A big data-based travel route recommendation method and system

Abstract

The application relates to the technical field of big data, and provides a tourism route recommendation method and system based on big data, which comprises the following steps: acquiring multi-source data including user dimension data, scene dimension data and resource dimension data, and constructing a tourism big data resource pool; constructing a user portrait model, constructing a three-dimensional user portrait containing static features, dynamic features and potential demand features according to the user dimension data; constructing a recommendation engine, extracting user core demand features based on the user portrait, matching the user core demand features with resource features in the tourism big data resource pool, preliminarily screening out resource combinations meeting the conditions, forming multiple candidate routes, and recommending the tourism routes according to the multiple candidate routes. The application not only breaks the limitation of single data dimension in the traditional scheme, but also can realize accurate feature matching, capture the core demand and potential preference of the user, thereby improving the accuracy of the tourism route recommendation and meeting the personalized demand of the user.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and more specifically, to a method and system for recommending travel routes based on big data. Background Technology

[0002] With the digital transformation of the cultural and tourism industry and the rapid iteration of big data technology, tourism consumption is gradually upgrading from standardized sightseeing to personalized experiences. Users' core needs for travel routes have shifted to a combination of adaptability, flexibility, cost-effectiveness, and experiential aspects. Currently, the mainstream travel route recommendation methods are mainly divided into three categories: collaborative filtering recommendations based on users' historical behavior, content recommendations based on attraction tag matching, and rule-based recommendations based on fixed itinerary templates.

[0003] However, the above methods all have obvious technical defects and implementation bottlenecks, making it difficult to meet the diversified needs of current tourism consumption. Specifically, firstly, the data dimensions are too limited. Most solutions only focus on online behavior data such as user browsing and ordering, ignoring offline scenario data and implicit demand data, resulting in insufficient recommendation accuracy. Secondly, the dynamic adaptation capability is weak. Traditional recommendation models are mostly static models, which cannot respond to changes in scenarios and adjustments in user needs in real time, resulting in rigid recommendation results with poor flexibility.

[0004] Therefore, it is necessary to develop a tourism route recommendation method based on big data, which can achieve accurate and personalized tourism route recommendations through multi-source data fusion and dynamic model iteration. Summary of the Invention

[0005] Therefore, in order to address the technical shortcomings of existing tourism route recommendation methods and improve the accuracy and personalization of tourism route recommendations, this invention provides a tourism route recommendation method and system based on big data, the specific technical solution of which is as follows:

[0006] A big data-based method for recommending travel routes includes the following steps: Acquire multi-source data, including user-dimensional data, scenario-dimensional data, and resource-dimensional data; preprocess the multi-source data; and construct a tourism big data resource pool. Build a user profile model, constructing a three-dimensional user profile that includes static features, dynamic features, and potential needs features based on user dimension data; A recommendation engine is built, which extracts the core needs of users based on user profiles, matches them with the resource features in the tourism big data resource pool, initially screens out resource combinations that meet the conditions, forms multiple candidate routes, and recommends tourism routes based on the multiple candidate routes.

[0007] The big data-based travel route recommendation method described above acquires multi-source data to construct a tourism big data resource pool, and builds a three-dimensional user profile containing static features, dynamic features, and potential demand features based on user-dimensional data. This not only breaks the limitation of the single data dimension in traditional solutions, but also achieves accurate feature matching, captures the core needs and potential preferences of users, thereby improving the accuracy of travel route recommendations and meeting users' personalized needs.

[0008] Preferably, recommending a tour route based on multiple candidate routes specifically includes the following steps: Real-time acquisition of scene dimension data, and generation of scene prediction data for the corresponding time period through a time series prediction model based on the preset travel time period of the candidate route; Based on scenario prediction data, candidate routes are pre-optimized and adjusted. When the prediction results of scenic spot congestion, traffic congestion, severe weather warning or temporary closure of scenic spot are detected during the travel period, the alternative resources are automatically replaced to ensure the pre-set feasibility of the candidate routes. After the candidate routes are output, scenario-dimensional data is acquired in real time during the user's trip execution, the route being executed is dynamically adjusted, and the rhythm and resource allocation of the candidate routes are dynamically optimized based on the user's real-time behavior feedback.

[0009] Preferably, recommending a travel route based on multiple candidate routes further includes the following steps: By connecting with offline tourism resource cooperation platforms, we verify the availability of various resources in the candidate routes, eliminate candidate routes with resource conflicts or those that cannot be implemented, and ensure the actual feasibility of the recommended routes.

[0010] Preferably, recommending a travel route based on multiple candidate routes further includes the following steps: Based on regional cultural characteristics, seasonal limited resources, and niche scene elements, we explored differentiated creative elements and incorporated them into candidate routes. We employ preliminary rules to screen core attractions and niche attractions based on geographical space, complementary experiences, and time compatibility, eliminating obviously unreasonable combinations. Then, for the pre-screened attraction combinations, we analyze the co-occurrence correlation between the two through association rule mining algorithms to discover strongly correlated and compatible combinations. Combining this with the characteristics of potential user needs, we select matching combinations from the strongly correlated and compatible combinations to design routes with unique experiential value, while optimizing the itinerary logic of the routes.

[0011] Preferably, recommending a travel route based on multiple candidate routes further includes the following steps: The optimal route is selected based on a comprehensive score obtained from user profile matching, route feasibility, and creative uniqueness, and then presented to the user. Collect user feedback data on recommended routes and incorporate this data into a tourism big data resource pool to optimize user profile models and recommendation engine parameters, thereby enabling continuous iteration of recommendation methods.

[0012] Preferably, the specific method for obtaining the comprehensive score includes the following steps: Obtain the feature importance weights for each dimension and the similarity between the user profile vector and the route resource vector, and obtain the matching score based on the feature importance weights and cosine similarity. Obtain resource availability score, transportation convenience score, and time reasonableness score, and obtain the landing score based on the weighted value of resource availability score, transportation convenience score, and time reasonableness score; Obtain the proportion of niche elements and the score for unique experience; then, obtain a score for creative uniqueness based on the proportion of niche elements and the score for unique experience. A comprehensive score is obtained by weighting the relevance score, practicality score, and creative uniqueness score.

[0013] Preferably, the specific method for obtaining feature importance weights includes the following steps: Initial feature importance weights are obtained based on the feature types of user profiles and combined with prior knowledge of the tourism industry. Based on historical user feedback data, the initial weights are iteratively optimized using a reinforcement learning algorithm to generate real-time feature importance weights, and the cosine similarity between the user profile vector and the route resource vector is calculated as the expected matching degree between the current recommended route and the user's needs. Based on the real-time feature importance weights and the expected matching degree, the final feature importance weights used to calculate the matching degree score are obtained.

[0014] A big data-based travel route recommendation system, used to implement the aforementioned travel route recommendation method, includes: The multi-source data acquisition module is used to acquire multi-source data, including user-dimensional data, scenario-dimensional data, and resource-dimensional data, preprocess the multi-source data, and build a tourism big data resource pool. User profile model is used to construct a three-dimensional user profile that includes static features, dynamic features and potential needs features based on user dimension data; The recommendation engine is used to extract core user needs based on user profiles, match them with resource features in the tourism big data resource pool, initially screen out resource combinations that meet the conditions, form multiple candidate routes, and recommend tourism routes based on the multiple candidate routes.

[0015] Preferably, the recommendation engine includes: The feature matching module is used to extract the core needs features of users based on user profiles, match them with the resource features in the tourism big data resource pool, and initially screen out resource combinations that meet the conditions to form multiple candidate routes. The route recommendation module is used to recommend travel routes based on multiple candidate routes.

[0016] Preferably, the recommendation engine further includes: The dynamic optimization module is used to acquire scene-dimensional data in real time and generate scene prediction data for the corresponding time period based on the preset travel time of the candidate routes using a time series prediction model. Based on the scene prediction data, the candidate routes are pre-optimized and adjusted. When the predicted results of scenic area congestion, traffic congestion, severe weather warnings, or temporary closure of scenic areas are detected during the travel time, the alternative resources are automatically replaced to ensure the preset executability of the candidate routes. After the candidate routes are output, scene-dimensional data is acquired in real time during the user's trip execution to dynamically adjust the route being executed and dynamically optimize the pace and resource allocation of the candidate routes based on real-time user behavior feedback. The resource verification module is used to connect with offline tourism resource cooperation platforms to verify the availability of various resources in the candidate routes, eliminate candidate routes with resource conflicts or that cannot be implemented, and ensure the actual feasibility of the recommended routes. The creative mining module is used to discover differentiated creative elements based on regional cultural characteristics, seasonal limited resources, and niche scene elements, and integrate them into candidate routes. It performs preliminary rule screening of core attractions and niche attractions based on geographical space, experiential complementarity, and time suitability, eliminating obviously unreasonable combinations. Then, for the pre-screened attraction combinations, it analyzes the co-occurrence correlation between the two through association rule mining algorithms, discovers strongly correlated and suitable combinations, and combines the potential needs of users to select matching combinations from the strongly correlated and suitable combinations to design routes with unique experiential value, while optimizing the itinerary logic of the routes. Attached Figure Description

[0017] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0018] Figure 1 This is a schematic diagram of the overall process of a tourism route recommendation method based on big data in one embodiment of the present invention; Figure 2 This is a flowchart illustrating one specific method for recommending tourist routes in an embodiment of the present invention; Figure 3 This is a second flowchart illustrating a specific method for recommending tourist routes in one embodiment of the present invention; Figure 4 This is a third flowchart illustrating a specific method for recommending tourist routes in one embodiment of the present invention; Figure 5 This is a flowchart illustrating a specific method for obtaining a comprehensive score in one embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0020] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.

[0023] Before describing the embodiments of the present invention, a brief introduction to the prior art will be given.

[0024] With the digital transformation of the cultural tourism industry and the rapid iteration of big data technology, tourism consumption is gradually upgrading from standardized sightseeing to personalized experiences. Users' core needs for travel routes have shifted to a combination of adaptability, flexibility, cost-effectiveness, and experiential appeal. Currently, mainstream travel route recommendation methods fall into three categories: collaborative filtering recommendations based on user history, content recommendations based on attraction tag matching, and rule-based recommendations based on fixed itinerary templates. However, all of these methods have significant technical shortcomings and implementation bottlenecks, making it difficult to meet the diversified needs of current tourism consumption. These shortcomings are mainly reflected in the following four aspects: First, the data dimensions are too narrow. Most solutions only focus on online behavioral data such as user browsing and ordering, ignoring offline scenario data such as real-time visitor flow, traffic conditions, weather dynamics, and merchant service quality, as well as implicit demand data such as user travel time, budget constraints, companion composition (elderly, children, etc.), and special preferences (intangible cultural heritage experience, food check-in, niche hidden places, etc.), resulting in insufficient recommendation accuracy.

[0025] Secondly, the dynamic adaptation capability is weak. Traditional recommendation models are mostly static models, which cannot respond in real time to changes in the scene (such as temporary closure of scenic spots, traffic congestion, sudden severe weather) and adjustments in user needs (such as changing the destination midway through the trip or increasing the length of stay). The recommendation results are rigid and lack flexibility.

[0026] Third, there is a lack of feasibility. Some innovative solutions rely on complex algorithm models or scarce data resources, making it difficult to achieve efficient integration with offline tourism resources such as scenic spots, transportation, and hotels. The recommended routes cannot guarantee actual feasibility (such as ticket reservation conflicts, inconvenient transportation connections, and lack of accommodation).

[0027] Fourth, there is a lack of creativity. Most of the solutions fall into the trap of being generic or similar recommendations, failing to incorporate elements such as regional cultural characteristics, seasonally limited resources, and niche special scenes, thus failing to provide users with route combinations that offer unique experiential value.

[0028] At the same time, the tourism industry already has a rich foundation of big data support: smart gates and monitoring equipment deployed in scenic spots can collect real-time data on visitor flow; traffic monitoring systems and public transportation scheduling data from transportation departments can provide references for travel timeliness; online travel agencies (OTAs) have accumulated massive amounts of user consumption behavior data; third-party review platforms (such as Dianping and Mafengwo) have accumulated a large amount of reputation data for attractions and merchants; and in addition, there is public data from meteorological and cultural tourism departments, which lays the foundation for multi-dimensional data integration and application.

[0029] Against this backdrop, there is an urgent need for a big data-based tourism route recommendation method. This method should integrate multi-source data and iterate dynamic models to achieve precise and personalized tourism route recommendations. At the same time, it should explore the value of regional characteristics and niche scenarios to provide users with differentiated tourism experiences and promote the high-quality development of the cultural tourism industry.

[0030] One of the objectives of this invention is to improve the accuracy and personalization of travel itinerary recommendations. To this end, such as... Figure 1 As shown, an embodiment of the present invention provides a travel route recommendation method based on big data, comprising the following steps: S1: Acquire multi-source data including user dimension data, scenario dimension data, and resource dimension data; preprocess the multi-source data; and construct a tourism big data resource pool.

[0031] Specifically, user dimension data is obtained through a combination of online collection and offline surveys. The online user dimension data includes users' browsing history, order data, review data, and favorite data on OTA platforms and travel social platforms. The offline user dimension data includes the composition of users' travel companions, special needs (including elderly care, child suitability, dietary restrictions, preferred scene types, etc.), travel budget, and time constraints. The data is converted into standardized data through structured forms and natural language recognition technology.

[0032] Scene-level data can be obtained by connecting to the open data platforms of scenic spots, transportation, meteorology, and cultural tourism departments through API interfaces. This includes real-time visitor density, opening status, instantaneous carrying capacity, and ticket reservation status of scenic spots; traffic conditions including highways, railways, and aviation; real-time information on public transportation schedules and connecting routes; real-time meteorological data (including temperature, precipitation, wind force, etc.); disaster early warning information; and dynamic scene data such as regional cultural festivals and seasonal limited-time events.

[0033] Resource-level data includes basic information about scenic spots, hotels, restaurants, and specialty merchants (such as location, business hours, services, and pricing), as well as information on cooperation resources (including reservation channels, group discounts, and exclusive service permissions).

[0034] The collected multi-source data is preprocessed, including cleaning (removing outliers and missing values), desensitization (hiding user privacy information and sensitive merchant data), and normalization, to unify the data format and indicator system, thereby constructing a structured tourism big data resource pool.

[0035] The collection frequency of the multi-source data can be dynamically adjusted according to the data type. For example, user behavior data is collected in real time, dynamic scene data such as scenic spot traffic flow, traffic conditions, and weather are collected every 5-15 minutes, basic resource data such as scenic spots and hotels are updated daily, and regional cultural festivals and seasonal limited-time event data are updated weekly.

[0036] S2, Construct a user profile model, and build a three-dimensional user profile that includes static features, dynamic features and potential needs features based on user dimension data.

[0037] Specifically, based on the user dimension data preprocessed in step S1, a three-dimensional user profile containing static features, dynamic features, and potential demand features is constructed using machine learning algorithms. The machine learning algorithm can integrate decision trees, collaborative filtering, and deep learning models.

[0038] Static features include, but are not limited to, user age, gender, place of residence, travel frequency, and historical spending level. Dynamic features include, but are not limited to, current travel time, type of companions, budget range, and real-time location.

[0039] Potential demand characteristics are obtained through text mining of user historical reviews and behavioral sequence analysis, including but not limited to preferred travel types (natural landscapes, historical sites, culinary experiences, family activities, niche adventures), service sensitivities (price, comfort, convenience), and itinerary pace (tight or relaxed). The BERT model can be used to perform semantic analysis on user historical review texts, extracting sentiment and demand keywords. Combined with user behavioral sequences such as continuously viewed attraction types and ordered services, association analysis algorithms can be used to determine potential user preferences, thus enabling the mining of potential demand characteristics for user profiling.

[0040] User profiles are updated in real time. When users generate new behavioral data, such as browsing new attractions, submitting reviews, adjusting their itineraries, or when scene data changes, user profile feature iteration is automatically triggered.

[0041] S3 builds a recommendation engine that extracts core user needs based on user profiles, matches them with resource features in the tourism big data resource pool, initially filters out suitable resource combinations, forms multiple candidate routes, and recommends tourism routes based on these candidate routes.

[0042] Specifically, based on the user profile constructed in step S2, core user needs are extracted and precisely matched with resource features such as attractions, transportation, accommodation, and dining in the tourism big data resource pool. This process initially filters out suitable resource combinations, forming multiple candidate routes. These core user needs include, but are not limited to, time, budget, preference type, travel companions, and service requirements.

[0043] After obtaining multiple candidate routes, you can recommend travel routes, such as randomly selecting a candidate route for recommendation.

[0044] The big data-based travel route recommendation method described above acquires multi-source data to construct a tourism big data resource pool, and builds a three-dimensional user profile containing static features, dynamic features, and potential demand features based on user-dimensional data. This not only breaks through the limitation of the single data dimension in traditional solutions, but also achieves accurate feature matching, captures users' core needs and potential preferences, improves the accuracy of travel route recommendations, meets users' personalized needs, and effectively solves the problem of insufficient accuracy in existing solutions.

[0045] In one embodiment, such as Figure 2 As shown, recommending a travel route based on multiple candidate routes includes the following steps: S31 acquires scene dimension data in real time and generates scene prediction data for the corresponding time period through a time series prediction model based on the preset travel time period of the candidate route.

[0046] S32 pre-optimizes and adjusts candidate routes based on scenario prediction data. When a prediction result of scenic area congestion, traffic congestion, severe weather warning, or temporary closure of the scenic area is detected during the travel period, the alternative resources are automatically replaced to ensure the pre-set feasibility of the candidate routes.

[0047] S33: After the candidate routes are output, scenario-dimensional data is acquired in real time during the user's trip execution. The route being executed is dynamically adjusted, and the candidate route rhythm and resource allocation are dynamically optimized based on the user's real-time behavior feedback.

[0048] Specifically, it acquires scene-specific data in real time, such as pedestrian flow, traffic, weather, and the opening status of scenic spots, and dynamically adjusts candidate routes.

[0049] First, data is collected and processed in different time periods. For the pre-set travel time periods, time series prediction models such as ARIMA and LSTM are used, combined with historical scene data (such as the flow of people, weather, and traffic during the same period in the past 30 days), holiday patterns, and cultural and tourism announcements, to generate scene prediction data for each time period of the trip, such as the predicted flow density of a certain scenic spot from 10:00 to 14:00 on March 20. During the execution of the trip, a real-time data collection frequency of 5-15 minutes is maintained to achieve dynamic adjustment.

[0050] Then, a two-layer dynamic optimization mechanism is implemented. The first is pre-optimization, which eliminates / replaces obviously infeasible resources based on the predicted travel time data during the candidate route screening stage. For example, if the scenic spot is predicted to be closed on the day of the trip, it will be directly replaced with a similar attraction. The second is real-time optimization, which adjusts the route based on the current real-time data during the user's actual trip. For example, if there is a sudden traffic jam during the trip, the transportation route will be changed immediately.

[0051] Finally, data timeliness update rules are implemented. After the candidate routes are output, the scenario prediction data is iteratively updated three times, 24 hours, 12 hours, and 1 hour before the trip begins, to gradually approximate the actual scenario.

[0052] In one embodiment, recommending a travel route based on multiple candidate routes further includes the following steps: connecting with an offline tourism resource cooperation platform to verify the availability of various resources in the candidate routes, eliminating candidate routes with resource conflicts or those that cannot be implemented, and ensuring the actual feasibility of the recommended routes.

[0053] Specifically, this involves connecting with offline tourism resource cooperation platforms such as scenic spot ticketing systems, hotel booking systems, and transportation ticketing systems to verify the availability of various resources in the candidate routes. Routes with resource conflicts or those that cannot be implemented are eliminated to ensure the actual feasibility of the recommended routes. The availability of candidate route resources includes, but is not limited to, ticket reservation slots, remaining hotel rooms, available transportation tickets, and compatibility of connecting services.

[0054] In this way, by collecting dynamic data of the scene in real time (people flow, traffic, weather, scenic area status) and combining it with the offline resource verification mechanism, the recommended routes can be dynamically adjusted and their availability verified, effectively avoiding problems such as scenic area congestion, traffic delays, and resource conflicts.

[0055] Even in the event of unforeseen circumstances such as temporary closure of scenic spots or severe weather, the proposed tourism route recommendation method can quickly replace alternative resources to ensure the smooth execution of the routes, thereby improving the actual success rate of the expected recommended routes and solving the core bottleneck of traditional static recommendation schemes, which are rigid and have poor implementation.

[0056] In one embodiment, such as Figure 3 As shown, recommending a travel route based on multiple candidate routes also includes the following steps: S34, based on regional cultural characteristics, seasonal limited resources and niche scene elements, explores differentiated creative elements and incorporates them into candidate routes.

[0057] S35 uses pre-selection rules to filter core attractions and niche attractions based on geographical space, complementary experiences, and time suitability, eliminating obviously unreasonable combinations. Then, for the pre-selected attraction combinations, the association rule mining algorithm is used to analyze the co-occurrence correlation between the two attractions, discovering strongly associated suitable combinations. Combined with the characteristics of potential user needs, the combination of matching combinations from the strongly associated suitable combinations is selected to design routes with unique experience value, while optimizing the itinerary logic of the routes.

[0058] Specifically, the pre-selection rule filtering is mainly used to eliminate obviously unreasonable combinations and to preprocess data for association rule mining. This includes: 1. Geographical filtering: Setting maximum transportation connection time thresholds for core attractions and less-known attractions, such as ≤1 hour in urban areas and ≤2 hours in suburbs. Actual connection times are calculated based on big data transportation data, and combinations exceeding these thresholds are eliminated; combinations involving route loops are also eliminated. 2. Complementary experience filtering: Based on the attraction's type tag system (e.g., nature / culture / intangible cultural heritage / family-friendly / food), complementary tag rules are set, such as cultural relics → intangible cultural heritage experience, natural landscapes → rural cuisine. Combinations with the same type of tags are eliminated to avoid homogenized experiences. 3. Time-matching filtering: Matching the playtime of core attractions and less-known attractions to ensure that the playtime per session after combination matches the user's itinerary rhythm, such as a total playtime of ≤4 hours for a compact itinerary and ≤6 hours for a relaxed itinerary. Combinations with mismatched playtimes are eliminated.

[0059] For the effective combinations after pre-screening, the support, confidence, and lift are calculated using association rule mining algorithms such as Apriori and FP-Growth to discover strong co-occurrence combinations in tourists' actual visits, such as "80% of tourists who visit core attraction A will also visit niche attraction B", forming a library of strong association combinations between core and niche attractions.

[0060] Finally, a second matching process is performed. The strongly correlated combination library is matched with users' potential needs and characteristics, such as preferences for intangible cultural heritage and parent-child relationships, to design themed routes, ensuring that the combinations not only conform to objective co-occurrence patterns but also match users' personalized needs.

[0061] Based on regional cultural characteristics, seasonally limited resources, and niche scene data, we uncover differentiated creative elements and integrate them into candidate routes. Through association rule mining algorithms, we analyze the compatibility of combinations between niche attractions and core attractions. Combining these with potential user needs, we design routes with unique experiential value, such as immersive intangible cultural heritage routes, seasonally limited natural landscape routes, parent-child interactive themed routes, and niche secret adventure routes. At the same time, we optimize the itinerary logic of the routes, such as avoiding route reversals, reasonably allocating the time spent at each attraction, and taking into account the needs for food and rest.

[0062] More specifically, from the standardized tourism big data resource pool, we extract full-dimensional feature data of core attractions (regional landmarks, popular scenic spots) and niche attractions (non-popular intangible cultural heritage sites, niche natural landscapes, distinctive cultural and creative / food vendors, etc.). The core features include attraction type (natural / cultural / family-friendly / intangible cultural heritage, etc.), geographical location, duration of visit, supporting services (parking / shuttle / family-friendly facilities, etc.), opening hours / seasonal limitations, and correlation features such as transportation connections between attractions and geographical distance. At the same time, we extract creative element data such as regional cultural festivals and seasonal limited-time events as added value support for the routes.

[0063] The core information of potential user needs is extracted from the three-dimensional user profile. The core information is obtained by semantic analysis and behavioral sequence analysis (browsing / ordering behavior) using the BERT model. This includes, but is not limited to, user preferences for travel type (natural landscape / cultural relics / parent-child interaction / intangible cultural heritage experience / niche adventure, etc.), itinerary pace (compact / relaxed), and service sensitivity (price / comfort / convenience). At the same time, personalized demand tags are formed by combining the user's dynamic characteristics (composition of companions, budget).

[0064] The system identifies strong correlations between core and niche attractions and matches them one-to-one or one-to-many with users' personalized needs tags. This allows for targeted selection of suitable attraction combinations. For example, users with a preference for intangible cultural heritage experiences are matched with a combination of "core historical sites + niche immersive intangible cultural heritage experiences," while families with children are matched with a combination of "core family-friendly scenic spots + niche family-friendly craft / nature study sites." The matched attraction combinations are then meticulously designed with a refined itinerary logic, incorporating regional and seasonal creative elements. This includes combining local cultural festivals such as folk temple fairs and seasonal resources such as flower fields / autumn foliage seasons, integrating relevant activities / scenes into the routes to create themed itineraries, such as immersive intangible cultural heritage routes, seasonal natural landscape routes, and niche hidden gem adventure routes.

[0065] In this way, by exploring regional cultural characteristics, seasonally limited resources, and niche scene elements, and combining them with association rule algorithms to design unique route combinations, the limitations of traditional solutions that rely on core attractions and lack creativity are broken. Furthermore, differentiated themed routes can be launched for different user groups, such as families, the elderly, and niche adventure enthusiasts, allowing users to obtain unique travel experiences. At the same time, it helps promote niche attractions and intangible cultural heritage resources, achieving diversified development of the cultural tourism industry and possessing significant industrial value.

[0066] In one embodiment, such as Figure 4 As shown, recommending a travel route based on multiple candidate routes also includes the following steps: S36 uses a comprehensive score based on user profile matching, route feasibility, and creative uniqueness to select the optimal route and output it to the user.

[0067] S37 collects user feedback data on recommended routes and incorporates this feedback data into the tourism big data resource pool to optimize user profile models and recommendation engine parameters, thereby enabling continuous iteration of recommendation methods.

[0068] Specifically, the recommendation engine uses a comprehensive scoring system based on three dimensions: user profile matching degree, route feasibility, and creative uniqueness, to select the top 3 to top 5 optimal routes and output them to the user.

[0069] The output includes, but is not limited to, a detailed itinerary (such as duration of stay, mode of transportation, restaurant recommendations, accommodation arrangements, etc.), resource booking links, cost details, and precautions.

[0070] Collect user feedback data on recommended routes, such as whether the route was adopted, satisfaction during the execution process, and suggestions for modification. Incorporate the feedback data into a big data resource pool to optimize user profile models and recommendation engine parameters, thereby enabling continuous iteration of recommendation methods.

[0071] By collecting user feedback data on recommended routes and continuously iterating user profile models and recommendation engine parameters, a closed-loop operation of data collection, profile construction, recommendation output, and feedback optimization can be achieved. With data accumulation and model optimization, the accuracy, applicability, and creativity of recommendations will continue to improve. In the long run, an adaptive intelligent recommendation system can be formed to adapt to the ever-changing tourism consumption needs, and it has good scalability and sustainability.

[0072] As a preferred technical solution, such as Figure 5 As shown, the specific method for obtaining the comprehensive score includes the following steps. S361: Obtain the feature importance weights for each dimension and the similarity between the user profile vector and the route resource vector, and obtain the matching score based on the feature importance weights and cosine similarity.

[0073] Specifically, the method for obtaining feature importance weights includes the following steps: First, based on the feature types of user profiles (static features, dynamic features, and potential demand features), and combined with prior knowledge of the tourism industry, initial feature importance weights are obtained. Then, based on historical user feedback data, the initial weights are iteratively optimized using a reinforcement learning algorithm to generate real-time feature importance weights. The cosine similarity between the user profile vector and the route resource vector is calculated as the expected matching degree between the current recommended route and the user's needs. Finally, based on the real-time feature importance weights and the expected matching degree, the final feature importance weights used to calculate the matching degree score are obtained.

[0074] Historical user feedback data includes, but is not limited to, the adoption rate and satisfaction level of recommended routes by users who have completed their trips.

[0075] Feature importance weights consist of two acquisition stages: 1. Initial weight acquisition. Based on prior industry knowledge, initial weights are set such as 0.6 for potential demand features, 0.2 for dynamic features, and 0.2 for static features, providing basic weights for single route selection; 2. Iterative weight optimization. Historical user feedback data, such as route adoption rate, satisfaction, and modification suggestions, are incorporated into the reinforcement learning model to continuously iterate the initial weights. The optimized weights are used for route selection for all subsequent users, achieving the goal of becoming more accurate with each use.

[0076] The expected matching degree is obtained by calculating the cosine similarity between the user profile vector and the route resource vector during the candidate route screening stage. It represents the theoretical matching degree between the candidate route and the user's needs, and the value ranges from 0 to 1, providing a basis for fine-tuning the feature importance weights.

[0077] For actual user feedback, before filtering, only historical user feedback data is used to optimize weights, without using the actual feedback from current users. After filtering and after the trip is completed, the actual feedback from current users on the output recommended routes is collected and included in the historical feedback database for subsequent weight iteration optimization, forming a closed loop of data collection - model optimization - route recommendation - feedback collection.

[0078] As a preferred technical solution, the features are first converted into numerical vectors based on user profiles and line resource features to complete the construction of feature vectors. Then, a multimodal similarity algorithm is used to calculate the matching degree between the user profile vector and the line resource vector. Here, weighted cosine similarity can be used to handle high-dimensional features. Finally, the similarity value is mapped to the range of 0-100 points to facilitate the calculation of the subsequent comprehensive score.

[0079] For example, matching score .in, Let represent the vectors of user profile and route resources in the i-th dimension, respectively. Let represent the cosine similarity between the user profile vector and the line resource vector in the i-th feature dimension. These represent the total number of feature dimensions (i.e., the total number of dimensions of the user profile and route resource feature vectors) and the feature importance weight of the i-th dimension, respectively. This matching score measures the degree to which the recommended routes match the user's needs; a higher value indicates a better match.

[0080] The feature importance weight of the i-th dimension .in, These represent the learning rate, actual feedback, and expected matching degree, respectively. The initial weights for feature importance weights are also shown. Based on the feature importance settings, such as setting the initial weight of potential demand features to 0.6, the initial weight of static features to 0.2, and the initial weight of dynamic features to 0.2, and optimizing in real time through the Q-learning algorithm.

[0081] Actual feedback uses binary, with 0 indicating rejection and 1 indicating acceptance. The expected matching degree represents the system's prediction of how well the currently recommended route matches the user's needs. It is generally used to measure the probability or confidence that the system expects the user to accept the recommended route. In essence, the expected matching degree is the probability or confidence that the system predicts the user will accept the recommended route, given the current user profile, route resource features, and weight settings. For example, for each feature dimension i, the cosine similarity between the user profile vector and the route resource vector is first calculated. Then, each similarity value is multiplied by the corresponding feature importance weight. Next, all weighted similarities are summed and divided by the total weight to obtain the weighted average similarity. Finally, this weighted average similarity is used as the expected matching degree.

[0082] For example, if a user profile shows a preference for "niche adventures" (a potential demand feature) and the route resources include "hidden scenic spots", then the cosine similarity is high, and the corresponding matching score is high, reaching over 90 points. If the user feedback is negative, the importance weight of the corresponding feature is automatically reduced to optimize subsequent recommendations.

[0083] Thus, by introducing reinforcement learning and dynamically adjusting weights based on user feedback, the matching accuracy of travel route recommendations can be improved. For example, when users frequently modify their itineraries, dynamic feature weights can be increased.

[0084] In addition, to improve information utilization by leveraging fine-grained feedback, and to avoid the inability to distinguish between reluctant adoption and high approval, the collection of actual user feedback is set after the recommended route is output. All feedback needs to be converted into standardized values ​​of 0-1 and cover the entire process from booking to the end of the trip. Feedback at different nodes corresponds to different matching degree verification scenarios. Collection nodes include, but are not limited to, route adoption / rejection, itinerary modification behavior, real-time feedback during the trip, and comprehensive evaluation after the trip. Collection methods include, but are not limited to, collection through OTA platforms, pop-ups and push questionnaires of cultural tourism APPs, as well as automatic synchronization with scenic spot ticketing systems and hotel booking systems.

[0085] It's important to note that actual user feedback is only used in the next round of weight iteration and does not participate in the overall score calculation for the current route. For example, after optimizing the weights based on feedback from multiple other users, this feedback is then used for route recommendations to the target user.

[0086] The selection of other users is based on the compatibility between their historical recommended routes and the target user's candidate routes. Specifically, a compatibility index is first defined, which represents the user's route constraints, including but not limited to travel time, attractions along the route, and season. Multiple compatibility indices are normalized and then weighted and summed to obtain the compatibility score between other users' historical recommended routes and the target user's candidate routes. The compatibility score is then used to select the routes. For example, the compatibility score for attractions along the route can be defined as: Compatibility score for attractions along the route = Number of overlapping attractions between other users' historical recommended routes and the target user's candidate routes / Number of attractions on the candidate routes; the compatibility score for travel time = 1 - |Travel time of other users' historical recommended routes - Travel time of the target user's candidate routes / Preset maximum travel time|; a compatibility score of 1 is required for the same travel season, otherwise 0.

[0087] S362, obtain resource availability score, transportation convenience score and time rationality score, and obtain landing score based on the weighted value of resource availability score, transportation convenience score and time rationality score.

[0088] Specifically, the resource availability score is based on factors such as the number of ticket reservations, remaining hotel rooms, and available transportation tickets, and is obtained by calculating availability rate (e.g., number of available resources / total number of required resources); the transportation convenience score is based on real-time traffic conditions and connecting routes, and is obtained by calculating average travel time score (e.g., ideal time / actual predicted time); and the time rationality score is based on factors such as itinerary compactness and time allocation, and is obtained by calculating time utilization rate (e.g., actual tour time / total tour time).

[0089] For example, implementation score .in, These represent the resource availability score, transportation convenience score, and time rationality score, respectively. All are weighting coefficients.

[0090] Resource availability score = Availability factor × Number of resources used / Total number of resources required. Availability factor = 1 - Current visitor flow / Maximum carrying capacity of the scenic area. It is used to adjust resource availability to reflect the impact of real-time scenarios and can dynamically weaken the resource availability score in high-risk scenarios.

[0091] Traffic Convenience Score = Congestion Factor × Ideal Time / Actual Predicted Time. The congestion factor, calculated as exp(-0.5 × Congestion Level), quantifies the impact of traffic congestion on time efficiency. It dynamically adjusts the traffic score, decreasing it during periods of congestion to trigger system optimization. The congestion level indicates the severity of traffic conditions, ranging from 0 to 5, with 0 representing no congestion. Data is obtained in real-time from the traffic condition API, including congestion indices for highways, railways, and aviation.

[0092] Time rationality score = Compactness factor × Actual sightseeing time / Total travel time, used to evaluate the time allocation efficiency and itinerary logic of the recommended route. Compactness factor = 1 - (Turnaround distance / Total travel distance), used to evaluate the logic of the itinerary, ranging from 0 to 1; a higher value indicates a more efficient route. The compactness factor can optimize itinerary logic and reduce unnecessary movement.

[0093] S363: Obtain the percentage of niche elements and the score for unique experience. Based on the percentage of niche elements and the score for unique experience, obtain the score for creative uniqueness.

[0094] Specifically, the proportion of niche elements can be obtained by calculating the ratio of niche attractions and distinctive merchants in the route (e.g., the number of niche resources / total resources); the experience uniqueness score can be obtained by analyzing user reviews and attraction descriptions through natural language processing, quantifying uniqueness such as emotional score and theme scarcity score.

[0095] For example, the proportion of niche elements, P = W1 × N1 / N2. Here, N1 represents the number of resources defined as "niche" or "unique" in the recommended route, directly measuring the density of creative elements on the route, such as intangible cultural heritage sites, seasonal limited-time activities, and niche adventure spots. Its purpose is to identify non-mainstream resources and avoid generic recommendation results. N2 represents the total number of resources in the recommended route, including all attractions, restaurants, accommodations, and transportation hubs. It is used to standardize the proportion of niche elements, preventing bias caused by differences in the total number of resources, such as an artificially high proportion of niche elements when there are few resources on short-distance routes. W1 represents the cultural weight factor, used to amplify the creative value of resources with cultural characteristics. A value greater than 1 (e.g., a weight of 1.2 for intangible cultural heritage sites) indicates enhanced importance, while a value less than 1 (e.g., a weight of 1.0 for ordinary resources) represents the baseline level. Its purpose is to strengthen the contribution of regional cultural elements and address the shortcomings of existing technologies in combining regional cultural characteristics.

[0096] The uniqueness score E is calculated as follows: Emotional Score + Scarcity Score. The emotional score represents a user's emotional inclination towards similar resources (range 0-1), where 0 represents negative emotion (e.g., bad reviews) and 1 represents strong positive emotion (e.g., good reviews). This emotional score is primarily used to predict a user's receptiveness to novel experiences and to avoid recommending unique resources that users may not like.

[0097] The scarcity score represents how rare a resource is in the database (ranging from 0 to 1, with higher values ​​indicating greater scarcity). It is used to ensure that recommended resources are unique, such as avoiding popular attractions. It can be determined by calculating the frequency of attractions in the tourism big data resource pool using the inverse document frequency algorithm.

[0098] Creativity and uniqueness score C = (λ × P + (1) λ)×E)×100. The balancing weight λ is used to adjust the relative importance of the proportion of niche elements P and the score of unique experience E in the creativity score.

[0099] S364 calculates a comprehensive score based on a weighted average of relevance score, feasibility score, and creative uniqueness score.

[0100] For example, the overall score = α × matching score + β × practicality score + γ × creativity score. Where α, β, and γ are weighting coefficients, with α ranging from 0.3 to 0.4, β ranging from 0.4 to 0.5, and γ ranging from 0.1 to 0.2, and α + β + γ = 1.

[0101] An embodiment of the present invention also provides a tourism route recommendation system based on big data, used to implement the aforementioned tourism route recommendation method, which includes a multi-source data acquisition module, a user profile model, and a recommendation engine.

[0102] The multi-source data acquisition module is used to acquire multi-source data including user dimension data, scenario dimension data, and resource dimension data, preprocess the multi-source data, and build a tourism big data resource pool; the user profile model is used to construct a three-dimensional user profile including static features, dynamic features, and potential demand features based on user dimension data.

[0103] The recommendation engine is used to extract core user needs based on user profiles, match them with resource features in the tourism big data resource pool, initially screen out resource combinations that meet the conditions, form multiple candidate routes, and recommend tourism routes based on the multiple candidate routes.

[0104] In this way, by integrating online and offline multi-dimensional data, a three-dimensional user profile is constructed, encompassing static, dynamic, and potential demand characteristics, breaking through the limitations of traditional solutions with their single data dimension. Based on precise feature matching, the recommendation engine can accurately capture users' core needs and potential preferences, avoiding generic recommendations, improving recommendation accuracy, and meeting users' personalized needs.

[0105] Specifically, the recommendation engine includes a feature matching module and a route recommendation module.

[0106] The feature matching module is used to extract core user needs features based on user profiles and match them with resource features in the tourism big data resource pool to initially screen out resource combinations that meet the conditions and form multiple candidate routes; the route recommendation module is used to recommend tourism routes based on multiple candidate routes.

[0107] The recommendation engine also includes a dynamic optimization module, a resource verification module, and a creative idea mining module.

[0108] The dynamic optimization module is used to acquire scene-dimensional data in real time and generate scene prediction data for the corresponding time period through a time series prediction model based on the preset travel time of the candidate routes. Based on the scene prediction data, the candidate routes are pre-optimized and adjusted. When the predicted results of scenic area congestion, traffic congestion, severe weather warnings, or temporary closure of scenic areas are detected during the travel time, the alternative resources are automatically replaced to ensure the preset executability of the candidate routes. After the candidate routes are output, scene-dimensional data is acquired in real time during the user's trip execution to dynamically adjust the route being executed. Based on the user's real-time behavior feedback, the rhythm and resource configuration of the candidate routes are dynamically optimized.

[0109] In this way, by constructing a time-segmented scenario data system, and obtaining predicted data for the trip period as well as real-time updated data before the trip, a two-layer dynamic optimization can be achieved, which uses predicted data to optimize the preset trip and real-time data to adjust the trip during execution, thus solving the problem of mismatch between the timeliness of travel route recommendations.

[0110] The resource verification module is used to connect with offline tourism resource cooperation platforms to verify the availability of various resources in the candidate routes, eliminate candidate routes with resource conflicts or those that cannot be implemented, and ensure the actual feasibility of the recommended routes.

[0111] By collecting dynamic data from real-time scenarios (people flow, traffic, weather, scenic area status) and combining it with offline resource verification mechanisms, the recommended routes can be dynamically adjusted and their availability verified. This can effectively avoid problems such as scenic area congestion, traffic delays, and resource conflicts. Even in the event of emergencies (such as temporary closure of scenic areas or severe weather), alternative resources can be quickly replaced to ensure the smooth execution of the routes. This improves the actual success rate of the expected recommended routes and solves the core bottleneck of traditional static recommendation schemes, which are rigid and have poor implementability.

[0112] The creative mining module is used to discover differentiated creative elements based on regional cultural characteristics, seasonal limited resources, and niche scene elements, and integrate them into candidate routes. It performs preliminary rule screening of core attractions and niche attractions based on geographical space, experiential complementarity, and time suitability, eliminating obviously unreasonable combinations. Then, for the pre-screened attraction combinations, it analyzes the co-occurrence correlation between the two through association rule mining algorithms, discovers strongly correlated and suitable combinations, and combines them with the characteristics of potential user needs to select matching combinations from the strongly correlated and suitable combinations to design routes with unique experiential value, while optimizing the itinerary logic of the routes.

[0113] By exploring regional cultural characteristics, seasonal resources, and niche scene elements through the creative mining module, and combining them with association rule algorithms to design unique route combinations, the traditional approach breaks through the limitations of relying on core attractions and lacking creativity.

[0114] In addition, offering differentiated themed routes for different user groups (families, seniors, niche adventure enthusiasts, etc.) allows users to have unique travel experiences, while also helping to promote niche attractions, intangible cultural heritage, and other resources, thus achieving diversified development of the cultural tourism industry and possessing significant industrial value.

[0115] The recommendation engine also includes a rating acquisition module and an iterative optimization module.

[0116] The rating acquisition module is used to obtain a comprehensive rating based on user profile matching degree, route feasibility, and creative uniqueness, and to select the best route and output it to the user; the iterative optimization module is used to collect user feedback data on recommended routes, and incorporate the feedback data into the tourism big data resource pool to optimize the user profile model and recommendation engine parameters, so as to achieve continuous iteration of the recommendation method.

[0117] In summary, the big data-based travel route recommendation system acquires multi-source data to construct a tourism big data resource pool, and builds a three-dimensional user profile containing static features, dynamic features, and potential demand features based on user-dimensional data. This not only breaks through the limitations of traditional solutions with their single data dimension, but also achieves accurate feature matching, captures users' core needs and potential preferences, improves the accuracy of travel route recommendations, meets users' personalized needs, and effectively solves the problem of insufficient accuracy in existing recommendation solutions.

[0118] The technical features of the embodiments described can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for recommending travel routes based on big data, characterized in that, Includes the following steps: Acquire multi-source data, including user-dimensional data, scenario-dimensional data, and resource-dimensional data; preprocess the multi-source data; and construct a tourism big data resource pool. Build a user profile model, constructing a three-dimensional user profile that includes static features, dynamic features, and potential needs features based on user dimension data; A recommendation engine is built, which extracts the core needs of users based on user profiles, matches them with the resource features in the tourism big data resource pool, initially screens out resource combinations that meet the conditions, forms multiple candidate routes, and recommends tourism routes based on the multiple candidate routes.

2. The method for recommending travel routes based on big data as described in claim 1, characterized in that, Recommending a travel itinerary based on multiple candidate routes includes the following steps: Real-time acquisition of scene dimension data, and generation of scene prediction data for the corresponding time period through a time series prediction model based on the preset travel time period of the candidate route; Based on scenario prediction data, candidate routes are pre-optimized and adjusted. When the prediction results of scenic spot congestion, traffic congestion, severe weather warning or temporary closure of scenic spot are detected during the travel period, the alternative resources are automatically replaced to ensure the pre-set feasibility of the candidate routes. After the candidate routes are output, scenario-dimensional data is acquired in real time during the user's trip execution, the route being executed is dynamically adjusted, and the rhythm and resource allocation of the candidate routes are dynamically optimized based on the user's real-time behavior feedback.

3. The method for recommending travel routes based on big data as described in claim 2, characterized in that, Recommending a travel itinerary based on multiple candidate routes also includes the following steps: By connecting with offline tourism resource cooperation platforms, we verify the availability of various resources in the candidate routes, eliminate candidate routes with resource conflicts or those that cannot be implemented, and ensure the actual feasibility of the recommended routes.

4. The method for recommending travel routes based on big data as described in claim 3, characterized in that, Recommending a travel itinerary based on multiple candidate routes also includes the following steps: Based on regional cultural characteristics, seasonal limited resources, and niche scene elements, we explored differentiated creative elements and incorporated them into candidate routes. We employ preliminary rules to screen core attractions and niche attractions based on geographical space, complementary experiences, and time compatibility, eliminating obviously unreasonable combinations. Then, for the pre-screened attraction combinations, we analyze the co-occurrence correlation between the two through association rule mining algorithms to discover strongly correlated and compatible combinations. Combining this with the characteristics of potential user needs, we select matching combinations from the strongly correlated and compatible combinations to design routes with unique experiential value, while optimizing the itinerary logic of the routes.

5. The method for recommending travel routes based on big data as described in claim 4, characterized in that, Recommending a travel itinerary based on multiple candidate routes also includes the following steps: The optimal route is selected based on a comprehensive score obtained from user profile matching, route feasibility, and creative uniqueness, and then presented to the user. Collect user feedback data on recommended routes and incorporate this data into a tourism big data resource pool to optimize user profile models and recommendation engine parameters, thereby enabling continuous iteration of recommendation methods.

6. The method for recommending travel routes based on big data as described in claim 5, characterized in that, The specific methods for obtaining the overall score include the following steps: Obtain the feature importance weights for each dimension and the similarity between the user profile vector and the route resource vector, and obtain the matching score based on the feature importance weights and cosine similarity. Obtain resource availability score, transportation convenience score, and time reasonableness score, and obtain the landing score based on the weighted value of resource availability score, transportation convenience score, and time reasonableness score; Obtain the proportion of niche elements and the score for unique experience; then, obtain a score for creative uniqueness based on the proportion of niche elements and the score for unique experience. A comprehensive score is obtained by weighting the relevance score, practicality score, and creative uniqueness score.

7. The method for recommending travel routes based on big data as described in claim 6, characterized in that, The specific method for obtaining feature importance weights includes the following steps: Initial feature importance weights are obtained based on the feature types of user profiles and combined with prior knowledge of the tourism industry. Based on historical user feedback data, the initial weights are iteratively optimized using a reinforcement learning algorithm to generate real-time feature importance weights, and the cosine similarity between the user profile vector and the route resource vector is calculated as the expected matching degree between the current recommended route and the user's needs. Based on the real-time feature importance weights and the expected matching degree, the final feature importance weights used to calculate the matching degree score are obtained.

8. A big data-based travel route recommendation system, used to implement the travel route recommendation method as described in any one of claims 1-7, characterized in that, The tourism route recommendation system includes: The multi-source data acquisition module is used to acquire multi-source data, including user-dimensional data, scenario-dimensional data, and resource-dimensional data, preprocess the multi-source data, and build a tourism big data resource pool. User profile model is used to construct a three-dimensional user profile that includes static features, dynamic features and potential needs features based on user dimension data; The recommendation engine is used to extract core user needs based on user profiles, match them with resource features in the tourism big data resource pool, initially screen out resource combinations that meet the conditions, form multiple candidate routes, and recommend tourism routes based on the multiple candidate routes.

9. A tourism route recommendation system based on big data as described in claim 8, characterized in that, The recommendation engine includes: The feature matching module is used to extract the core needs features of users based on user profiles, match them with the resource features in the tourism big data resource pool, and initially screen out resource combinations that meet the conditions to form multiple candidate routes. The route recommendation module is used to recommend travel routes based on multiple candidate routes.

10. A travel route recommendation system based on big data as described in claim 9, characterized in that, The recommendation engine also includes: The dynamic optimization module is used to acquire scene-dimensional data in real time and generate scene prediction data for the corresponding time period based on the preset travel time of the candidate routes using a time series prediction model. Based on the scene prediction data, the candidate routes are pre-optimized and adjusted. When the predicted results of scenic area congestion, traffic congestion, severe weather warnings, or temporary closure of scenic areas are detected during the travel time, the alternative resources are automatically replaced to ensure the preset executability of the candidate routes. After the candidate routes are output, scene-dimensional data is acquired in real time during the user's trip execution to dynamically adjust the route being executed and dynamically optimize the pace and resource allocation of the candidate routes based on real-time user behavior feedback. The resource verification module is used to connect with offline tourism resource cooperation platforms to verify the availability of various resources in the candidate routes, eliminate candidate routes with resource conflicts or that cannot be implemented, and ensure the actual feasibility of the recommended routes. The creative mining module is used to discover differentiated creative elements based on regional cultural characteristics, seasonal limited resources, and niche scene elements, and integrate them into candidate routes. It performs preliminary rule screening of core attractions and niche attractions based on geographical space, experiential complementarity, and time suitability, eliminating obviously unreasonable combinations. Then, for the pre-screened attraction combinations, it analyzes the co-occurrence correlation between the two through association rule mining algorithms, discovers strongly correlated and suitable combinations, and combines the potential needs of users to select matching combinations from the strongly correlated and suitable combinations to design routes with unique experiential value, while optimizing the itinerary logic of the routes.

Images