings, and facility integrity.

[0021] The time distance unit obtains the distance parameters between the tourism industry in the area to be promoted and each known city in the surrounding area, and determines the time distance between each associated city and the tourism industry in the area to be promoted.

[0022] The consumption probability evaluation unit establishes a Huff tourism industry spatial radiation range model based on the comprehensive attraction index of the tourism industry in the area to be promoted and the time distance between the tourism industry in the area to be promoted and each city, and calculates the consumption probability of the crowd choosing the tourism industry in the area to be promoted.

[0023] Further, the crowd division module specifically includes:

[0024] The preference crowd division unit divides the tourism crowd according to the consumption records using K-means clustering based on the historical consumption records in the desensitized tourism digital footprint of the crowd, and obtains a set of tourism crowd consumption preference divisions.

[0025] The consumption demand unit determines the consumption demand of the tourism industry in the area to be promoted based on the economic status of the city where the tourism industry in the area to be promoted is located.

[0026] The consumption potential unit establishes a Huff tourism industry economic radiation range model based on the consumption demand of the tourism industry in the area to be promoted, the consumption probability of the crowd choosing the tourism industry in the area to be promoted, and the set of tourism crowd consumption preference divisions, and calculates the consumption potential index of the crowd choosing the tourism industry in the area to be promoted.

[0027] Further, the population division module specifically includes:

[0028] An initial publicity conversion unit that determines the publicity conversion rate of the tourism industry in the historical areas to be publicized.

[0029] A publicity conversion weight unit that uses the analytic hierarchy process to assign weights to the consumption potential index, consumption probability, and publicity conversion rate of the tourism industry in the areas to be publicized for the population selection.

[0030] A publicity priority unit that calculates the publicity priority score for the corresponding population based on the consumption potential index, consumption probability, and publicity conversion rate of the tourism industry in the areas to be publicized for the population selection and the weights of the consumption potential index, consumption probability, and publicity conversion rate of the tourism industry in the areas to be publicized for the population selection.

[0031] A population division unit that divides the population according to the publicity priority score of the population to obtain the target population and the non-target population.

[0032] Further, the population preference module specifically includes:

[0033] A preference feature unit that marks the tourism preference feature parameters of the target population and the non-target population based on the de-identified tourism digital footprints of the target population and the non-target population.

[0034] A population preference feature unit that performs vector conversion on the tourism preference feature parameters of the target population and the non-target population according to linear mapping to obtain the tourism preference feature vectors of the target population and the non-target population.

[0035] A tourism resource structure unit that obtains the tourism industry in the areas to be publicized, extracts the tourism resources included in the tourism industry, and establishes the structured data of the tourism resources of the tourism industry in the areas to be publicized.

[0036] A keyword screening unit that extracts the keywords corresponding to the tourism resources of the tourism industry in the areas to be publicized based on the structured data of the tourism resources of the tourism industry in the areas to be publicized using the TF-IDF algorithm.

[0037] Further, the population preference module specifically includes:

[0038] A keyword matrix unit that establishes a keyword matrix of the tourism resources of the tourism industry in the areas to be publicized based on the keywords corresponding to the tourism resources of the tourism industry in the areas to be publicized.

[0039] A tourism resource vector unit that uses the Top-N algorithm to screen out the keywords in the keyword matrix corresponding to the tourism resources of the tourism industry in the areas to be publicized as label data, and calculates the tourism resource label vector of the tourism industry in the areas to be publicized with the TF-IDF normalized value of the keyword as the weight.

[0040] Furthermore, the population preference module specifically includes:

[0041] A population publicity weight unit that normalizes the publicity priority scores of the target population and the non-target population, denoted as the publicity weights of the target population and the non-target population;

[0042] A personalized population tourism resource vector unit that uses the publicity weights of the target population and the non-target population as influencing factors to calculate the Euclidean distance between the tourism preference feature vectors of the target population and the non-target population and the tourism resource label vector of the tourism industry in the area to be publicized, and determines the tourism resource label vectors matching the target population and the non-target population;

[0043] A personalized population publicity vector unit that, based on the tourism resource label vectors matching the target population and the non-target population, establishes a play strategy publicity database for the target population and a play culture publicity database for the non-target population, and generates the tourism preference publicity directions of the target population and the non-target population.

[0044] Furthermore, the publicity update module specifically includes:

[0045] A personalized population publicity unstructured unit that determines the tourism publicity unstructured parameters of the tourism preference publicity directions of the target population and the non-target population based on the tourism preference publicity directions of the target population and the non-target population;

[0046] A publicity plan generation unit that substitutes the tourism publicity unstructured parameters of the tourism preference publicity directions of the target population and the non-target population into the Internet tourism dynamic timeliness preference model, and updates the tourism publicity unstructured parameters of the tourism preference publicity directions of the target population and the non-target population according to the Internet cross-modal tourism dynamic interest representation, and generates the optimal publicity directions for the corresponding tourism preference populations.

[0047] The present invention proposes a method for generating a publicity plan for tourism preference populations based on tourism big data learning. Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] By integrating Internet unstructured tourism big data for multi-modal analysis, constructing a dynamic interest representation model, accurately dividing the target population and the non-target population, and generating a differentiated publicity strategy in combination with the characteristics of the tourism industry in the area to be publicized. This solution can capture the changes in user interests in real time, push highly matched personalized strategies for the target population to improve the conversion rate, and at the same time stimulate the potential interests of the non-target population through cultural content, achieve targeted user mining, and improve the accuracy and coverage of tourism marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flowchart of a method for generating a publicity direction for tourism preference populations based on tourism big data learning;

[0050] Figure 2 It is a system framework diagram for generating publicity directions for tourism preference groups based on learning from tourism big data; Specific implementation manners

[0051] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0052] Refer to Figure 1 As shown, the present invention provides a method for generating publicity directions for tourism preference groups based on learning from tourism big data, including:

[0053] Step 1: Obtain unstructured tourism big data integrated from the Internet for multi-modal analysis and extraction, and generate Internet cross-modal tourism dynamic interest representations;

[0054] The above Step 1 includes the following contents:

[0055] Step 101: Based on the unstructured tourism big data integrated from the Internet, divide it according to each dimension of the data to obtain unstructured tourism multi-dimensional type data; each dimension of the data includes: text data, video data, and spatio-temporal data;

[0056] Establish corresponding data encoders according to each dimension of the data, and perform vector conversion on the unstructured tourism multi-dimensional type data to obtain unstructured tourism multi-dimensional type vector data;

[0057] Step 102: Use multivariate analysis - canonical correlation analysis method to construct a triple loss function, calculate the maximum similarity between unstructured tourism multi-dimensional type vector data for cross-modal projection, and obtain unstructured tourism multi-dimensional type unified vector data, in the following way:

[0058] ;

[0059] Among them, is the triple loss function, is the feature extraction function for text data of, is the feature extraction function for video data of, is the feature extraction function for spatio-temporal data of, is the similarity metric function, is the weight of the similarity between text features and video features, is the weight of the similarity between video features and spatio-temporal features, is the weight of the similarity between spatio-temporal features and text features;

[0060] Based on the unstructured tourism multi-dimensional type unified vector data for normalization processing, train the Transformer deep learning network. Use the video feature segment as the query vector, the spatio-temporal feature as the key-value variable, and perform local fusion according to the cross-attention mechanism to generate the video-spatio-temporal local fusion feature; use the text feature segment as the query variable, the video-spatio-temporal local fusion feature as the key, and the original video feature as the value, and perform global fusion according to the multi-attention mechanism to generate the text-video-spatio-temporal global fusion feature. Combine the time-aware attention mechanism, take the Internet integrated unstructured tourism big data as the input, and the Internet cross-modal tourism dynamic interest representation as the output to obtain the Internet tourism dynamic timeliness preference model;

[0061] When in use, combine the content in steps 101 to 102:

[0062] As a further content, for the feature methods of data in each dimension, by using natural language processing, convolutional neural network and spatio-temporal convolutional neural network, establish a data encoder corresponding to the Internet integrated unstructured tourism big data, extract the corresponding dimension features and input them into the cross-modal mapping network, and determine the similarity difference between the data of each dimension feature according to the Euclidean distance between the dimension features, map the features of each modality to a common space, and realize the data alignment between modalities to improve the effect of multi-modal learning of the subsequent model; and because tourism preferences are extremely susceptible to external environment and network trends, therefore, in the process of training the model, it is necessary to consider the season and month and geographical location factors, and in order to ensure the accuracy of the prediction results, in the process of feature annotation of the data, focus on considering the coarse-grained feature parameters that affect tourism preferences to avoid inaccurate output results due to excessive branches of the data.

[0063] Step 2: According to the public platform, obtain the de-identified tourism digital footprints of the population, and divide the population according to the tourism industry radiation range of the area to be publicized to obtain the target population and the non-target population;

[0064] The said step 2 includes the following content:

[0065] Step 201: Statistically analyze the basic factors that affect the population's decision-making in the de-identified tourism digital footprints of the population, and estimate the corresponding comprehensive attraction index for the tourism industry in the area to be publicized; the basic factors that affect the population's decision-making include: the number of scenic spots, tourism ratings, and facility integrity;

[0066] Obtain the distance parameters between the tourism industry in the area to be publicized and each known city in the surrounding area, and determine the time distance between each associated city and the tourism industry in the area to be publicized;

[0067] Based on the comprehensive attractiveness index of the tourism industry in the area to be promoted and the time distance between the tourism industry in the area to be promoted and each city, establish a Huff tourism industry spatial radiation range model to calculate the consumption probability of the crowd choosing the tourism industry in the area to be promoted. The method is as follows:

[0068] ;

[0069] Among them, is the consumption probability of the i-th tourism crowd choosing the tourism industry in the j-th area to be promoted, is the comprehensive attractiveness index of the tourism industry in the j-th area to be promoted, is the time distance between the i-th tourism crowd in the associated city and the tourism industry in the j-th area to be promoted, is the regression coefficient of comprehensive attractiveness, is the regression coefficient of time distance;

[0070] Step 202: Based on the historical consumption records in the de-identified tourism digital footprints of the crowd, use K-means clustering to divide the tourism crowd according to the consumption records to obtain a set of divided tourism crowd consumption preferences;

[0071] Based on the economic status of the cities where the tourism industry in the area to be promoted is located, determine the consumption demand of the tourism industry in the area to be promoted;

[0072] Based on the consumption demand of the tourism industry in the area to be promoted, the consumption probability of the crowd choosing the tourism industry in the area to be promoted, and the set of divided tourism crowd consumption preferences, establish a Huff tourism industry economic radiation range model to calculate the consumption potential index of the crowd choosing the tourism industry in the area to be promoted. The method is as follows:

[0073] ;

[0074] Among them, is the consumption potential index of the i-th tourism crowd choosing the j-th area to be promoted, is the consumption demand of the tourism industry in the j-th area to be promoted, is the consumption ability of the i-th tourism crowd in the set of divided tourism crowd consumption preferences;

[0075] Step 203: Determine the promotion conversion rate of the tourism industry in the historical area to be promoted;

[0076] Use the analytic hierarchy process to assign weights to the consumption potential index, consumption probability, and promotion conversion rate of the crowd choosing the tourism industry in the area to be promoted;

[0077] According to the consumption potential index, consumption probability and promotion conversion rate of the tourism industry in the area to be promoted by the group of people, and the consumption potential index weight, consumption probability weight and promotion conversion rate weight of the tourism industry in the area to be promoted by the group of people, the promotion priority score of the corresponding group of people is calculated as follows:

[0078] ;

[0079] in, Give priority score to the publicity of the i-th tourist group. is the promotion conversion rate of the i-th tourist group, , , They are consumption potential index weight, consumption probability weight and promotion conversion rate weight.

[0080] Divide the population into target and non-target populations according to their publicity and promotion priority scores;

[0081] When using, combine the contents in steps 201 to 103:

[0082] As a further content, in order to ensure the timeliness of the publicity and promotion priority score, the crowd-anonymous tourism digital footprint needs to be updated according to the season to ensure the accuracy of the crowd segmentation;

[0083] The mainstream method of existing tourism promotion and marketing is to establish corresponding user portraits based on users' historical data, match tourism industries according to the user portraits, and implement promotion and marketing. Although targeted marketing can be achieved, due to the serious homogeneity of the tourism industry, users will choose similar tourism industries nearby to visit, resulting in a poor actual conversion rate of tourism promotion and marketing based on user portraits.

[0084] Step 3: Analyze the tourism preferences of the target and non-target groups and the tourism industry of the area to be promoted, and generate the tourism preference promotion direction of the target and non-target groups;

[0085] The step three includes the following contents:

[0086] Step 301: based on the desensitized tourism digital footprints of the target population and the non-target population, marking the tourism preference characteristic parameters of the target population and the non-target population;

[0087] According to linear mapping, the tourism preference characteristic parameters of the target group and the non-target group are transformed into vectors to obtain the tourism preference characteristic vectors of the target group and the non-target group;

[0088] Acquire the tourism industry of the area to be promoted, extract the tourism resources contained in the tourism industry, and establish the tourism resource structured data of the tourism industry in the area to be promoted;

[0089] Based on the structured data of tourism resources of the tourism industry in the area to be promoted, use the TF-IDF algorithm to extract the keywords corresponding to the tourism resources of the tourism industry in the area to be promoted;

[0090] Step 302: Based on the keywords corresponding to the tourism resources of the tourism industry in the area to be promoted, establish a keyword matrix of the tourism resources of the tourism industry in the area to be promoted;

[0091] Use the Top-N algorithm to screen out the keywords in the keyword matrix corresponding to the tourism resources of the tourism industry in the area to be promoted as label data, and use the TF-IDF normalized value of the keywords as weights to calculate the tourism resource label vector of the tourism industry in the area to be promoted. The method is as follows:

[0092] ;

[0093] Among them, is the tourism resource label vector of the j-th tourism industry in the area to be promoted, is the local importance of the word in the tourism resource document , is the number of tourism resource documents containing the word , is the keyword matrix of the tourism resources of the tourism industry in the area to be promoted;

[0094] Step 303: Perform normalization processing on the publicity priority scores of the target population and the non-target population, and record them as the publicity weights of the target population and the non-target population;

[0095] Use the publicity weights of the target population and the non-target population as influencing factors to calculate the Euclidean distance between the tourism preference feature vectors of the target population and the non-target population and the tourism resource label vector of the tourism industry in the area to be promoted, and determine the tourism resource label vectors matched by the target population and the non-target population;

[0096] Based on the tourism resource label vectors matched by the target population and the non-target population, establish a publicity database for the travel guides of the target population and a publicity database for the tourism culture of the non-target population, and generate the publicity directions of the tourism preferences of the target population and the non-target population;

[0097] When in use, combine the content in Steps 301 to 303:

[0098] As a further content, by dividing the population into target population and non-target population, since the attention preferences of the target population match the tourism resources of the tourism industry in the area to be promoted, therefore, corresponding travel guides are generated based on the tourism resources of the tourism industry in the area to be promoted as the promotion direction. While the attention preferences of the non-target population do not match the tourism resources of the tourism industry in the area to be promoted, so the play culture corresponding to the tourism resources of the tourism industry in the area to be promoted is promoted to increase the attention of the non-target population.

[0099] Step 4: Use the Internet cross-modal tourism dynamic interest representation to update the tourism preference promotion directions for the target population and the non-target population, and generate the optimal promotion directions for the corresponding tourism preference populations;

[0100] The above Step 4 includes the following contents:

[0101] Step 401: Based on the tourism preference promotion directions of the target population and the non-target population, determine the tourism promotion unstructured parameters of the tourism preference promotion directions of the target population and the non-target population;

[0102] Step 402: According to the tourism promotion unstructured parameters of the tourism preference promotion directions of the target population and the non-target population, substitute them into the Internet tourism dynamic timeliness preference model, and update the tourism promotion unstructured parameters of the tourism preference promotion directions of the target population and the non-target population according to the Internet cross-modal tourism dynamic interest representation, and generate the optimal promotion directions for the corresponding tourism preference populations.

[0103] Combining the contents in 401 to 402:

[0104] As a further content, since the Internet cross-modal tourism dynamic interest representation usually changes dynamically with the season and the network popularity, and the interest directions of the corresponding populations' tourism also change dynamically, therefore, using the Internet cross-modal tourism dynamic interest representation to update the tourism promotion unstructured parameters of the tourism preference promotion directions of the target population and the non-target population can effectively improve the effect of tourism marketing and the participation degree of users.

[0105] Refer to Figure 2 As shown, the promotion direction system for tourism preference populations generated based on tourism big data learning includes:

[0106] Tourism trend module, population division module, population preference module, promotion update module;

[0107] The tourism trend module is used to obtain the Internet integrated unstructured tourism big data for multi-modal analysis and extraction, and generate the Internet cross-modal tourism dynamic interest representation;

[0108] The population division module is used to obtain the desensitized digital footprints of tourism for the population based on the public platform, and divide the population according to the radiation range of the tourism industry in the area to be publicized, so as to obtain the target population and the non-target population;

[0109] The population preference module is electrically connected to the population division module. The population preference module is used to analyze the tourism preferences of the target population and the non-target population and fit and screen them with the tourism industry in the area to be publicized, so as to generate the publicity directions of the tourism preferences of the target population and the non-target population;

[0110] The publicity update module is electrically connected to the tourism trend module and the population preference module. The publicity update module is used to update the publicity directions of the tourism preferences of the target population and the non-target population by using the cross-modal tourism dynamic interest representation on the Internet, so as to generate the optimal publicity directions for the population with corresponding tourism preferences.

[0111] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A publicity direction system for learning and generating tourism preference groups based on tourism big data, characterized in that Including: A tourism trend module, a population segmentation module, a population preference module, and a publicity update module; The tourism trend module is used to obtain and extract unstructured tourism big data integrated from the Internet through multimodal analysis, and generate an Internet cross-modal tourism dynamic interest representation; The population segmentation module is used to obtain the desensitized tourism digital footprints of the population based on a public platform, and segment the population according to the radiation range of the tourism industry in the area to be publicized, obtaining the target population and the non-target population; The population preference module is electrically connected to the population segmentation module. The population preference module is used to analyze the tourism preferences of the target population and the non-target population and fit and screen them with the tourism industry in the area to be publicized, generating the publicity directions of the tourism preferences of the target population and the non-target population; The publicity update module is electrically connected to the tourism trend module and the population preference module. The publicity update module is used to update the publicity directions of the tourism preferences of the target population and the non-target population by using the Internet cross-modal tourism dynamic interest representation, generating the optimal publicity directions for the corresponding tourism preference populations; Among them, the publicity direction of the tourism preferences of the target population is to generate corresponding play strategies from the tourism resources of the tourism industry in the area to be publicized as the publicity direction, while the publicity direction of the tourism preferences of the non-target population is the play culture corresponding to the tourism resources of the tourism industry in the area to be publicized as the publicity direction; Among them, the tourism trend module includes: A data division unit, which divides the unstructured tourism big data integrated from the Internet according to each dimension of the data, obtaining unstructured tourism multi-dimensional type data; each dimension of the data includes: text data, video data, and spatio-temporal data; A data dimension conversion unit, which establishes corresponding data encoders according to each dimension of the data, and performs vector conversion on the unstructured tourism multi-dimensional type data, obtaining unstructured tourism multi-dimensional type vector data; A data unification unit, which uses multivariate analysis - canonical correlation analysis method to construct a ternary loss function, calculates the maximum similarity between the unstructured tourism multi-dimensional type vector data for cross-modal projection, obtaining unstructured tourism multi-dimensional type unified vector data, and the method is as follows: ； Among them, is to construct a triple loss function, is the feature extraction function for text data , is the feature extraction function for video data , is the feature extraction function for spatio-temporal data , is the similarity measurement function, is the weight of the similarity between text features and video features, is the weight of the similarity between video features and spatio-temporal features, is the weight of the similarity between spatio-temporal features and text features; A timeliness preference unit, which performs normalization processing based on the unstructured tourism multi-dimensional type unified vector data, trains a Transformer deep learning network, uses video feature segments as query vectors, uses spatio-temporal features as key-value variables, and performs local fusion according to the cross-attention mechanism to generate video-spatio-temporal local fusion features; uses text feature segments as query variables, uses video-spatio-temporal local fusion features as keys, and uses the original video features as values, and performs global fusion according to the multi-attention mechanism to generate text-video-spatio-temporal global fusion features. Combining the time-aware attention mechanism, taking the Internet integrated unstructured tourism big data as the input and the Internet cross-modal tourism dynamic interest representation as the output, an Internet tourism dynamic timeliness preference model is obtained.

2. The publicity direction system for generating tourist preference groups based on tourism big data according to claim 1, characterized in that, The tourism trend module also includes: An attraction estimation unit that counts the basic factors affecting the choices of the crowd in the desensitized tourism digital footprints of the crowd, and estimates the corresponding comprehensive attraction index for the tourism industry in the area to be promoted; the basic factors affecting the choices of the crowd include: the number of scenic spots, tourism ratings, and facility integrity; A time distance unit that obtains the distance parameters between the tourism industry in the area to be promoted and each known city in the surrounding area, and determines the time distance between each associated city and the tourism industry in the area to be promoted; A consumption probability evaluation unit that establishes a Huff tourism industry spatial radiation range model based on the comprehensive attraction index of the tourism industry in the area to be promoted and the time distance between the tourism industry in the area to be promoted and each city, and calculates the consumption probability of the crowd choosing the tourism industry in the area to be promoted.

3. The publicity direction system for generating tourism preference groups based on tourism big data learning according to claim 2, wherein, The crowd division module includes: A preference crowd division unit that divides the tourism crowd based on the historical consumption records in the desensitized tourism digital footprints of the crowd, and uses K-means clustering to obtain a set of divisions of the consumption preferences of the tourism crowd according to the consumption records; A consumption demand unit that determines the consumption demand of the tourism industry in the area to be promoted based on the economic status of the city where the tourism industry in the area to be promoted is located; A consumption potential unit that establishes a Huff tourism industry economic radiation range model based on the consumption demand of the tourism industry in the area to be promoted, the consumption probability of the crowd choosing the tourism industry in the area to be promoted, and the set of divisions of the consumption preferences of the tourism crowd, and calculates the consumption potential index of the crowd choosing the tourism industry in the area to be promoted.

4. The publicity direction system for generating tourism preference groups based on tourism big data according to claim 3, characterized in that The crowd division module further includes: An initial promotion conversion unit that determines the promotion conversion rate of the tourism industry in the historical area to be promoted; A promotion conversion weight unit that assigns weights to the consumption potential index, consumption probability, and promotion conversion rate of the crowd choosing the tourism industry in the area to be promoted using the analytic hierarchy process; A promotion priority unit that calculates the promotion priority score for the corresponding crowd based on the consumption potential index, consumption probability, and promotion conversion rate of the crowd choosing the tourism industry in the area to be promoted and the weights of the consumption potential index, consumption probability, and promotion conversion rate of the crowd choosing the tourism industry in the area to be promoted; A crowd division unit that divides the crowd according to the promotion priority score of the crowd to obtain the target crowd and the non-target crowd.

5. The publicity direction system for generating tourism preference groups based on tourism big data according to claim 4, wherein The crowd preference module includes: A preference feature unit that marks the tourism preference feature parameters of the target crowd and the non-target crowd based on the desensitized tourism digital footprints of the target crowd and the non-target crowd; A crowd preference feature unit that performs vector conversion on the tourism preference feature parameters of the target crowd and the non-target crowd according to linear mapping to obtain the tourism preference feature vectors of the target crowd and the non-target crowd; A tourism resource structure unit that obtains the tourism industry in the area to be promoted, extracts the tourism resources included in the tourism industry, and establishes the structured data of the tourism resources of the tourism industry in the area to be promoted; A keyword screening unit that extracts the keywords corresponding to the tourism resources of the tourism industry in the area to be promoted using the TF-IDF algorithm based on the structured data of the tourism resources of the tourism industry in the area to be promoted.

6. The publicity direction system for generating travel preference groups based on travel big data according to claim 5, characterized in that, The crowd preference module further includes: Keyword matrix unit, based on keywords corresponding to tourism resources of the tourism industry in the area to be promoted, establish a keyword matrix of tourism resources of the tourism industry in the area to be promoted; Tourism resource vector unit, use the Top-N algorithm to screen out keywords in the keyword matrix corresponding to tourism resources of the tourism industry in the area to be promoted as label data, use the TF-IDF normalized value of the keywords as weights, and calculate the tourism resource label vector of the tourism industry in the area to be promoted.

7. The publicity direction system for generating tourism preference groups based on tourism big data learning according to claim 6, wherein The population preference module further includes: Population promotion weight unit, perform normalization processing on the promotion priority scores of the target population and the non-target population, denoted as the promotion weights of the target population and the non-target population; Personalized population tourism resource vector unit, use the promotion weights of the target population and the non-target population as influencing factors, calculate the Euclidean distance between the tourism preference feature vectors of the target population and the non-target population and the tourism resource label vector of the tourism industry in the area to be promoted, and determine the tourism resource label vectors matched by the target population and the non-target population; Personalized population promotion vector unit, based on the tourism resource label vectors matched by the target population and the non-target population, establish a promotion database for the travel guides of the target population and a promotion database for the tourism culture of the non-target population, and generate the tourism preference promotion directions of the target population and the non-target population.

Images