Collaborative recommendation method and system for tourist self-created itineraries and officially selected itineraries

Abstract

The invention is suitable for the technical field of itinerary recommendation, and provides a collaborative recommendation method and system for a tourist self-created itinerary and an official selected itinerary, and the method comprises the steps: obtaining behavior data of the user self-created itinerary, multi-dimensional evaluation data of the official selected itinerary, and environment context data; the user interest matching degree is calculated according to the behavior data, a travel freedom degree index is generated based on modification frequency, cooperation modes are analyzed and determined through preset conditions in combination with the environment context data, and the cooperation modes comprise a user dominant mode, an official dominant mode, a balanced cooperation mode, a theme guide mode and a real-time emergency mode; according to the cooperation mode, a corresponding recommendation strategy is adopted to adjust a user self-created journey; and generating an optimized journey scheme, carrying out visual display, and updating user data and an official data source. According to the invention, efficient cooperation of diversified travel itinerary schemes is realized, and the intelligent level and user experience of travel itinerary planning are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of itinerary recommendation, and in particular to a method and system for collaboratively recommending tourist-created itineraries and officially selected itineraries. Background Art

[0002] With the rapid development of the tourism industry and the growing demand for personalized travel, tourists are increasingly turning to self-created itinerary planning tools to design personalized itineraries. Meanwhile, official itineraries provided by scenic area management departments have become an important reference for tourists due to their professionalism, safety, and thematic coherence. However, existing technologies still lack a coherent recommendation mechanism between user-created itineraries and officially recommended itineraries. For example, existing systems often employ a static weighting mechanism, failing to adjust recommendation patterns based on dynamic factors such as the frequency of user itinerary modifications and changes in environmental risks. Furthermore, the behavioral data of user-created itineraries lacks synergy with the multi-dimensional evaluation data of official attractions, resulting in recommendations that lack actionability for itineraries. Furthermore, users' acceptance or rejection of recommendations is not effectively leveraged in the iterative optimization model, leading to a decline in long-term recommendation quality. These issues can easily lead to a conflict between personalization, practicality, and safety in recommendation results.

[0003] In view of this, a method and system for collaboratively recommending tourist-created itineraries and official selected itineraries is needed. Summary of the Invention

[0004] The embodiment of the present application provides a collaborative recommendation method for tourist-created itineraries and official selected itineraries, which is used to solve the contradiction between personalization, practicality and security of recommendation results.

[0005] A first aspect of the embodiments of the present application provides a collaborative recommendation method for tourist-created itineraries and official selected itineraries, including:

[0006] Obtain behavioral data on user-created itineraries, multi-dimensional evaluation data on official selected itineraries, and environmental context data;

[0007] Calculate the user interest matching degree based on the behavior data, generate a travel freedom index based on the modification frequency, and determine the collaborative mode through preset condition analysis in combination with the environmental context data. The collaborative modes include user-led mode, official-led mode, balanced collaborative mode, theme-guided mode, and real-time emergency mode.

[0008] According to the collaborative model, a corresponding recommendation strategy is used to adjust the user's self-created itinerary, including:

[0009] In the user-driven mode, the user's core attraction sequence is retained, and official attractions are recommended based on the user's interest matching degree;

[0010] In the official-led mode, the highest priority official attractions are inserted based on the attraction scores and thematic relevance in the multi-dimensional evaluation data;

[0011] Dynamically allocate weights between user-created attractions and official attractions in a balanced collaborative mode;

[0012] In the topic-guided mode, a coherent route is generated by combining the topic relevance of the multi-dimensional assessment data;

[0013] Replace risky scenic spots based on the safety status of the multi-dimensional assessment data in real-time emergency mode;

[0014] Generate optimized itinerary plans and present them visually, updating user data and official data sources.

[0015] Furthermore, the calculation of user interest matching based on the behavior data, the generation of a travel freedom index based on the modification frequency, and the determination of a collaborative mode through analysis of preset conditions in combination with the environmental context data include:

[0016] Perform word segmentation on the search records, favorite tags, and official attraction description texts in the user's historical behavior data to generate a joint corpus, which includes a user behavior corpus sub-corpus and an official attraction corpus sub-corpus;

[0017] Based on the joint corpus, an improved TF-IDF algorithm is used to generate user interest vectors and official attraction vectors respectively, and the user historical behavior data is weighted by a preset time decay factor to generate a weighted user interest vector;

[0018] The cosine similarity between the weighted user interest vector and the official scenic spot vector is calculated, and the calculated cosine similarity is integrated with the proportion of stay time and the score deviation in the user behavior characteristics to output an interest matching score.

[0019] Furthermore, the calculation of user interest matching based on the behavior data, the generation of a travel freedom index based on the modification frequency, and the determination of a collaborative mode based on preset condition analysis in combination with the environmental context data may also include:

[0020] Extract the dynamic frequency, semantic adjustment depth, and spatial originality characteristics of user itinerary modification behavior. The semantic adjustment depth is quantified by the change ratio of attraction types and the increase or decrease of theme keywords. The spatial originality is calculated based on the geographical distribution dispersion of attractions.

[0021] Combined with the real-time traffic prediction level and pedestrian flow in the environmental context data, the behavioral characteristics are dynamically weighted and fused through a gating mechanism to generate an initial degree of freedom index;

[0022] The initial degree of freedom index is dynamically graded, and when an unexpected risk is detected, a forced downgrade operation is performed to output five levels of travel degree of freedom index.

[0023] Furthermore, the preset conditions include:

[0024] When the user interest matching degree is higher than the preset first matching degree threshold and the freedom rating reaches the highest level of the preset evaluation system, the user-led mode is activated;

[0025] When the user interest matching degree is lower than the preset second matching degree threshold, and the official scenic spot score exceeds the official scenic spot score benchmark value and the theme relevance meets the preset relevance standard, the official dominant mode is triggered;

[0026] When the user interest matching degree is between the first matching degree threshold and the second matching degree threshold, and the spatial distribution overlap between the user's planned attractions and the officially recommended attractions reaches a preset overlap ratio, the balanced collaboration mode is enabled;

[0027] When it is detected that the user's itinerary contains multiple different theme keywords that meet a preset number of conditions, and there are recommended routes in the official theme library that meet the preset association conditions with the theme keywords, the theme guidance mode is activated;

[0028] When environmental monitoring data triggers the scenic area safety warning standard or the real-time passenger flow data exceeds the scenic area carrying capacity warning threshold, it switches to real-time emergency mode.

[0029] Furthermore, in the user-driven mode, the user's core attraction sequence is retained, and official attractions are recommended based on the user's interest matching, including:

[0030] Based on the dwell time distribution, visit frequency statistics and user-initiated annotation information in the user's historical behavior data, a set of scenic spots that meet the preset dwell time threshold, visit frequency threshold or core mark identification is extracted to construct an unchangeable core scenic spot sequence;

[0031] Based on the user interest matching degree, a set of candidate attractions whose textual semantic similarity with the core attraction sequence is higher than a preset semantic threshold is selected from the official attraction database, and collaborative filtering and sorting is performed in combination with the user's historical preference tags;

[0032] Based on the real-time spatial distribution characteristics of the environmental context data, candidate attractions are screened for geographic proximity, and official attractions whose path distance to at least one attraction in the core attraction sequence is less than a preset proximity threshold are retained;

[0033] The filtered official attractions are inserted into the idle time of non-core attractions in the user's itinerary in descending order of interest matching, and the recommendation strategy is adjusted based on user feedback behavior.

[0034] Furthermore, in the official-led mode, the highest priority official attractions are inserted based on the attraction scores and theme relevance in the multi-dimensional evaluation data, including:

[0035] Generate priority weights based on real-time dynamic scores, historical comprehensive scores, and topic relevance scores in the constructed multi-dimensional evaluation data;

[0036] Screen candidate official attractions that meet the preset dynamic score threshold and theme relevance threshold, and generate a ranked list in descending order of priority weight;

[0037] Based on the time and space distribution gaps of non-core attractions in the user-created itinerary, the highest priority official attractions are inserted into the idle time period that meets the following conditions: the commuting time between the inserted attraction and the adjacent user attractions is less than a preset time threshold, and the total time deviation of the inserted itinerary does not exceed a preset time tolerance range;

[0038] The priority evaluation model is updated according to the real-time feedback behavior of the user on the inserted scenic spots.

[0039] Furthermore, the weights of user-created attractions and official attractions are dynamically allocated in the balanced collaboration mode, including:

[0040] Generate an initial weight distribution ratio based on the user's interest matching and the theme relevance score in the official attractions' multi-dimensional evaluation data;

[0041] The weight ratio is adjusted based on the travel time redundancy and spatial distribution dispersion in the environmental context data. The adjustment includes: when the time redundancy is higher than the preset redundancy threshold, the weight of the user-created attractions is increased by linear interpolation; when the spatial dispersion exceeds the preset dispersion threshold, the weight of the official attractions is increased based on the path continuity optimization algorithm;

[0042] According to the adjusted weight ratio, user-created attractions and official attractions are mixed and sorted to generate a set of candidate itinerary plans that meet the weight constraints.

[0043] Furthermore, in the topic-guided mode, generating a coherent route by combining the topic relevance of the multi-dimensional evaluation data includes:

[0044] Model the topic probability distribution of the attraction description text in the user's self-created itinerary, extract the latent topic vector and the inter-topic transition probability matrix;

[0045] Perform similarity matching between the potential topic vector and the feature vectors of predefined topic routes in the official topic library, and select candidate topic routes with a relevance score higher than a preset matching threshold;

[0046] Calculating the inter-attraction logical correlation based on the inter-theme transition probability matrix and the spatiotemporal commuting time of the candidate theme routes;

[0047] When the correlation exceeds the preset coherence threshold, an enhanced route containing temporal and spatial sequence and thematic narrative logic is generated, and the guide commentary data in the official thematic library is embedded.

[0048] Furthermore, in the real-time emergency mode, replacing risky scenic spots based on the safety status of the multi-dimensional assessment data includes:

[0049] IoT sensors collect real-time data on geological activity intensity, facility failure signals, and heat maps of crowd density at scenic spots to calculate dynamic safety risk indexes.

[0050] When the risk index continues to exceed the preset safety threshold for a monitoring window period, the official safe scenic spot database is screened for candidate scenic spots that meet the conditions. The conditions include: the theme relevance score with the risky scenic spot is higher than the preset emergency relevance threshold, and the commuting time from the user's current location is less than the preset emergency time threshold;

[0051] Based on the tabu search algorithm and dynamic time warping model, the optimal path sequence of alternative attractions is generated.

[0052] A second aspect of the present application provides a collaborative recommendation system for tourist-created itineraries and official selected itineraries, including:

[0053] A data acquisition unit, used to obtain behavioral data of user-created itineraries, multi-dimensional evaluation data of official selected itineraries, and environmental context data;

[0054] a collaborative mode determination unit, configured to calculate a user interest matching degree based on the behavioral data, generate a travel freedom index based on the modification frequency, and determine a collaborative mode based on preset conditions in combination with the environmental context data. The collaborative modes include user-led mode, official-led mode, balanced collaborative mode, theme-guided mode, and real-time emergency mode;

[0055] An itinerary adjustment unit is configured to adjust the user's self-created itinerary using a corresponding recommendation strategy according to the collaborative mode, including: retaining the user's core attraction sequence in the user-led mode and recommending official attractions based on the user's interest matching degree; inserting the highest priority official attractions based on the attraction scores and theme relevance in the multi-dimensional evaluation data in the official-led mode; dynamically allocating the weights of user-created attractions and official attractions in the balanced collaborative mode; generating a coherent route based on the theme relevance of the multi-dimensional evaluation data in the theme-guided mode; and replacing risky attractions based on the safety status of the multi-dimensional evaluation data in the real-time emergency mode.

[0056] The optimized itinerary plan generation unit is used to generate the optimized itinerary plan and display it visually, and update user data and official data sources.

[0057] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0058] After acquiring data, this invention quantifies user preference strength through interest matching, dynamically assesses user control needs through an itinerary freedom indicator, and switches collaborative modes in real time based on environmental risks. This allows for adaptive matching of user intent with environmental conditions, balancing personalization and safety. It also adjusts user-created itineraries based on different collaborative modes, enabling the planning of diverse itinerary scenarios. Finally, it provides real-time feedback to the model on user interactions with recommended options, updating user preference profiles and official data ratings to improve recommendation accuracy. Through multi-source data fusion, dynamic mode switching, and feedback-driven adaptive optimization, this approach overcomes technical drawbacks such as rigid policies, fragmented data, lack of environmental awareness, and weak feedback, enabling efficient collaboration across diverse itinerary options and effectively enhancing the intelligence and user experience of travel itinerary planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The figure is a flow chart of an embodiment of a method for collaboratively recommending tourist-created itineraries and official selected itineraries in the present invention. DETAILED DESCRIPTION

[0060] The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "corresponding to," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0061] Example 1

[0062] The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, and the specific implementation is not clearly limited. The following will introduce the collaborative recommendation method of the tourist's self-created itinerary and the official selected itinerary in this application from the perspective of system implementation. Figure 1 , the method provided in the embodiment of the present application includes the following steps:

[0063] S11. Obtaining behavioral data on user-created itineraries, multi-dimensional evaluation data on official selected itineraries, and environmental context data;

[0064] In this embodiment, the user's itinerary creation, modification, and deletion events are collected through the mobile app or web terminal, and the operation type, timestamp, and difference in the sequence of attractions before and after the modification are recorded. GPS positioning and matching with the electronic fence of the attraction are used to count the user's stay time, click frequency, favorite mark, and 1-5 star rating for each attraction. Itinerary tags filled in by users, such as "family travel" and "historical exploration", and free text descriptions such as "ancient town suitable for taking pictures" are parsed. The acquired behavioral data is normalized by time window and abnormal behavior is filtered to obtain pre-processed data.

[0065] The multi-dimensional evaluation data for official selected itineraries includes both static and dynamic data. Static evaluation data is obtained from the scenic area management department's API, including historical comprehensive ratings of attractions, theme tags such as "cultural heritage" and "natural scenery," opening schedules, and safety ratings. A third-party platform is used to crawl real-time attraction reputation ratings, daily reservations, and emergency announcements. This multi-dimensional evaluation data is then used to calculate thematic relevance. This relevance score is generated based on the semantic similarity between the attraction's theme tags and the officially defined theme routes.

[0066] Environmental context data includes data collected by IoT sensors, such as temperature and humidity sensors deployed in scenic areas, crowd counting cameras, and geological monitoring equipment that detects vibration frequency. These sensors collect real-time information on weather conditions, crowd density, and facility operating status. Furthermore, a traffic data interface, connected to the AutoNavi / Baidu Maps API, is used to obtain real-time traffic control zones, road congestion indices, and public transportation schedules. This data is standardized, for example, to calculate crowd load factors or synthesize environmental risk indices.

[0067] S12. Calculate user interest matching based on behavioral data, generate a travel freedom index based on modification frequency, and combine it with environmental context data to determine a collaborative model through pre-condition analysis. Collaborative models include user-led, official-led, balanced, theme-guided, and real-time emergency response.

[0068] Step S12 includes the following:

[0069] S121. Perform word segmentation on the search history, favorite tags, and official attraction description text in the user's historical behavior data to generate a joint corpus, which includes a user behavior corpus sub-corpus and an official attraction corpus sub-corpus;

[0070] Perform multimodal text preprocessing on the search history, collection tags and official scenic spot description text in the user's historical behavior data: segment the original text using the Chinese word segmentation tool, remove irrelevant words in the preset stop word list, and use the Porter stem extraction algorithm to standardize the remaining words. Construct user interest corpus D user With the official attraction corpus D off , generate a joint vocabulary V = {w1,w2,…w n}, generating vector representation based on the improved TF-IDF algorithm.

[0071] S122. Based on the joint corpus, an improved TF-IDF algorithm is used to generate user interest vectors and official attraction vectors, and the user historical behavior data is weighted by a preset time decay factor to generate a weighted user interest vector;

[0072] Calculate user-side text d user Chinese vocabulary w i word frequency (i=1,2,…,n) Calculate global inverse document frequency And smooth the low-frequency words; generate the user interest vector v user =[TF-IDF(w1),…,TF-IDF(w n )] and official attractions vector v off , and perform L2 normalization.

[0073] Introduce the time decay factor to dynamically modify the user interest vector: according to the user behavior timestamp t k , calculate the attenuation weight Where λ is set according to the user activity level, and the weighted user interest vector is updated to

[0074] Specifically, when the user behavior frequency is higher than the set frequency value, set λ = 0.01 days -1 , to reduce the decay rate of historical data; when the user behavior frequency is lower than the set frequency value, set λ = 0.05 days -1 , accelerating the weight decay of old behavior data; if it is detected that the user has recently made intensive itinerary modification operations, λ will be temporarily adjusted to 0.03 days -1 , to balance short-term preferences with long-term interests.

[0075] S123. Calculate the cosine similarity between the weighted user interest vector and the official attraction vector, combine the calculated cosine similarity with the proportion of stay time and the score deviation in the user behavior characteristics, and output the interest matching score.

[0076] Calculate the semantic similarity between the user interest vector and the official attraction vector: using the cosine similarity formula The result is linearly mapped to the range of 0-1 as the final interest matching score. Specifically, the multimodal fusion of the cosine similarity calculation results includes:

[0077] Extract the proportion of stay time in user behavior characteristics s stay and rating deviation Δ score , through linear interpolation, the behavior similarity Sim behavior =0.6s stay +0.4(1-Δ score ) and semantic similarity Sim cosine Fusion, generate the final interest matching score Score final :

[0078] Score final =γ·Sim cosine +(1-γ)·Sim behavior (γ∈[0.7,0.9])

[0079] Here, γ is automatically adjusted according to the completeness of the text description in the user's self-created itinerary: when the number of words in the user's itinerary text is <20, γ = 0.7, otherwise γ = 0.9.

[0080] Step S12 also includes the following:

[0081] S124. Extract the dynamic frequency, semantic adjustment depth, and spatial originality characteristics of user itinerary modification behavior. The semantic adjustment depth is quantified by the change ratio of attraction types and the increase or decrease of theme keywords. The spatial originality is calculated based on the geographical distribution dispersion of attractions.

[0082] Count the number of times a user modifies their itinerary within different time periods, and use a time decay factor to make recent modifications contribute more significantly to the frequency value. For example, a user who modified it three times yesterday is weighted higher than one who modified it five times last week. By dividing the time window and weighting the modification amplitude, a dynamic frequency index reflecting the user's real-time activity is generated. Analyze the change ratio of attraction types before and after each user modification. For example, changing "museum" to "amusement park" is considered a type change, and count the increase or decrease in thematic keywords. Combined with the type change ratio and keyword change amount, calculate the semantic impact depth of the user's modification behavior, distinguishing between superficial adjustments and substantive changes. Identify user-defined attractions and calculate their geographical dispersion. For example, when user-defined attractions are scattered across different regions, they are more original than when they are concentrated. Combine the proportion of customized attractions and the degree of geographical dispersion to generate a spatial originality parameter to reflect the originality and exploratory nature of the user's itinerary.

[0083] S125. Combine the real-time traffic prediction level and pedestrian flow in the environmental context data, dynamically weight the behavioral characteristics through the gating mechanism, and generate an initial degree of freedom index;

[0084] Collect holiday traffic levels and current passenger flow data in real time to generate dynamic weight coefficients. For example, during peak traffic hours, reduce the user's freedom weight to avoid recommending congested attractions. Combined with the overlap ratio between traffic control areas and user travel routes, set control impact factors in different grades. Through the set gating function, dynamically adjust the contribution ratio of behavioral characteristics based on the user's historical freedom mean and environmental factors including traffic and control. For example, in a high-risk environment, the weight of the user's dynamic frequency is suppressed. The dynamic frequency, semantic adjustment depth, and spatial originality are fused according to the weights after environmental adaptation to generate an initial freedom index that reflects the user's editable permissions in the current environment.

[0085] S126. Dynamically grade the initial degree of freedom index, perform forced downgrade operations when sudden risks are detected, and output five levels of travel degree of freedom index.

[0086] The initial degree of freedom index is input into the deep reinforcement learning model, and combined with the stability of user historical behavior such as modification frequency fluctuations and environmental risk levels, it is divided into 1-5 levels of freedom. The model optimizes the classification strategy through user acceptance rate and itinerary safety risk value. For example, a high acceptance rate increases the degree of freedom level, and high-risk events trigger downgrades. Real-time monitoring of sudden risks such as geological activities, facility failures, or excessive passenger flow. When the risk level reaches the threshold, the degree of freedom level is forced to be lowered. For example, if a heavy rain warning (risk level 3) is detected, the original level 4 is downgraded to level 2. After downgrading, it automatically switches to the corresponding collaborative mode (such as switching to the official dominant mode), locks the editing permissions of high-risk attractions, and quickly generates safe alternative routes through the path optimization algorithm.

[0087] Specifically, the preset conditions in this step include the following:

[0088] 1. When the user's interest matching degree exceeds the preset first matching degree threshold and the degree of freedom rating reaches the highest level of the preset evaluation system, the user-led mode is activated;

[0089] The first match threshold is a user interest match greater than 0.8 (range 0-1), calculated by fusing user behavior data with the semantic similarity of official attractions. The threshold is recalibrated quarterly based on changes in user group behavior. Itinerary freedom is rated on a five-level scale, based on user modification frequency, originality, and environmental constraints. The highest level is ≥50% user-defined attractions with an environmental correction value greater than 0.9.

[0090] 2. When the user interest matching degree is lower than the preset second matching degree threshold, and the official scenic spot score exceeds the official scenic spot score benchmark value and the theme relevance meets the preset relevance standard, the official dominant mode is triggered;

[0091] The second matching threshold is that the user interest matching degree is lower than 0.3, indicating that the user interest is vague or deviates greatly from the official recommendation. Judgment is made based on the sparse user behavior data (such as the length of stay is less than 5 minutes / attraction) or the score is significantly conflicting with the official data (such as the user score is less than 5 points, the official score is greater than 9 points). The official rating benchmark value of the scenic spot is the official minimum safety and quality score set by the scenic spot, such as a safety score ≥8 points and a comprehensive score ≥9 points. Attractions that do not meet the standards are prohibited from being recommended. The theme relevance standard is that the correlation between the official route and the user's itinerary theme keywords is ≥0.85. For example, if the user's itinerary contains "parent-child" and "entertainment", the official "family entertainment" theme route with a correlation of 0.92 is matched.

[0092] 3. When the user's interest matching degree is between the first matching degree threshold and the second matching degree threshold, and the spatial distribution overlap between the user's planned attractions and the officially recommended attractions reaches a preset overlap ratio, the balanced collaboration mode is activated;

[0093] The range between the first and second match thresholds indicates a user interest match between 0.4 and 0.7, indicating a partial match but room for improvement. This is suitable for users with clear preferences who need to incorporate official resources to optimize their routes. The spatial overlap threshold is ≥60% overlap between the geographic distribution of the user's attractions and the official recommended attractions. GIS is used to calculate the percentage of overlapping bounding boxes, map the user's itinerary and the official route to a grid coordinate system, and calculate the percentage of overlapping grid cells.

[0094] 4. When it is detected that the user's itinerary contains multiple different theme keywords that meet the preset number conditions, and there are recommended routes in the official theme library that meet the preset association conditions with the theme keywords, the theme guidance mode is activated;

[0095] The number of topic keywords required is that the user's itinerary must contain ≥3 keywords with the same topic (e.g., "history," "culture," and "historical sites"), which are extracted and clustered using TF-IDF. The topic relevance requirement is that the official theme route must meet the following two conditions: (1) semantic relevance with the user's keyword ≥ 0.8; (2) route coherence score ≥ 0.9.

[0096] 5. When environmental monitoring data triggers the scenic area safety warning standard or real-time passenger flow data exceeds the scenic area carrying capacity warning threshold, switch to real-time emergency mode.

[0097] The safety warning standard is a scenic area safety level ≥ 2 (levels 1-5). Trigger conditions include geological risks, equipment failure, or meteorological disasters. The scenic area's capacity warning threshold is a real-time visitor flow ≥ 80% of the capacity. This threshold is calculated in real time using camera and gate data. The threshold is raised to 90% on weekends and holidays to prevent frequent false triggers.

[0098] By clearly defining multi-dimensional thresholds and their dynamic management rules, the above steps enable the system to accurately identify user needs, environmental risks, and resource constraints, achieve adaptive switching among five collaborative modes, and effectively solve the problems of policy rigidity, data fragmentation, and delayed security response in traditional recommendation systems.

[0099] S13. Adopt corresponding recommendation strategies based on the collaborative model to adjust the user's self-created itinerary, including:

[0100] S131. In the user-driven mode, the user's core attraction sequence is retained, and official attractions are recommended based on the user's interest matching degree;

[0101] 1. Based on the dwell time distribution, visit frequency statistics, and user-initiated annotation information in the user's historical behavior data, extract the set of attractions that meet the preset dwell time threshold, visit frequency threshold, or core marker identification, and construct an unchangeable core attraction sequence;

[0102] Core attractions are identified based on user duration, visit frequency, or active tagging. A duration threshold is used to filter out low-interest points, ensuring that core attractions represent rigid user demand. For example, if a user's duration is 30 minutes or longer, a sequence of core attractions that meet this requirement is extracted.

[0103] 2. Based on the user's interest matching, a set of candidate attractions with a textual semantic similarity with the core attraction sequence exceeding a preset semantic threshold is screened from the official attraction database. The candidate attractions are then ranked by collaborative filtering based on the user's historical preference tags.

[0104] We screen candidate sets from official attractions whose text description similarity with core attractions is greater than 0.7, such as "Bronze Drum Culture" matching "Zhuang Festival Exhibition" and "Guangxi National Museum" matching "Qingxiu Mountain". We also combine user historical preference tags, such as "Parent-Child Travel", to prioritize family-friendly attractions to ensure theme consistency.

[0105] 3. Based on the real-time spatial distribution characteristics of the environmental context data, candidate attractions are screened for geographic proximity, retaining official attractions whose path distance to at least one attraction in the core attraction sequence is less than a preset proximity threshold;

[0106] Only official attractions with a distance of less than 5 kilometers or a commuting time of less than 30 minutes from core attractions will be retained. Avoid recommending geographically dispersed attractions to ensure the continuity of the itinerary.

[0107] 4. Insert the selected official attractions into the idle time of non-core attractions in the user's itinerary in descending order of interest matching, and adjust the recommendation strategy based on user feedback behavior.

[0108] The candidate attractions are sorted by matching degree and inserted into the idle time period. For example, if the travel gap is greater than 1 hour, the semantic weight will be increased after the user accepts the recommendation, and the proximity will be prioritized after the user rejects it, thus achieving adaptive adjustment of the strategy.

[0109] In the user-driven mode, the system locks on core attractions through rigid standards, and recommends highly relevant official attractions based on semantics, preferences and location constraints, intelligently optimizing the quality of the itinerary while retaining user autonomy.

[0110] S132. In the official-led mode, insert the highest priority official attractions based on the attraction scores and theme relevance in the multi-dimensional evaluation data;

[0111] 1. Generate priority weights based on the real-time dynamic scores, historical comprehensive scores, and topic relevance scores in the constructed multi-dimensional evaluation data;

[0112] Based on the real-time dynamic rating of official attractions (such as the reputation of the day ≥ 8 points), historical comprehensive rating (such as annual average ≥ 9 points) and theme relevance rating (such as similarity with user itinerary keywords ≥ 0.8), weights are dynamically allocated (such as dynamic rating 50%, historical rating 30%, theme relevance 20%) to ensure that high-security and strong-relevance attractions are given priority.

[0113] 2. Screen candidate official attractions that meet the preset dynamic score threshold and theme relevance threshold, and generate a ranked list in descending order of priority weight;

[0114] Only official attractions with a dynamic score of 8 or higher and a thematic relevance of 0.8 or higher will be retained, sorted by priority weight. For example, an attraction with a dynamic score of 9 and a thematic relevance of 0.85 will have a higher priority than an attraction with a score of 8.5 and a thematic relevance of 0.9.

[0115] 3. Based on the spatial and temporal distribution gaps of non-core attractions in user-created itineraries, the highest-priority official attractions are inserted into idle time slots that meet the following conditions: the commuting time between the inserted attraction and the adjacent user attractions is less than a preset time threshold, and the total time deviation after the insertion does not exceed a preset time tolerance range;

[0116] The time-space constraint insertion method requires that the highest-priority attraction be inserted during an idle period in the user's itinerary (e.g., the interval between two attractions is greater than one hour). The commuting time to the adjacent attraction must be less than 30 minutes (time threshold), and the total travel time deviation must be less than 15% (tolerance range). For example, after inserting the "Zhuang Festival Exhibition," the commuting time is 25 minutes, extending the total travel time by 12%, meeting the requirements.

[0117] 4. Update the priority evaluation model according to the real-time feedback behavior of users on the inserted attractions.

[0118] If the user accepts the recommendation, the theme relevance weight of the attraction will be increased; if the user rejects it, the dynamic score weight will be increased and the alternative attractions will be re-matched to ensure real-time optimization of the recommendation strategy.

[0119] Under the official-led mode, the system prioritizes inserting attractions with high scores and strong thematic relevance, strictly constrains the rationality of time and space, and dynamically adjusts the recommendation strategy through feedback to ensure the safety and professionalism of the itinerary.

[0120] S133. Dynamically allocate weights between user-created attractions and official attractions in a balanced collaborative mode;

[0121] 1. Generate an initial weight distribution ratio based on the user's interest matching and the theme relevance score in the official attractions' multi-dimensional evaluation data;

[0122] An initial weight ratio (e.g., 7:3) is generated based on the user's interest matching degree of 0.7 and the official attraction's theme relevance score of 0.8. The theme relevance must be greater than 0.6 to be included in the calculation to ensure that the quality of the official attractions meets the standards.

[0123] 2. Adjust the weight ratio based on the travel time redundancy and spatial distribution dispersion in the environmental context data. The adjustment includes: when the time redundancy exceeds the preset redundancy threshold, the weight of user-created attractions is increased by linear interpolation; when the spatial dispersion exceeds the preset dispersion threshold, the weight of official attractions is increased based on the path continuity optimization algorithm;

[0124] Time redundancy adjustment: If the idle time of a user's trip accounts for more than 20% (redundancy threshold), the user weight will be increased proportionally to 8:2, allowing more self-created attractions to fill the idle time.

[0125] Spatial discrete optimization: If the average distance between attractions is greater than 5 kilometers (discrete threshold), the path planning algorithm is called to optimize the route, and the official weight is increased to 5:5 to improve the continuity of the trip.

[0126] 3. Sort the user-created attractions and official attractions in a mixed order based on the adjusted weight ratio to generate a set of candidate itinerary plans that meet the weight constraints.

[0127] Mixed sorting is performed according to the adjusted weights (e.g. 6:4), with user attractions retaining their originality and official attractions supplemented with highly relevant locations, generating multiple sets of candidate options for users to choose from.

[0128] In the balanced collaborative mode, a dynamic weight allocation mechanism is used to flexibly adjust between user creativity and official resources, ensuring personalized itineraries while optimizing time utilization and route rationality.

[0129] S134. Generate a coherent route by combining the thematic relevance of the multidimensional evaluation data in the theme-guided mode;

[0130] 1. Model the topic probability distribution of the attraction description text in the user's self-created itinerary, extract the latent topic vector and the inter-topic transition probability matrix;

[0131] Perform LDA topic modeling on the descriptions of scenic spots in the user's itinerary (such as "singing festival culture" and "hydrangea making"), extract the topic distribution vector, such as history: 0.6, culture: 0.3, art: 0.1, calculate the probability of switching themes of adjacent scenic spots, such as the probability of "history switching to culture" is 0.8, and construct a transfer matrix.

[0132] 2. Perform similarity matching between the potential topic vector and the feature vectors of predefined topic routes in the official topic library, and select candidate topic routes with a relevance score higher than the preset matching threshold;

[0133] Perform cosine similarity matching between the user topic vector and the official topic library (such as "March 3rd"), and screen candidate routes with similarity ≥ 0.8 to ensure strong topic relevance.

[0134] 3. Calculate the logical correlation between attractions based on the inter-theme transition probability matrix and the spatiotemporal commuting time of the candidate theme routes;

[0135] Taking into account the topic transfer probability (e.g., 0.8) and commuting time (e.g., 30 minutes), routes with a correlation greater than 0.75 were selected to ensure that the topic was coherent and the time was reasonable.

[0136] 4. When the degree of relevance exceeds the preset coherence threshold, an enhanced route is generated that includes temporal and spatial order and thematic narrative logic, and is embedded with the guide and commentary data in the official thematic library.

[0137] Sorted by timeline (such as "ancient → modern → contemporary") and narrative logic (such as "Zhuang cultural context"), embedded with official guide data (such as AI singing system, light and shadow show), an immersive theme route is generated.

[0138] In the theme guidance mode, the system connects scattered attractions into a logically complete thematic narrative route through semantic analysis and time and space optimization, combines the guide data to enhance the cultural experience, and solves the problem of isolated attractions in traditional recommendations.

[0139] S135. Replace risky attractions based on the safety status of the multi-dimensional assessment data in real-time emergency mode;

[0140] 1. IoT sensors collect real-time data on geological activity intensity, facility failure signals, and heat maps of crowd density at scenic spots to calculate a dynamic safety risk index.

[0141] Geological sensors monitor seismic wave intensity (e.g., ≥ 4), facility sensors detect fault signals, and cameras measure crowd density. A real-time risk index is calculated using weights (50% for geology, 30% for facilities, and 20% for crowds). An emergency response is triggered when the index is ≥ 0.7 (safety threshold) and persists for 10 minutes (monitoring window).

[0142] 2. When the risk index continues to exceed the preset safety threshold for the monitoring window period, the official safe attraction database will be used to select alternative attractions that meet the requirements. The requirements include: the theme relevance score of the risk attraction is higher than the preset emergency relevance threshold, and the commuting time from the user's current location is less than the preset emergency time threshold;

[0143] Select alternative attractions from the official safety database with a theme relevance greater than 0.75 and a commuting time less than 20 minutes, giving priority to high-quality attractions with a score of ≥9 points.

[0144] 3. Based on the tabu search algorithm and dynamic time warping model, the optimal path sequence of alternative attractions is generated.

[0145] Use the tabu search algorithm to generate the shortest path to alternative attractions, such as avoiding congested sections. Combined with the DTW model, align the original travel timeline. For example, if the original travel time is 8 hours, the deviation after replacement is ≤12%, ensuring that the user's travel rhythm is not significantly affected.

[0146] In real-time emergency mode, the system accurately identifies risks through multi-source data fusion, quickly replaces highly relevant safety attractions, and uses intelligent algorithms to ensure the rationality of routes and time, minimizing the interference of sudden risks on the itinerary.

[0147] S14. Generate an optimized itinerary plan and display it visually, updating user data and official data sources.

[0148] The system integrates optimized itineraries into a spatially and temporally ordered sequence, annotating attraction locations, commuting routes, and timelines through a map visualization interface, and embedding official guide data. After users confirm their plans, their interactions—such as accepting or rejecting recommendations or adjusting their stay duration—are synchronized and updated in the user profile database, which is then used to iterate the recommendation model. Finally, the system feeds environmental risk data from itineraries back to official data sources, dynamically updating attraction safety scores and thematic relevance, creating a closed-loop data ecosystem.

[0149] Example 2

[0150] An embodiment of a collaborative recommendation system for tourist-created itineraries and official selected itineraries in the present invention includes the following steps:

[0151] A data acquisition unit, used to obtain behavioral data of user-created itineraries, multi-dimensional evaluation data of official selected itineraries, and environmental context data;

[0152] The collaboration mode determination unit is used to calculate the user interest matching degree based on behavioral data, generate the itinerary freedom index based on the modification frequency, and determine the collaboration mode through preset condition analysis based on the environmental context data. The collaboration modes include user-led mode, official-led mode, balanced collaboration mode, theme-guided mode, and real-time emergency mode.

[0153] The itinerary adjustment unit is used to adjust the user's self-created itinerary using the corresponding recommendation strategy according to the collaborative mode, including: retaining the user's core attraction sequence in the user-led mode and recommending official attractions based on the user's interest matching; inserting the highest priority official attractions based on the attraction scores and theme relevance in the official-led mode; dynamically allocating the weights of user-created attractions and official attractions in the balanced collaborative mode; generating a coherent route in combination with the theme relevance of the multi-dimensional evaluation data in the theme-guided mode; and replacing risky attractions based on the safety status of the multi-dimensional evaluation data in the real-time emergency mode.

[0154] The optimized itinerary plan generation unit is used to generate the optimized itinerary plan and display it visually, and update user data and official data sources.

[0155] The specific functions and uses of the units in this embodiment are similar to the steps in the aforementioned embodiment 1 and are not described in detail here.

[0156] It is understandable that those skilled in the art can, under the guidance of the above embodiments, combine various implementation methods in the above embodiments to obtain technical solutions of multiple implementation methods.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A collaborative recommendation method for tourist-created itineraries and official selected itineraries, characterized in that: include: Obtain behavioral data on user-created itineraries, multi-dimensional evaluation data on official selected itineraries, and environmental context data; Calculate the user interest matching degree based on the behavior data, generate a travel freedom index based on the modification frequency, and determine the collaborative mode through preset condition analysis in combination with the environmental context data. The collaborative modes include user-led mode, official-led mode, balanced collaborative mode, theme-guided mode, and real-time emergency mode. According to the collaborative model, a corresponding recommendation strategy is used to adjust the user's self-created itinerary, including: In the user-driven mode, the user's core attraction sequence is retained, and official attractions are recommended based on the user's interest matching degree; In the official-led mode, the highest priority official attractions are inserted based on the attraction scores and thematic relevance in the multi-dimensional evaluation data; Dynamically allocate weights between user-created attractions and official attractions in a balanced collaborative mode; In the topic-guided mode, a coherent route is generated by combining the topic relevance of the multi-dimensional assessment data; Replace risky scenic spots based on the safety status of the multi-dimensional assessment data in real-time emergency mode; Generate optimized itinerary plans and present them visually, updating user data and official data sources.

2. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 1, characterized in that: The calculation of user interest matching based on the behavior data, the generation of a travel freedom index based on the modification frequency, and the determination of a collaborative mode based on preset condition analysis in combination with the environmental context data include: Perform word segmentation on the search records, favorite tags, and official attraction description texts in the user's historical behavior data to generate a joint corpus, which includes a user behavior corpus sub-corpus and an official attraction corpus sub-corpus; Based on the joint corpus, an improved TF-IDF algorithm is used to generate user interest vectors and official attraction vectors respectively, and the user historical behavior data is weighted by a preset time decay factor to generate a weighted user interest vector; The cosine similarity between the weighted user interest vector and the official scenic spot vector is calculated, and the calculated cosine similarity is integrated with the proportion of stay time and the score deviation in the user behavior characteristics to output an interest matching score.

3. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 2, characterized in that: The calculation of user interest matching based on the behavior data, the generation of a travel freedom index based on the modification frequency, and the determination of a collaborative mode based on preset condition analysis in combination with the environmental context data further include: Extract the dynamic frequency, semantic adjustment depth, and spatial originality characteristics of user itinerary modification behavior. The semantic adjustment depth is quantified by the change ratio of attraction types and the increase or decrease of theme keywords. The spatial originality is calculated based on the geographical distribution dispersion of attractions. Combined with the real-time traffic prediction level and pedestrian flow in the environmental context data, the behavioral characteristics are dynamically weighted and fused through a gating mechanism to generate an initial degree of freedom index; The initial degree of freedom index is dynamically graded, and when an unexpected risk is detected, a forced downgrade operation is performed to output five levels of travel degree of freedom index.

4. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to any one of claims 1 to 3, characterized in that: The preset conditions include: When the user interest matching degree is higher than the preset first matching degree threshold and the freedom rating reaches the highest level of the preset evaluation system, the user-led mode is activated; When the user interest matching degree is lower than the preset second matching degree threshold, and the official scenic spot score exceeds the official scenic spot score benchmark value and the theme relevance meets the preset relevance standard, the official dominant mode is triggered; When the user interest matching degree is between the first matching degree threshold and the second matching degree threshold, and the spatial distribution overlap between the user's planned attractions and the officially recommended attractions reaches a preset overlap ratio, the balanced collaboration mode is enabled; When it is detected that the user's itinerary contains multiple different theme keywords that meet a preset number of conditions, and there are recommended routes in the official theme library that meet the preset association conditions with the theme keywords, the theme guidance mode is activated; When environmental monitoring data triggers the scenic area safety warning standard or the real-time passenger flow data exceeds the scenic area carrying capacity warning threshold, it switches to real-time emergency mode.

5. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 1, characterized in that: In the user-driven mode, the user's core attraction sequence is retained, and official attractions are recommended based on the user's interest matching, including: Based on the dwell time distribution, visit frequency statistics and user-initiated annotation information in the user's historical behavior data, a set of scenic spots that meet the preset dwell time threshold, visit frequency threshold or core mark identification is extracted to construct an unchangeable core scenic spot sequence; Based on the user interest matching degree, a set of candidate attractions whose textual semantic similarity with the core attraction sequence is higher than a preset semantic threshold is selected from the official attraction database, and collaborative filtering and sorting is performed in combination with the user's historical preference tags; Based on the real-time spatial distribution characteristics of the environmental context data, candidate attractions are screened for geographic proximity, and official attractions whose path distance to at least one attraction in the core attraction sequence is less than a preset proximity threshold are retained; The filtered official attractions are inserted into the idle time of non-core attractions in the user's itinerary in descending order of interest matching, and the recommendation strategy is adjusted based on user feedback behavior.

6. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 1, characterized in that: The official-led mode inserts the highest priority official attractions based on the attraction scores and theme relevance in the multi-dimensional evaluation data, including: Generate priority weights based on real-time dynamic scores, historical comprehensive scores, and topic relevance scores in the constructed multi-dimensional evaluation data; Screen candidate official attractions that meet the preset dynamic score threshold and theme relevance threshold, and generate a ranked list in descending order of priority weight; Based on the time and space distribution gaps of non-core attractions in the user-created itinerary, the highest priority official attractions are inserted into the idle time period that meets the following conditions: the commuting time between the inserted attraction and the adjacent user attractions is less than a preset time threshold, and the total time deviation of the inserted itinerary does not exceed a preset time tolerance range; The priority evaluation model is updated according to the real-time feedback behavior of the user on the inserted scenic spots.

7. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 1, characterized in that: The dynamic allocation of weights between user-created attractions and official attractions in the balanced collaboration mode includes: Generate an initial weight distribution ratio based on the user's interest matching and the theme relevance score in the official attractions' multi-dimensional evaluation data; The weight ratio is adjusted based on the travel time redundancy and spatial distribution dispersion in the environmental context data. The adjustment includes: when the time redundancy is higher than the preset redundancy threshold, the weight of the user-created attractions is increased by linear interpolation; when the spatial dispersion exceeds the preset dispersion threshold, the weight of the official attractions is increased based on the path continuity optimization algorithm; According to the adjusted weight ratio, user-created attractions and official attractions are mixed and sorted to generate a set of candidate itinerary plans that meet the weight constraints.

8. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 1, characterized in that: In the topic-guided mode, a coherent route is generated by combining the topic relevance of the multi-dimensional evaluation data, including: Model the topic probability distribution of the attraction description text in the user's self-created itinerary, extract the latent topic vector and the inter-topic transition probability matrix; Perform similarity matching between the potential topic vector and the feature vectors of predefined topic routes in the official topic library, and select candidate topic routes with a relevance score higher than a preset matching threshold; Calculating the inter-attraction logical correlation based on the inter-theme transition probability matrix and the spatiotemporal commuting time of the candidate theme routes; When the correlation exceeds the preset coherence threshold, an enhanced route containing temporal and spatial sequence and thematic narrative logic is generated, and the guide commentary data in the official thematic library is embedded.

9. The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to claim 1, characterized in that: Replacing risky scenic spots based on the safety status of the multi-dimensional assessment data in the real-time emergency mode includes: IoT sensors collect real-time data on geological activity intensity, facility failure signals, and heat maps of crowd density at scenic spots to calculate dynamic safety risk indexes. When the risk index continues to exceed the preset safety threshold for a monitoring window period, the official safe scenic spot database is screened for candidate scenic spots that meet the conditions. The conditions include: the theme relevance score with the risky scenic spot is higher than the preset emergency relevance threshold, and the commuting time from the user's current location is less than the preset emergency time threshold; Based on the tabu search algorithm and dynamic time warping model, the optimal path sequence of alternative attractions is generated.

10. A collaborative recommendation system for tourist-created itineraries and official selected itineraries, characterized in that: The collaborative recommendation method for tourist-created itineraries and official selected itineraries according to any one of claims 1 to 9 comprises: A data acquisition unit, used to obtain behavioral data of user-created itineraries, multi-dimensional evaluation data of official selected itineraries, and environmental context data; a collaborative mode determination unit, configured to calculate a user interest matching degree based on the behavioral data, generate a travel freedom index based on the modification frequency, and determine a collaborative mode based on preset conditions in combination with the environmental context data. The collaborative modes include user-led mode, official-led mode, balanced collaborative mode, theme-guided mode, and real-time emergency mode; An itinerary adjustment unit is configured to adjust the user's self-created itinerary using a corresponding recommendation strategy according to the collaborative mode, including: retaining the user's core attraction sequence in the user-led mode and recommending official attractions based on the user's interest matching degree; inserting the highest priority official attractions based on the attraction scores and theme relevance in the multi-dimensional evaluation data in the official-led mode; dynamically allocating the weights of user-created attractions and official attractions in the balanced collaborative mode; generating a coherent route based on the theme relevance of the multi-dimensional evaluation data in the theme-guided mode; and replacing risky attractions based on the safety status of the multi-dimensional evaluation data in the real-time emergency mode. The optimized itinerary plan generation unit is used to generate the optimized itinerary plan and display it visually, and update user data and official data sources.

Images