Multi-dimensional dynamic price comparison and personalized recommendation method and system for cruise ship products
By constructing multi-dimensional price comparison feature vectors and adaptively generating weights, combined with an LSTM price prediction model, the information silo effect of cruise products was solved, personalized recommendations and resource optimization were achieved, and user satisfaction was improved.
Patent Information
- Application Number
- CN202511063845.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-31
- Publication Date
- 2025-11-14
AI Technical Summary
Existing cruise product information systems suffer from information silos, making it difficult for users to achieve multi-dimensional dynamic price comparisons and personalized recommendations. This results in the inability to obtain dynamic discount information in real time, the inability to quantify service added value, a disconnect between recommendation results and user needs, and the failure to incorporate price fluctuation signals into the price comparison system, leading to missed opportunities to purchase tickets.
By constructing a multi-dimensional price comparison feature vector Vcomp, combined with an LSTM price prediction model and adaptive weight generation, a personalized recommendation ranking is generated. The discount rate and weight vector are updated through a closed-loop collaborative optimization mechanism to achieve multi-dimensional dynamic price comparison and personalized recommendation of cruise products.
It enables multi-dimensional dynamic price comparison and personalized recommendations for cruise products, breaks down data silos, enables scenario-aware decision-making, improves user satisfaction, optimizes resource allocation, and meets user needs.
Smart Images
Figure CN120952173A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of product recommendation technology, specifically to a method and system for multi-dimensional dynamic price comparison and personalized recommendation of cruise products. Background Technology
[0002] As a complex tourism product, the cruise industry involves multi-source heterogeneous data on itinerary planning, including voyage days, port sequence, cabin pricing models, and entertainment packages. Currently, the decentralized information systems independently built by over 400 operators globally create a severe information silo effect. Because operators use unstructured HTML pages and private API interfaces to publish itinerary information, and lack unified data parsing standards, consumers are forced to manually search and compare prices across platforms. This not only makes it difficult to obtain real-time dynamic discount information (such as limited-time cabin upgrade offers), but also makes it impossible to quantify and compare the added value of key services (such as catering standards and entertainment facility density). This data fragmentation directly leads to three major decision-making dilemmas: First, basic voyage parameters (such as number of days and ports) and service resource data remain disconnected, making it impossible for users to assess the rationality of high-premium services; second, static weighted recommendation models ignore scenario-based needs, causing recommendation results to be out of touch with actual decision-making scenarios; and third, price fluctuation signals are not incorporated into the price comparison system, causing users to miss the optimal ticketing time. Summary of the Invention
[0003] One of the objectives of this invention is to provide a method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products, which can realize multi-dimensional dynamic price comparison and personalized recommendation of cruise products, meet user needs, and improve user satisfaction.
[0004] To achieve the above objectives, a method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products is provided, including the following steps:
[0005] Multi-dimensional price comparison feature construction steps: Obtain the basic dimensional parameters of cruise products, including cruise duration (D), number of ports of call (P), scenario tag (L), user preference tag (F / S), and real-time price (P). now User preference tags F / S include family and parent-child (F) and senior citizen (S); service added value dimension is calculated based on flight range basic dimension parameters, and the service added value dimension includes catering density index R. d and entertainment facility density index E d ; Obtain the discount rate δt output by the LSTM price prediction model as the time dimension object; Generate a price comparison feature vector V by combining the basic distance dimension parameters, the service added value dimension, and the time dimension object representing the price trend. comp =(D,P,P) now ,Rd,Ed,δt,F,S);
[0006] The adaptive weight generation steps are as follows: A static baseline weight vector W0 is generated based on the scene label L, and a convergent weight vector W is generated using an iterative formula with session behavior data as input. K The scenario tag L includes family and parent-child activities (F) and elderly care (S); the scenario tag L is optimized through historical behavior data; the session behavior data includes search term frequency and page dwell time.
[0007] Recommendation decision generation steps: Linearly combine the price comparison feature vector and the weight vector to output a personalized recommendation ranking;
[0008] Closed-loop collaborative optimization steps: Real-time collection of user feedback data regarding personalized recommendations and rankings of cruise products, including the recommendation conversion rate C; analysis of the cruise product recommendation conversion rate C; when the recommendation conversion rate C... <C thres At that time, the feedback data is used as historical behavior data to update the discount rate δt and the calibration weight vector W. k The optimized δt is then compared with the calibration weight vector W. k Feedback is fed back to the multi-dimensional price comparison feature construction step and the weight adaptive generation step for the next round of decision-making.
[0009] Furthermore, the catering density index R d The calculation formula is as follows:
[0010] R d =(∑(x*t)) / n*Φ1(D,P),Φ1(D,P)=D / 7*(1-0.08P)
[0011] Where x is the restaurant rating, t is the opening hours, n is the passenger capacity, and Φ1(D,P) is the dynamic adjustment function.
[0012] Furthermore, the density index E of the entertainment facilities d The calculation formula is as follows:
[0013]
[0014] Where y is the number of facilities, m is the predicted usage frequency, n is the passenger capacity, and Φ2(D,P) is the dynamic adjustment function.
[0015] Furthermore, the generation of the scene label L and user preference label F / S includes the following steps:
[0016] Hierarchical tag generation steps: Using users' historical order data and real-time search keywords as input, execute a classification algorithm based on random forest to output discrete scene tags L∈{family and parenting (F), elderly care (S), business meetings (B)};
[0017] Using real-time page clickstream data and facility dwell time as input, a time-decay weighted calculation is performed: F = ∑(parent-child related clicks × e) -0.1t ), output continuous user preference labels F∈[0,1],S∈[0,1].
[0018] Furthermore, in the weight adaptive generation step, a static reference vector is generated using the scene label L as input:
[0019]
[0020] in,
[0021]
[0022] PriceSens∈[0,10] represents the user's price sensitivity score;
[0023]
[0024] include and
[0025]
[0026] Furthermore, the calculation formula for the iterative formula in the adaptive weight generation step is as follows:
[0027]
[0028] Where k = 1, 2, 3…m, the cycle continues until ||W k -W k-1 ||<ε,K i For search term frequency, T i K represents the page dwell time; α and β are adaptive coefficients based on user identity type. max T is the cardinality for normalized search term frequencies. max γ is the normalized base for dwell time; F Parent-child preference moderating coefficient, γ S This is the adjustment coefficient for silver-haired healthcare.
[0029] Furthermore, the recommendation decision generation step includes the following steps:
[0030] The price comparison feature vector and weight vector of cruise products are linearly combined, and the specific calculation formula is as follows:
[0031] S user =V comp *W K , among which, S user For personalized scoring;
[0032] Personalized recommendations are generated and ranked based on the individual scores of different cruise products, and the personalized recommendation ranking is output.
[0033] The second objective of this invention is to provide a multi-dimensional dynamic price comparison and personalized recommendation system for cruise products. This system utilizes the multi-dimensional dynamic price comparison and personalized recommendation method for cruise products described in any one of claims 1-7, and specifically includes the following modules:
[0034] Multi-dimensional price comparison feature construction module: used to obtain the basic dimension parameters of cruise products and historical behavior data. The basic dimension parameters of cruise products include cruise days D, number of ports of call P, scenario label L, user preference label F / S, and real-time price P. now User preference tags F / S include family and parent-child (F) and senior citizen (S); service added value dimension is calculated based on flight range basic dimension parameters, and the service added value dimension includes catering density index R. d and entertainment facility density index E d ; Obtain the discount rate δt output by the LSTM price prediction model as the time dimension object; Generate a price comparison feature vector V by combining the basic distance dimension parameters, the service added value dimension, and the time dimension object representing the price trend. comp =(D,P,P) now ,Rd,Ed,δt,F,S);
[0035] The adaptive weight generation module generates a static baseline weight vector W0 based on the scene label L, and uses session behavior data as input to generate a convergent weight vector W through an iterative formula. K The scenario tag L includes family and parent-child activities (F) and elderly care (S); the scenario tag L is optimized through historical behavior data; the session behavior data includes search term frequency and page dwell time.
[0036] Recommendation decision generation module: used to linearly combine the price comparison feature vector and the weight vector to output personalized recommendation ranking;
[0037] Closed-loop collaborative optimization module: used to collect feedback data on user feedback regarding personalized recommendations and rankings of cruise products in real time, including the recommendation conversion rate C; analyzes the recommendation conversion rate C of cruise products, and when the recommendation conversion rate C... <C thres At that time, the feedback data is used as historical behavior data to update the discount rate δt and the calibration weight vector W. k The optimized δt is then compared with the calibration weight vector W. k Feedback is fed back to the multi-dimensional price comparison feature construction step and the weight adaptive generation step for the next round of decision-making.
[0038] Principles and advantages:
[0039] 1. Break down data barriers: Integrate basic dimensions of flight range (D, P, Pnow), service added value (Rd, Ed), price trend variables (δt), and user preference tags (F / S) into a unified and comparable feature vector Vcomp, transforming traditionally unquantifiable unstructured services such as "exclusive privileges" into standardized decision parameters.
[0040] 2. Achieve scene-aware decision-making: Initialize the baseline weight W0 based on the scene tag L (family parenting / elderly care), and dynamically generate the convergent weight Wk through conversation behavior data (search term frequency, page dwell time), so that the weight allocation adaptively matches the user's real-time decision intention.
[0041] 3. Constructing a self-evolving closed loop: When the recommendation conversion rate falls below a threshold, the discount rate δt is automatically updated and the weight vector Wk is calibrated. The optimization results are then fed back to the price comparison feature construction and weight generation module, forming a "decision-feedback-optimization" technology enhancement loop. This ensures the best recommendation ranking for cruise products, thereby meeting user needs and improving user satisfaction.
[0042] 4. Driving industry value upgrading: Significantly reducing the decision-making burden of highly complex cruise products for users, and enabling operators to accurately identify service shortcomings (such as insufficient entertainment facilities on family cruises or lack of medical facilities on senior cruises) by relying on closed-loop data, thereby optimizing resource allocation efficiency and service innovation direction. Attached Figure Description
[0043] Figure 1 This is a flowchart illustrating a method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to an embodiment of the present invention.
[0044] Figure 2 This is a diagram of the LSTM price prediction model architecture. Detailed Implementation
[0045] The following detailed description illustrates the specific implementation method:
[0046] Example
[0047] A method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products, basically as follows: Figure 1 As shown, it includes the following steps:
[0048] Multi-dimensional price comparison feature construction steps: Obtain the basic dimensional parameters of the cruise product's itinerary and historical behavioral data. The historical behavioral data is extracted through analysis of user historical order data, including user behavioral choices and feedback made during order processing. The basic dimensional parameters of the itinerary include the number of cruise days (D), the number of ports of call (P), the scenario label (L), the user preference label (F / S), and the real-time price (P). nowUser preference tags F / S include family and parent-child (F) and senior citizen (S); service added value dimension is calculated based on flight range basic dimension parameters, and the service added value dimension includes catering density index R. d and entertainment facility density index E d The discount rate δt output by the LSTM price prediction model is obtained as the time dimension object; the historical behavior data can be used to train and optimize the LSTM price prediction model; for example, by analyzing user discount sensitivity records from historical behavior data and combining price series and historical cabin booking rates for each time period, the LSTM price prediction model can be optimized.
[0049] The basic parameters of flight distance, the service added value dimension, and the time dimension representing price trends are used to generate a price comparison feature vector V. comp =(D,P,P) now ,Rd,Ed,δt,F,S);
[0050] The catering density index R d The calculation formula is as follows:
[0051] R d =(∑(x*t)) / n*Φ1(D,P),Φ1(D,P)=D / 7*(1-0.08P)
[0052] Where x is the restaurant rating, t is the opening hours, n is the passenger capacity, and Φ1(D,P) is the dynamic adjustment function.
[0053] The density index of entertainment facilities E d The calculation formula is as follows:
[0054]
[0055] Where y is the number of facilities, m is the predicted usage frequency, n is the passenger capacity, and Φ2(D,P) is the dynamic adjustment function.
[0056] The generation of the scene label L and user preference label F / S includes the following hierarchical label generation steps:
[0057] Hierarchical tag generation steps: Using users' historical order data and real-time search keywords as input, execute a classification algorithm based on random forest to output discrete scene tags L∈{family and parenting (F), elderly care (S), business meetings (B)};
[0058] Using real-time page clickstream data and facility dwell time as input, a time-decay weighted calculation is performed: F = ∑(parent-child related clicks × e) -0.1t ), output continuous user preference labels F∈[0,1],S∈[0,1].
[0059] like Figure 2 As shown, the core structure of the LSTM price prediction model includes:
[0060] Input layer: Receives a four-dimensional temporal feature vector X t =(P hist O rate ,S season C fuel )
[0061] P hist Historical price series
[0062] O rate Real-time cabin booking rate
[0063] S season Seasonal index (Holidays / Peak Season = 1.5, Off-Season = 0.7)
[0064] C fuel Fuel cost volatility
[0065] Hidden layer:
[0066] Three-layer LSTM unit (forget gate: determines information to discard; input gate: determines information to update; output gate: determines information to output) Loss function: MSE;
[0067] Fully connected layer: Mapped to price forecasts for the next 30 days
[0068] Output layer:
[0069] Output ticket price trends and best time to buy tickets, as well as the lowest price range and the discount rate sequence for the next 30 days δt=(δ1,δ2,…,δ 30 Information such as )
[0070] The adaptive weight generation steps are as follows: A static baseline weight vector W0 is generated based on the scene label L, and a convergent weight vector W is generated using an iterative formula with session behavior data as input. K The scenario tag L includes family and parent-child activities (F) and elderly care (S); the scenario tag L is optimized through historical behavior data; the session behavior data includes search term frequency and page dwell time.
[0071] In the weight adaptive generation step, the scene label L is used as input to generate a static baseline vector:
[0072]
[0073] in,
[0074]
[0075] PriceSens∈[0,10] represents the user's price sensitivity score;
[0076]
[0077] include and
[0078]
[0079] The calculation formula for the iterative formula in the adaptive weight generation step is as follows:
[0080]
[0081] Where k = 1, 2, 3…m, the cycle continues until ||W k -W k-1 ||<ε,K i For search term frequency, T i K represents the page dwell time; α and β are adaptive coefficients based on user identity type. max T is the cardinality for normalized search term frequencies. max γ is the normalized base for dwell time; F Parent-child preference moderating coefficient, γ S This is the adjustment coefficient for silver-haired healthcare.
[0082] Recommendation decision generation steps: The price comparison feature vector and weight vector are linearly combined to output a personalized recommendation ranking; the recommendation decision generation steps include the following steps:
[0083] The price comparison feature vector and weight vector of cruise products are linearly combined, and the specific calculation formula is as follows:
[0084] S user =V comp *W K , among which, S user For personalized scoring;
[0085] Personalized recommendations are generated and ranked based on the individual scores of different cruise products, and the personalized recommendation ranking is output.
[0086] Closed-loop collaborative optimization steps: Real-time collection of user feedback data regarding personalized recommendations and rankings of cruise products, including the recommendation conversion rate C; analysis of the cruise product recommendation conversion rate C; when the recommendation conversion rate C... <C thres At that time, the feedback data is used as historical behavior data to update the discount rate δt and the calibration weight vector W. k The optimized δt is then compared with the calibration weight vector W. kFeedback is fed back to the multi-dimensional price comparison feature construction step and the weight adaptive generation step for the next round of decision-making.
[0087] This solution implements a dual-loop decision-making architecture for cruise products: the inner loop dynamically optimizes price comparison weights through real-time behavioral data, achieving personalized recommendations with sub-second response times; the outer loop, based on recommendation conversion rate feedback, collaboratively calibrates the price prediction model δt and the weight vector Wk, driving the self-evolution of multi-dimensional price comparison features. This nested dual-loop architecture deeply couples voyage parameters, service density, and price trends, forming an inseparable closed-loop intelligent entity, completely resolving the industry pain point of the disconnect between static price comparison and dynamic demand in traditional solutions. It achieves multi-dimensional dynamic price comparison and personalized recommendations for cruise products, meeting user needs and improving user satisfaction.
[0088] A multi-dimensional dynamic price comparison and personalized recommendation system for cruise products, the system employs the multi-dimensional dynamic price comparison and personalized recommendation method for cruise products as described above, and specifically includes the following modules:
[0089] Multi-dimensional price comparison feature construction module: used to obtain the basic dimension parameters of cruise products and historical behavior data. The basic dimension parameters of cruise products include cruise days D, number of ports of call P, scenario label L, user preference label F / S, and real-time price P. now User preference tags F / S include family and parent-child (F) and senior citizen (S); service added value dimension is calculated based on flight range basic dimension parameters, and the service added value dimension includes catering density index R. d and entertainment facility density index E d The discount rate δt output by the LSTM price prediction model is obtained as the time dimension object, and the historical behavioral data is used to train and optimize the LSTM price prediction model; the basic distance dimension parameter, the service added value dimension, and the time dimension object representing the price trend are used to generate a price comparison feature vector V. comp =(D,P,P) now ,Rd,Ed,δt,F,S);
[0090] The catering density index R d The calculation formula is as follows:
[0091] R d =(∑(x*t)) / n*Φ1(D,P),Φ1(D,P)=D / 7*(1-0.08P)
[0092] Where x is the restaurant rating, t is the opening hours, n is the passenger capacity, and Φ1(D,P) is the dynamic adjustment function.
[0093] The density index of entertainment facilities E d The calculation formula is as follows:
[0094]
[0095] Where y is the number of facilities, m is the predicted usage frequency, n is the passenger capacity, and Φ2(D,P) is the dynamic adjustment function.
[0096] The generation of the scene label L and user preference label F / S is performed through the following hierarchical label generation module:
[0097] Layered tag generation module: This module takes user historical order data and real-time search keywords as input, executes a random forest-based classification algorithm, and outputs discrete scene tags L∈{Family and Parenting (F), Senior Care (S), Business Meeting (B)}.
[0098] Using real-time page clickstream data and facility dwell time as input, a time-decay weighted calculation is performed: F = ∑(parent-child related clicks × e) -0.1t ), output continuous user preference labels F∈[0,1],S∈[0,1].
[0099] The adaptive weight generation module generates a static baseline weight vector W0 based on the scene label L, and uses session behavior data as input to generate a convergent weight vector W through an iterative formula. K The scene tags L include family and parent-child activities (F) and elderly care (S); the conversation behavior data includes search term frequency and page dwell time; in the weight adaptive generation step, the scene tags L are used as input to generate a static baseline vector.
[0100]
[0101] in,
[0102]
[0103] PriceSens∈[0,10] represents the user's price sensitivity score;
[0104]
[0105] include and
[0106]
[0107] The calculation formula for the iterative formula in the adaptive weight generation step is as follows:
[0108]
[0109] Where k = 1, 2, 3…m, the cycle continues until ||W k-W k-1 ||<ε,K i For search term frequency, T i K represents the page dwell time; α and β are adaptive coefficients based on user identity type. max T is the cardinality for normalized search term frequencies. max γ is the normalized base for dwell time; F Parent-child preference moderating coefficient, γ S This is the adjustment coefficient for silver-haired healthcare.
[0110] Recommendation decision generation module: used to linearly combine price comparison feature vectors and weight vectors to output personalized recommendation rankings; the recommendation decision generation module includes the following sub-modules:
[0111] The personalized score calculation submodule is used to linearly combine the price comparison feature vector and weight vector of cruise products. The specific calculation formula is as follows:
[0112] S user =V comp *W K , among which, S user For personalized scoring;
[0113] The Personalized Recommendation and Ranking Submodule is used to perform personalized recommendation and ranking based on the personalization scores of different cruise products, and output the personalized recommendation ranking.
[0114] Closed-loop collaborative optimization module: used to collect feedback data on user feedback regarding personalized recommendations and rankings of cruise products in real time, including the recommendation conversion rate C; analyzes the recommendation conversion rate C of cruise products, and when the recommendation conversion rate C... <C thres At that time, the feedback data is used as historical behavior data to update the discount rate δt and the calibration weight vector W. k The optimized δt is then compared with the calibration weight vector W. k Feedback is fed back to the multi-dimensional price comparison feature construction step and the weight adaptive generation step for the next round of decision-making.
[0115] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. A method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products, characterized in that, Includes the following steps: Multi-dimensional price comparison feature construction steps: Obtain the basic dimensional parameters of cruise products, including cruise duration (D), number of ports of call (P), scenario tag (L), user preference tag (F / S), and real-time price (P). now User preference tags F / S include family and parent-child (F) and senior citizen (S); service added value dimension is calculated based on flight range basic dimension parameters, and the service added value dimension includes catering density index R. d and entertainment facility density index E d ; Obtain the discount rate δt output by the LSTM price prediction model as the time dimension object; Generate a price comparison feature vector V by combining the basic distance dimension parameters, the service added value dimension, and the time dimension object representing the price trend. comp =(D,P,P) now ,Rd,Ed,δt,F,S); The adaptive weight generation steps are as follows: A static baseline weight vector W0 is generated based on the scene label L, and a convergent weight vector W is generated using an iterative formula with session behavior data as input. K The scenario label L includes family and parent-child activities (F) and elderly care (S). The scene tag L is optimized using historical behavior data; the session behavior data includes search term frequency and page dwell time. Recommendation decision generation steps: Linearly combine the price comparison feature vector and the weight vector to output a personalized recommendation ranking; Closed-loop collaborative optimization steps: Real-time collection of user feedback data regarding personalized recommendations and rankings of cruise products, including the recommendation conversion rate C; analysis of the cruise product recommendation conversion rate C; when the recommendation conversion rate C... <C thres At that time, the feedback data is used as historical behavior data to update the discount rate δt and the calibration weight vector W. k The optimized δt is then compared with the calibration weight vector W. k Feedback is fed back to the multi-dimensional price comparison feature construction step and the weight adaptive generation step for the next round of decision-making.
2. The method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to claim 1, characterized in that: The catering density index R d The calculation formula is as follows: R d =(∑(x*t)) / n*Φ1(D,P),Φ1(D,P)=D / 7*(1-0.08P) Where x is the restaurant rating, t is the opening hours, n is the passenger capacity, and Φ1(D,P) is the dynamic adjustment function.
3. The method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to claim 2, characterized in that: The density index of entertainment facilities E d The calculation formula is as follows: Where y is the number of facilities, m is the predicted usage frequency, n is the passenger capacity, and Φ2(D,P) is the dynamic adjustment function.
4. The method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to claim 3, characterized in that: The generation of the scene label L and user preference label F / S includes the following steps: Hierarchical tag generation steps: Using users' historical order data and real-time search keywords as input, execute a classification algorithm based on random forest to output discrete scene tags L∈{family and parenting (F), elderly care (S), business meetings (B)}; Using real-time page clickstream data and facility dwell time as input, a time-decay weighted calculation is performed: F = ∑(parent-child related clicks × e) -0.1t ), output continuous user preference labels F∈[0,1],S∈[0,1].
5. The method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to claim 4, characterized in that: In the weight adaptive generation step, the scene label L is used as input to generate a static baseline vector: in, PriceSens∈[0,10] represents the user's price sensitivity score; include and 6. The method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to claim 5, characterized in that: The calculation formula for the iterative formula in the adaptive weight generation step is as follows: Where k = 1, 2, 3…m, the cycle continues until ||W k -W k-1 ||<ε,K i For search term frequency, T i K represents the page dwell time; α and β are adaptive coefficients based on user identity type. max T is the cardinality for normalized search term frequencies. max γ is the normalized base for dwell time; F Parent-child preference moderating coefficient, γ S This is the adjustment coefficient for elderly healthcare.
7. The method for multi-dimensional dynamic price comparison and personalized recommendation of cruise products according to claim 6, characterized in that: The recommendation decision generation step includes the following steps: The price comparison feature vector and weight vector of cruise products are linearly combined, and the specific calculation formula is as follows: S user =V comp *W K , among which, S user For personalized scoring; Personalized recommendations are generated and ranked based on the individual scores of different cruise products, and the personalized recommendation ranking is output.
8. A multi-dimensional dynamic price comparison and personalized recommendation system for cruise products, characterized in that, The system employs a multi-dimensional dynamic price comparison and personalized recommendation method for cruise products as described in any one of claims 1-7, specifically including the following modules: Multi-dimensional price comparison feature construction module: used to obtain the basic dimension parameters of cruise products and historical behavior data. The basic dimension parameters of cruise products include cruise days D, number of ports of call P, scenario label L, user preference label F / S, and real-time price P. now User preference tags F / S include family and parent-child (F) and senior citizen (S); service added value dimension is calculated based on flight range basic dimension parameters, and the service added value dimension includes catering density index R. d and entertainment facility density index E d ; Obtain the discount rate δt output by the LSTM price prediction model as the time dimension object; Generate a price comparison feature vector V by combining the basic distance dimension parameters, the service added value dimension, and the time dimension object representing the price trend. comp =(D,P,P) now ,Rd,Ed,δt,F,S); The adaptive weight generation module generates a static baseline weight vector W0 based on the scene label L, and uses session behavior data as input to generate a convergent weight vector W through an iterative formula. K The scenario label L includes family and parent-child activities (F) and elderly care (S). The scene tag L is optimized using historical behavior data; the session behavior data includes search term frequency and page dwell time. Recommendation decision generation module: used to linearly combine the price comparison feature vector and the weight vector to output personalized recommendation ranking; Closed-loop collaborative optimization module: used to collect feedback data on user feedback regarding personalized recommendations and rankings of cruise products in real time, including the recommendation conversion rate C; analyzes the recommendation conversion rate C of cruise products, and when the recommendation conversion rate C... <C thres At that time, the feedback data is used as historical behavior data to update the discount rate δt and the calibration weight vector W. k The optimized δt is then compared with the calibration weight vector W. k Feedback is fed back to the multi-dimensional price comparison feature construction step and the weight adaptive generation step for the next round of decision-making.