Parent-child tourist attraction recommendation system based on AIGC

Abstract

The invention discloses an AIGC-based parent-child tourism area recommendation system, and belongs to the technical field of tourism recommendation systems, the AIGC-based parent-child tourism area recommendation system comprises a data acquisition module, an intelligent classification module, an expert component module and a decision push module, and the expert component module comprises a user analysis module, a subject education module, a quality science popularization module and a traffic environment module; the data acquisition module is used for acquiring scenic spot data, user data and user attribute data; the intelligent classification module performs theme mining and classification on the data in the data acquisition module according to the theme demand of the expert component module; the expert component module obtains an overall scenic spot sequence according to the scenic spot rough sorting results of the subject education module and the quality science popularization module, and selects the first M scenic spots as rough sorting results; the traffic environment module performs weighted calculation on the first M scenic spots selected by the expert component module to obtain a sorting result of the recommended M scenic spots; and the decision pushing module is used for determining an optimal scenic spot recommendation sequence.

Description

Technical Field

[0001] The present invention belongs to the field of tourism recommendation, and in particular to an AIGC-based parent-child travel scenic spot recommendation system. Background Art

[0002] In recent years, families with children have increasingly higher requirements for travel experiences. They not only hope to enjoy parent-child time during the trip, but also hope that their children can gain knowledge and improve their quality during the trip. The traditional scenic spot recommendation system has many limitations.

[0003] When it comes to data collection, most systems rely on a single or limited number of data sources, making it difficult to fully capture multi-dimensional user information. For example, analyzing preferences solely based on user search history on travel platforms fails to capture the potential interest reflected in discussions and sharing about family travel on social media. It also struggles to integrate behavioral data such as real-time user location and travel methods captured by operator data platforms and IoT devices. This results in inaccurate user profiles and recommendations that deviate significantly from actual user needs.

[0004] There's a lack of efficient and intelligent data classification mechanisms for data categorization and processing. A common practice is to simply classify data based on scenic spot type or basic user attributes, failing to deeply explore the underlying thematic attributes of the data. For example, with massive amounts of textual data related to scenic spots, it's difficult to accurately distinguish which data is relevant to the concrete learning of subject knowledge in compulsory education and which is related to the quality-oriented education practices of cultural, museum, and science and technology venues. This makes it impossible for subsequent recommendation modules to conduct targeted, in-depth analysis and utilization of the data.

[0005] Traditional systems are even more inadequate when it comes to meeting the diverse needs of parents and children. They struggle to effectively integrate and meet the comprehensive demands of parent-child travel, including knowledge acquisition, quality development, and travel security. For example, when recommending scenic spots, they often only consider their popularity, overlooking their specific characteristics in terms of subject knowledge education, quality development, and transportation comfort. Regarding travel security, they are unable to accurately predict traffic conditions based on travel time, providing reliable travel recommendations for families. Summary of the Invention

[0006] Purpose of the invention: To provide a parent-child travel scenic spot recommendation system based on AIGC to solve the above-mentioned problems existing in the prior art.

[0007] Technical solution: An AIGC-based parent-child travel scenic spot recommendation system, including a data collection module, an intelligent classification module, an expert component module, and a decision-making push module. The expert component module includes a user analysis module, a subject education module, a quality science popularization module, and a traffic environment module.

[0008] The data acquisition module is used to obtain scenic area data and user data, and combine the user input data to generate data for analyzing user attributes. User attributes include static attributes and dynamic attributes;

[0009] The intelligent classification module conducts topic mining and classification on the data in the data collection module according to the theme requirements of the expert component module, and transmits user attribute theme data to the user analysis module, humanities and history theme data to the subject education module, cultural and museum science theme data to the quality science popularization module, and traffic environment theme data to the traffic environment module;

[0010] The user analysis module is used to calculate the recommendation index, the subject education module is used to roughly sort the scenic spots in the humanities and history theme data, and the quality science popularization module is used to roughly sort the scenic spots in the cultural, museum and technology theme data;

[0011] The expert component module obtains the overall scenic spot ranking based on the rough ranking results of the subject education module and the quality science popularization module, and selects the top M scenic spots as the rough ranking results;

[0012] The traffic environment module performs weighted calculation on the top M scenic spots selected by the expert component module to obtain the ranking results of the recommended M scenic spots;

[0013] The decision-making push module performs spatial clustering judgment on the ranking results of the recommended M scenic spots (such as calculating the Moran index or based on K-means), and determines the optimal scenic spot recommendation ranking based on the user's travel time and the number of scenic spots visited.

[0014] In one embodiment, scenic spot data is crawled from websites such as official cultural and tourism media, social media UGC content, and OTA platforms. The scenic spot data is related to scenic spot characteristics, tourist characteristics, emotional tendencies (such as user likes and comments), and interest preferences (such as user collections).

[0015] In one embodiment, scenic spot data includes text content and multimodal content (such as pictures, videos, audio, and text); for text content: 1) NLP-based data preprocessing removes irrelevant characters and stop words, and completes text cleaning and standardization operations; 2) Chinese word segmentation is performed based on Jieba, and the text is divided into words to form a vocabulary; 3) Part-of-speech tagging is completed based on the Orange visual text analysis tool; 4) Named entity recognition (NER) is implemented based on a deep learning model; for multimodal content: in-depth analysis is performed based on AIGC large models (such as the Gemini model and the CLIP model), for example, analyzing tourist characteristics and scenic spot characteristics from videos and pictures.

[0016] In one embodiment, user data is crawled on platforms such as operator data platforms, Internet of Things devices (such as passenger flow sensing sensors) and UnionPay consumption data service platforms. The user data is related to the user's location, travel trajectory, travel mode, behavioral preferences, etc. For example, the data crawled on the operator data platform can obtain the user's location and travel trajectory based on LBS.

[0017] In one embodiment, static attributes include age, education level, occupation, family structure, source of tourists, consumption capacity (such as income level, travel budget), etc.; dynamic attributes include departure place, travel trajectory, travel mode (such as self-driving, public transportation), travel preferences (such as independent travel, group travel), length of stay, consumption preferences, cultural interests, activity preferences, dining preferences, seasonal preferences, digital service preferences, technology preferences, scenic spot preferences, etc.

[0018] In one embodiment, the intelligent classification module performs topic mining and classification on the data acquired by the data collection module according to the topic requirements of the expert component module, including the following steps:

[0019] The data acquired by the data collection module is analyzed using a topic model algorithm (such as LDA) to generate a document-topic matrix (DT matrix). The document-topic matrix is ​​a two-dimensional matrix U of Q×V, where Q represents the number of documents. A document is a data set consisting of multiple words, and V represents the number of topics. Each row corresponds to an analyzed document data, and each column represents a topic. Each element u in the matrix q,v Indicates the relevance of the qth document to the vth topic (usually a probability value);

[0020] When the relevance exceeds the set threshold, the qth document data in the matrix is ​​considered to belong to the vth topic, and the topic types are user attribute topics, humanities and history topics, cultural and scientific topics, and traffic environment topics. For example, if the threshold is set to 0.8, that is, when the value of an element in the two-dimensional matrix is ​​≥0.8, it means that the document belongs to a certain topic;

[0021] The document data belonging to each topic is transmitted as a whole to the topic module corresponding to the expert component module.

[0022] In one embodiment, the steps for the user analysis module to calculate the recommendation index are as follows:

[0023] The dynamic feature vectors of dynamic attributes are obtained by using the feature extraction algorithm in NLP, and the static feature vectors of static attributes are obtained by using demographic feature encoding;

[0024] The user attribute feature vector p is obtained by combining the dynamic feature vector and the static feature vector;

[0025] The scenic spot feature vector f1 or f2 extracted from the subject education module or quality science popularization module is used to calculate the recommendation index α using the cosine similarity algorithm i , the calculation formula is as follows:

[0026]

[0027] Among them, i=1,2, and this value is determined based on the characteristic values ​​of subject education and quality science popularization preference analysis.

[0028] In one embodiment, the dimensions of the scenic spot feature vector and the user attribute feature vector are consistent. If there is no relevant information data in a certain component dimension of the scenic spot feature vector, the component value in this dimension is 0. When the non-zero data feature vector data contains positive or negative emotions, the difference between the vector feature values ​​is large, and the recommendation index cannot be directly calculated. In this case, the "maximum-minimum" normalization method is used to process the original data into the [0,1] interval. The calculation formula is as follows:

[0029]

[0030] Among them, p′ j Indicates that p j To achieve standardized results, p j is the component of p, max(p) and min(p) represent the maximum and minimum values ​​of the original data.

[0031] In one embodiment, the steps for the subject education module to roughly sort the scenic spots in the humanities and history theme data are as follows:

[0032] Establish a POI database for subject education scenic spots:

[0033] Through the API interface, the location of scenic spot addresses in the humanities and history theme data is retrieved to achieve geographic analysis and complete the establishment of the subject education scenic spot POI database;

[0034] Extract the subject education scenic spot feature vector f1:

[0035] The feature extraction algorithm in NLP is used to extract the feature vector of the scenic spot in the POI database of subject education scenic spots, and the feature vector is preliminarily standardized (UnitLength standardization) to obtain the feature vector f1 of the subject education scenic spot;

[0036] Dynamic NPS score of subject education scenic spots:

[0037] Based on NLTK, sentiment analysis is performed on the scenic spots in the POI database of subject education scenic spots, and the number of positive sentiment evaluations indicating recommendation (N1), the number of negative sentiment evaluations indicating non-recommendation (N2), and the total number of evaluation samples (N) are obtained. The dynamic NPS (Net Recommendation Score) score β of the subject education scenic spots is calculated.NPS , the calculation formula is as follows:

[0038]

[0039] Rough ranking of subject education scenic spots:

[0040] By calculating the recommendation coefficient γ1, the scenic spots in the subject education scenic spot POI database are roughly sorted. The calculation formula is as follows:

[0041] γ1=α1·β NPS

[0042] In one embodiment, the steps for the quality science popularization module to roughly sort the scenic spots in the cultural, museum, and technology theme data are as follows:

[0043] Establish a POI database for quality science popularization scenic spots:

[0044] Through the API interface, the address name of the scenic spot in the cultural and scientific theme data is retrieved to achieve geographic analysis and complete the establishment of the quality science scenic spot POI database;

[0045] Extract the characteristic vector f2 of the quality science popularization scenic spot:

[0046] The feature extraction algorithm in NLP is used to extract the feature vector of the scenic spot in the POI database of quality science popularization scenic spot, and the feature vector is preliminarily normalized (Unit Length normalization) to obtain the feature vector f2 of the quality science popularization scenic spot;

[0047] Dynamic NPS score of quality science popularization scenic spots:

[0048] Based on NLTK, sentiment analysis is performed on the scenic spots in the POI database of quality science popularization scenic spots, and the number of positive sentiment evaluations indicating recommendation (N1), the number of negative sentiment evaluations indicating non-recommendation (N2), and the total number of evaluation samples (N) are obtained. The dynamic NPS (net recommendation value) score β of quality science popularization scenic spots is calculated. NPS , the calculation formula is as follows:

[0049]

[0050] Rough ranking of quality science popularization scenic spots:

[0051] By calculating the recommendation coefficient γ2, the scenic spots in the POI database of quality science popularization scenic spots are roughly sorted. The calculation formula is as follows:

[0052] γ2=α2·β NPS

[0053] In one embodiment, the traffic environment module performs weighted calculation on the first M scenic spots selected by the expert component module as follows:

[0054] Combined with the user's expected travel time, the traffic environment data is predicted by deep learning neural network, and the traffic environment comfort index ε is weightedly calculated for the γ values ​​of the first M scenic spots selected by the expert component module. The expression is: γ·ε;

[0055] The calculation formula of traffic environment comfort index ε is as follows:

[0056]

[0057] Among them, ω k represents the weight of the comfort index (determined by entropy weight method, AHP method, etc.), s k represents the eigenvalue of the comfort index, K represents the total number of eigenvalues, and the comfort index includes traffic comfort, climate comfort, air comfort, volume comfort, dining comfort, accommodation comfort, etc.

[0058] According to the weighted calculation results, the top M scenic spots selected by the expert component module are re-ranked.

[0059] In summary, the beneficial effects of the present invention are:

[0060] 1. The present invention obtains scenic spot data and user data through the data acquisition module, and combines the user input data to jointly generate data for analyzing user attributes, which can build a more three-dimensional user portrait and make the subsequent scenic spot recommendation results more in line with the user's actual needs.

[0061] 2. The user analysis module, subject education module, and quality science popularization module of the present invention collaborate through data interaction and algorithm to form a closed loop of "user preference identification → educational needs matching → rough ranking of scenic spots", ensuring that the top M scenic spots selected by the expert component not only meet the actual needs of users but also take into account the reputation and quality of the scenic spots.

[0062] 3. The traffic module of the present invention calculates the recommendation coefficient values ​​of the top M scenic spots selected by the expert component based on the weighted traffic environment comfort index, and re-ranks the top M scenic spots selected by the expert component based on the weighted calculation results. This solves the problem of traditional recommendation systems ignoring travel guarantees and improves the actual experience of parent-child travel.

[0063] 4. The decision-making push module of the present invention performs spatial clustering judgment on the ranking results of the recommended M scenic spots, and at the same time limits the user's travel time and the number of scenic spots visited, so as to determine the optimal scenic spot recommendation ranking, which can ensure the rationality of the recommendation plan in terms of geographical distribution and time arrangement. DETAILED DESCRIPTION

[0064] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art are not described to avoid confusion with the present invention.

[0065] Example 1

[0066] The AIGC-based parent-child travel scenic spot recommendation system disclosed in this embodiment includes a data collection module, an intelligent classification module, an expert component module, and a decision push module. The expert component module includes a user analysis module, a subject education module, a quality science popularization module, and a traffic environment module.

[0067] The data acquisition module is used to obtain scenic area data and user data, and combine the user input data to generate data for analyzing user attributes. User attributes include static attributes and dynamic attributes;

[0068] The intelligent classification module conducts topic mining and classification on the data in the data collection module according to the theme requirements of the expert component module, and transmits user attribute theme data to the user analysis module, humanities and history theme data to the subject education module, cultural and museum science theme data to the quality science popularization module, and traffic environment theme data to the traffic environment module;

[0069] The user analysis module is used to calculate the recommendation index, the subject education module is used to roughly sort the scenic spots in the humanities and history theme data, and the quality science popularization module is used to roughly sort the scenic spots in the cultural, museum and technology theme data;

[0070] The expert component module obtains the overall scenic spot ranking based on the rough ranking results of the subject education module and the quality science popularization module, and selects the top M scenic spots as the rough ranking results;

[0071] The traffic environment module performs weighted calculation on the top M scenic spots selected by the expert component module to obtain the ranking results of the recommended M scenic spots;

[0072] The decision-making push module performs spatial clustering judgment on the ranking results of the recommended M scenic spots (such as calculating the Moran index or based on K-means), and determines the optimal scenic spot recommendation ranking based on the user's travel time and the number of scenic spots visited.

[0073] Scenic spot data is crawled from websites such as official cultural and tourism media, social media UGC content and OTA platforms. The scenic spot data is related to scenic spot characteristics, tourist characteristics, emotional tendencies (such as user likes and comments) and interest preferences (such as user collections).

[0074] Scenic spot data includes text content and multimodal content (such as pictures, videos, audio, and text); for text content: 1) NLP-based data preprocessing removes irrelevant characters and stop words, and completes text cleaning and standardization operations; 2) Chinese word segmentation is performed based on Jieba, dividing the text into words to form a vocabulary; 3) Part-of-speech tagging is completed based on the Orange visual text analysis tool; 4) Named entity recognition (NER) is implemented based on deep learning models; for multimodal content: in-depth analysis is performed based on AIGC large models (such as the Gemini model and the CLIP model), for example, analyzing tourist characteristics and scenic spot characteristics from videos and pictures.

[0075] User data is crawled from platforms such as operator data platforms, IoT devices (such as passenger flow sensing sensors), and UnionPay consumption data service platforms. This user data is related to user location, travel trajectory, travel mode, behavioral preferences, etc. For example, data crawled from operator data platforms can obtain user location and travel trajectory based on LBS.

[0076] Static attributes include age, education, occupation, family structure, source of tourists, consumption capacity (such as income level, travel budget), etc.; dynamic attributes include departure place, travel trajectory, travel mode (such as self-driving, public transportation), travel preferences (such as independent travel, group travel), length of stay, consumption preferences, cultural interests, activity preferences, dining preferences, seasonal preferences, digital service preferences, technology preferences, scenic spot preferences, etc.

[0077] The intelligent classification module is used to perform topic mining and classification on the data acquired by the data collection module according to the topic requirements of the expert component module, including the following steps:

[0078] The data acquired by the data collection module is analyzed using a topic model algorithm (such as LDA) to generate a document-topic matrix (DT matrix). The document-topic matrix is ​​a two-dimensional matrix U of Q×V, where Q represents the number of documents. A document is a data set consisting of multiple words, and V represents the number of topics. Each row corresponds to an analyzed document data, and each column represents a topic. Each element u in the matrix q,v Indicates the relevance of the qth document to the vth topic (usually a probability value);

[0079] When the relevance exceeds the set threshold, the qth document data in the matrix is ​​considered to belong to the vth topic, and the topic types are user attribute topics, humanities and history topics, cultural and scientific topics, and traffic environment topics. For example, if the threshold is set to 0.8, that is, when the value of an element in the two-dimensional matrix is ​​≥0.8, it means that the document belongs to a certain topic;

[0080] The document data belonging to each topic is transmitted as a whole to the topic module corresponding to the expert component module.

[0081] In one embodiment, the intelligent classification module performs topic mining and classification on the data acquired by the data collection module according to the topic requirements of the expert component module, including the following steps:

[0082] The data acquired by the data collection module is analyzed using a topic model algorithm (such as LDA) to generate a document-topic matrix (DT matrix). The document-topic matrix is ​​a two-dimensional matrix U of Q×V, where Q represents the number of documents. A document is a data set consisting of multiple words, and V represents the number of topics. Each row corresponds to an analyzed document data, and each column represents a topic. Each element u in the matrix q,v Indicates the relevance of the qth document to the vth topic (usually a probability value);

[0083] When the relevance exceeds the set threshold, the qth document data in the matrix is ​​considered to belong to the vth topic, and the topic types are user attribute topics, humanities and history topics, cultural and scientific topics, and traffic environment topics. For example, if the threshold is set to 0.8, that is, when the value of an element in the two-dimensional matrix is ​​≥0.8, it means that the document belongs to a certain topic;

[0084] The document data belonging to each topic is transmitted as a whole to the topic module corresponding to the expert component module.

[0085] The steps for the user analysis module to calculate the recommendation index are as follows:

[0086] The dynamic feature vectors of dynamic attributes are obtained by using the feature extraction algorithm in NLP, and the static feature vectors of static attributes are obtained by using demographic feature encoding;

[0087] The user attribute feature vector p is obtained by combining the dynamic feature vector and the static feature vector;

[0088] The scenic spot feature vector f1 or f2 extracted from the subject education module or quality science popularization module is used to calculate the recommendation index α using the cosine similarity algorithm i , the calculation formula is as follows:

[0089]

[0090] Among them, i=1,2, and this value is determined based on the characteristic values ​​of subject education and quality science popularization preference analysis.

[0091] The dimensions of the scenic spot feature vector and the user attribute feature vector are consistent. If there is no relevant information data in a certain component dimension of the scenic spot feature vector, the component value in this dimension is 0. When the non-zero data feature vector data has positive or negative emotions, the difference between the vector eigenvalues ​​is large, and the recommendation index cannot be directly calculated, the "maximum-minimum" normalization method is used to process the original data into the [0,1] interval. The calculation formula is as follows:

[0092]

[0093] Among them, p′ j Indicates that p j To achieve standardized results, p j is the component of p, max(p) and min(p) represent the maximum and minimum values ​​of the original data.

[0094] The steps for the subject education module to roughly sort the scenic spots in the humanities and history theme data are as follows:

[0095] Establish a POI database for subject education scenic spots:

[0096] Through the API interface, the location of scenic spot addresses in the humanities and history theme data is retrieved to achieve geographic analysis and complete the establishment of the subject education scenic spot POI database;

[0097] The POI database for subject-based educational scenic spots covers scenic spots related to subject knowledge in the compulsory education stage (primary and junior high schools). There are over 300 scenic spots related to parent-child travel (88 in primary school Chinese textbooks, over 20 in junior high school Chinese textbooks, 180 related to Tang poetry, and 110 related to Song poetry), and over 1,600 scenic spots related to celebrity relics. These scenic spots focus on concrete learning and experience of subject knowledge.

[0098] Extract the subject education scenic spot feature vector f1:

[0099] The feature extraction algorithm in NLP is used to extract the feature vector of the scenic spot in the POI database of subject education scenic spots, and the feature vector is preliminarily standardized (Unit Length standardization) to obtain the feature vector f1 of the subject education scenic spot;

[0100] Dynamic NPS score of subject education scenic spots:

[0101] Based on NLTK, sentiment analysis is performed on the scenic spots in the POI database of subject education scenic spots, and the number of positive sentiment evaluations indicating recommendation (N1), the number of negative sentiment evaluations indicating non-recommendation (N2), and the total number of evaluation samples (N) are obtained. The dynamic NPS (Net Recommendation Score) score β of the subject education scenic spots is calculated. NPS , the calculation formula is as follows:

[0102]

[0103] Rough ranking of subject education scenic spots:

[0104] By calculating the recommendation coefficient γ1, the scenic spots in the subject education scenic spot POI database are roughly sorted. The calculation formula is as follows:

[0105] γ1=α1·β NPS

[0106] The steps for the quality science popularization module to roughly sort the scenic spots in the cultural, museum and technology theme data are as follows:

[0107] Establish a POI database for quality science popularization scenic spots:

[0108] Through the API interface, the address name of the scenic spot in the cultural and scientific theme data is retrieved to achieve geographic analysis and complete the establishment of the quality science scenic spot POI database;

[0109] According to the latest data released by the State Administration of Cultural Heritage at the 2024 International Museum Day China Main Venue Event, as of the end of 2023, the total number of registered museums in China has reached 6,833, including 203 first-level, 448 second-level, and 566 third-level museums. The total number of revolutionary history museums and memorial halls in China has exceeded 1,600, and the number of science and technology museums built and open to the public is 477. These scenic spots focus on the comprehensive cultivation and practice of quality education.

[0110] Extract the characteristic vector f2 of the quality science popularization scenic spot:

[0111] The feature extraction algorithm in NLP is used to extract the feature vector of the scenic spot in the POI database of quality science popularization scenic spot, and the feature vector is preliminarily normalized (Unit Length normalization) to obtain the feature vector f2 of the quality science popularization scenic spot;

[0112] Dynamic NPS score of quality science popularization scenic spots:

[0113] Based on NLTK, sentiment analysis is performed on the scenic spots in the POI database of quality science popularization scenic spots, and the number of positive sentiment evaluations indicating recommendation (N1), the number of negative sentiment evaluations indicating non-recommendation (N2), and the total number of evaluation samples (N) are obtained. The dynamic NPS (net recommendation value) score β of quality science popularization scenic spots is calculated. NPS , the calculation formula is as follows:

[0114]

[0115] Rough ranking of quality science popularization scenic spots:

[0116] By calculating the recommendation coefficient γ2, the scenic spots in the POI database of quality science popularization scenic spots are roughly sorted. The calculation formula is as follows:

[0117] γ2=α2·β NPS

[0118] In one embodiment, the traffic environment module performs weighted calculation on the first M scenic spots selected by the expert component module as follows:

[0119] Combined with the user's expected travel time, the traffic environment data is predicted by deep learning neural network, and the traffic environment comfort index ε is weightedly calculated for the γ values ​​of the first M scenic spots selected by the expert component module. The expression is: γ·ε;

[0120] The calculation formula of traffic environment comfort index ε is as follows:

[0121]

[0122] Among them, ω k represents the weight of the comfort index (determined by entropy weight method, AHP method, etc.), s k represents the eigenvalue of the comfort index, K represents the total number of eigenvalues, and the comfort index includes traffic comfort, climate comfort, air comfort, volume comfort, dining comfort, accommodation comfort, etc.

[0123] According to the weighted calculation results, the top M scenic spots selected by the expert component module are re-ranked.

[0124] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A parent-child travel scenic spot recommendation system based on AIGC, characterized by: It includes data collection module, intelligent classification module, expert component module and decision push module. The expert component module includes user analysis module, subject education module, quality science popularization module and traffic environment module. The data acquisition module is used to obtain scenic area data and user data, and combine the user input data to generate data for analyzing user attributes. User attributes include static attributes and dynamic attributes; The intelligent classification module conducts topic mining and classification on the data in the data collection module according to the theme requirements of the expert component module, and transmits user attribute theme data to the user analysis module, humanities and history theme data to the subject education module, cultural and museum science theme data to the quality science popularization module, and traffic environment theme data to the traffic environment module; The user analysis module is used to calculate the recommendation index, the subject education module is used to roughly sort the scenic spots in the humanities and history theme data, and the quality science popularization module is used to roughly sort the scenic spots in the cultural, museum and technology theme data; The expert component module obtains the overall scenic spot ranking based on the rough ranking results of the subject education module and the quality science popularization module, and selects the top M scenic spots as the rough ranking results; The traffic environment module performs weighted calculation on the top M scenic spots selected by the expert component module to obtain the ranking results of the recommended M scenic spots; The decision-making push module performs spatial clustering judgment on the ranking results of the recommended M scenic spots, and determines the optimal recommended ranking of scenic spots based on the user's travel time and the number of scenic spots visited.

2. The AIGC-based parent-child travel scenic spot recommendation system according to claim 1 is characterized in that: The intelligent classification module conducts topic mining and classification on the data in the data collection module according to the topic requirements of the expert component module, including the following steps: The topic model algorithm is used to analyze the data obtained by the data collection module to generate a document-topic matrix. The document-topic matrix is ​​a two-dimensional matrix U of Q×V, where Q represents the number of documents. A document is a data set consisting of multiple words. V represents the number of topics. Each row corresponds to an analyzed document data, and each column represents a topic. Each element u in the matrix q,v Indicates the relevance between the qth document and the vth topic; When the relevance exceeds the set threshold, the qth document data in the matrix is ​​considered to belong to the vth topic; The document data belonging to each topic is transmitted as a whole to the topic module corresponding to the expert component module.

3. The AIGC-based parent-child travel scenic spot recommendation system according to claim 1 is characterized in that: The steps for the user analysis module to calculate the recommendation index are as follows: The dynamic feature vectors of dynamic attributes are obtained by using the feature extraction algorithm in NLP, and the static feature vectors of static attributes are obtained by using demographic feature encoding; The user attribute feature vector p is obtained by combining the dynamic feature vector and the static feature vector; The scenic spot feature vector f1 or f2 extracted from the subject education module or quality science popularization module is used to calculate the recommendation index α using the cosine similarity algorithm i , the calculation formula is as follows: Among them, i=1,2, and this value is determined based on the characteristic values ​​of subject education and quality science popularization preference analysis.

4. The AIGC-based parent-child travel scenic spot recommendation system according to claim 3 is characterized in that: The dimensions of the scenic spot feature vector and the user attribute feature vector are consistent. If there is no relevant information data in a certain component dimension of the scenic spot feature vector, the component value in this dimension is 0. When the non-zero data feature vector data contains positive or negative emotions, the difference between the vector feature values ​​is large, and the recommendation index cannot be directly calculated. The "maximum-minimum" normalization method is used to process the original data into the [0,1] interval. The calculation formula is as follows: Among them, p′ j Indicates that p j To achieve standardized results, p j is the component of p, max(p) and min(p) represent the maximum and minimum values ​​of the original data.

5. The AIGC-based parent-child travel scenic spot recommendation system according to claim 3 is characterized in that: The steps for the subject education module to roughly sort the scenic spots in the humanities and history theme data are as follows: Establish a POI database for subject education scenic spots: Through the API interface, the location of scenic spot addresses in the humanities and history theme data is retrieved to achieve geographic analysis and complete the establishment of the subject education scenic spot POI database; Extract the subject education scenic spot feature vector f1: The feature extraction algorithm in NLP is used to extract the feature vector of the scenic spot in the POI database of subject education scenic spots, and the feature vector is preliminarily standardized to obtain the feature vector f1 of the subject education scenic spot; Dynamic NPS score of subject education scenic spots: Based on NLTK, sentiment analysis is performed on the scenic spots in the POI database of subject education scenic spots, and the number of positive sentiment evaluations indicating recommendation (N1), the number of negative sentiment evaluations indicating non-recommendation (N2), and the total number of evaluation samples (N) are obtained to calculate the dynamic NPS score β of the subject education scenic spot. NPS , the calculation formula is as follows: Rough ranking of subject education scenic spots: By calculating the recommendation coefficient γ1, the scenic spots in the subject education scenic spot POI database are roughly sorted. The calculation formula is as follows: γ1=α1·β NPS。 6. The AIGC-based parent-child travel scenic spot recommendation system according to claim 3 is characterized in that: The steps for the quality science popularization module to roughly sort the scenic spots in the cultural, museum and technology theme data are as follows: Establish a POI database for quality science popularization scenic spots: Through the API interface, the address name of the scenic spot in the cultural and scientific theme data is retrieved to achieve geographic analysis and complete the establishment of the quality science scenic spot POI database; Extract the characteristic vector f2 of the quality science popularization scenic spot: The feature extraction algorithm in NLP is used to extract the feature vector of the scenic spot in the POI database of quality science popularization scenic spots, and the feature vector is preliminarily standardized to obtain the feature vector f2 of the quality science popularization scenic spot; Dynamic NPS score of quality science popularization scenic spots: Based on NLTK, sentiment analysis is performed on the scenic spots in the POI database of quality science popularization scenic spots to obtain the number of positive sentiment evaluations N1 indicating recommendation, the number of negative sentiment evaluations N2 indicating non-recommendation, and the total number of evaluation samples N, and the dynamic NPS score β of the quality science popularization scenic spots is calculated. NPS , the calculation formula is as follows: Rough ranking of quality science popularization scenic spots: By calculating the recommendation coefficient γ2, the scenic spots in the POI database of quality science popularization scenic spots are roughly sorted. The calculation formula is as follows: γ2=α2·β NPS 。 7. The AIGC-based parent-child travel scenic spot recommendation system according to claim 1 is characterized in that: The steps for the traffic environment module to perform weighted calculation on the first M scenic spots selected by the expert component module are as follows: Combined with the user's expected travel time, the traffic environment data is predicted by deep learning neural network, and the traffic environment comfort index ε is weightedly calculated for the γ values ​​of the first M scenic spots selected by the expert component module. The expression is: γ·ε; The calculation formula of traffic environment comfort index ε is as follows: Among them, ω k Indicates the weight of the comfort index, s k represents the eigenvalue of the comfort index, and K represents the total number of eigenvalues; According to the weighted calculation results, the top M scenic spots selected by the expert component module are re-ranked.

Images