Personalized tourism recommendation system based on AI

Abstract

The invention relates to the technical field of recommendation systems, in particular to an AI-based personalized tourism recommendation system, which comprises a video data analysis module for analyzing and monitoring video data on the basis of an image recognition technology, detecting a tourist dense area in a scenic area, converting video image data into a scenic spot people flow density index, and sending the scenic spot people flow density index to a recommendation module; and then matching the scenic spot people flow density index with an actual map position through a geographic information system, and generating people flow data in a positioning area. According to the invention, through integration of real-time image identification and a geographic information system, rapid measurement of the scenic spot people flow density is realized, the data analysis speed and precision are improved, such optimization improves the people flow monitoring efficiency, effectively prevents congestion, allows real-time adjustment of management strategies, and the system accurately analyzes the user behavior and predicts the behavior mode, thereby improving the user experience. And meanwhile, the time sequence analysis technology is utilized to predict the people flow change, scientifically adjust the travel route and time, optimize the resource allocation and enhance the operation rationality.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation systems, and in particular to an AI-based personalized tourism recommendation system. Background Art

[0002] A recommendation system is an information filtering system that aims to predict and recommend products or services that users may be interested in. These systems usually rely on batch data analysis, including user history, preferences, behavior of similar users, and content characteristics. In terms of technical implementation, recommendation systems may use a variety of algorithms, such as collaborative filtering, content-based recommendation, and hybrid recommendation. The key to this technical field is to optimize the relevance and personalization of recommendations through algorithms to improve user experience and business efficiency.

[0003] The personalized tourism recommendation system is a specific example of the application of recommendation system technology. The system focuses on providing customized travel destinations, activity suggestions and route planning based on the user's travel history, preferences, budget and time. Its main purpose is to help users find the travel experience that best suits their personal interests and needs, while providing tourism operators with an effective tool to attract and serve target customers and enhance customer satisfaction and loyalty.

[0004] Existing recommendation systems mostly rely on historical data and static preference analysis, and lack the ability to dynamically respond to real-time events and changes in traffic. In an environment with high demand changes, this limitation prevents the system from updating recommendations in a timely manner, affecting management efficiency and user satisfaction. For example, when encountering emergencies or drastic changes in traffic during holidays, traditional systems often cannot respond effectively, resulting in unbalanced resource allocation and reduced service quality. In addition, existing systems often cannot accurately match users' real-time needs in terms of personalized recommendations, especially when user behaviors are diverse and expectations are changing rapidly. The lack of real-time data analysis capabilities and highly personalized services means that existing technologies need to be improved in terms of accuracy and adaptability. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose an AI-based personalized tourism recommendation system.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solution: The AI-based personalized tourism recommendation system includes:

[0007] The video data analysis module analyzes the surveillance video data based on image recognition technology, detects the tourist-dense areas in the scenic area, converts the video image data into the density index of the scenic spot flow, and then matches the density index of the scenic spot flow with the actual map location through the geographic information system to generate the flow data of the positioning area;

[0008] The user behavior analysis module integrates the user's historical access records and behavior preferences based on the pedestrian flow data in the positioning area, performs user grouping, distinguishes multiple user categories, establishes a behavior prediction model for each category, predicts future behavior patterns and preference trends, and generates behavior trend prediction results;

[0009] The crowd flow prediction module uses time series analysis technology based on the behavior trend prediction results to predict the crowd flow of the scenic spot in multiple time periods in the future, compare the capacity of the scenic spot, analyze the potential congestion, predict the congestion probability of the scenic spot in multiple time periods in the future, and generate congestion prediction results;

[0010] The recommendation strategy adjustment module optimizes and adjusts the travel route and visit time based on the congestion prediction result, updates the recommendation model to match the real-time passenger flow and user preference data, updates the adjusted travel route and time, and establishes a recommendation plan.

[0011] As a further solution of the present invention, the steps for obtaining the tourist flow density index of the scenic spot are specifically as follows:

[0012] Use image recognition technology to analyze the surveillance video data frame by frame and identify the number of tourists N in each frame i , for video frame i, use the formula:

[0013]

[0014] Calculate the total number of visitors N in all frames within the total number of video frames F total , generate the total number of visitors in the video;

[0015] Among them, N i represents the number of tourists in the i-th frame, c i is the clarity weight of the i-th frame, and the visitor count of each frame is adjusted to reflect the change of visual clarity;

[0016] The number of visitors per frame N i Converted into density index, refer to the screen area A and field of view coverage v i , using the formula:

[0017]

[0018] Calculate the density index D for each frame i , generate the density index D of all frames i ;

[0019] Calculate the density index D of all frames i The weighted average of i and stability factor s i , using the formula:

[0020]

[0021] Calculate the weighted average density and generate the weighted average traffic density D avg ;

[0022] Among them, D i Represents the density index of each frame, w i represents the temporal weight of the i-th frame, s i represents the stability factor, referring to the stability of time changes and frames;

[0023] Through the geographic information system, the weighted average flow density D avg Paired with the actual map location L, an innovative formula is used:

[0024]

[0025] Generate the tourist flow density index D location ;

[0026] Among them, L represents the geographic coordinates, L0 represents the central geographic coordinates, and k is the sensitivity adjustment parameter.

[0027] As a further solution of the present invention, the steps of acquiring the crowd flow data in the positioning area are specifically as follows:

[0028] Extract the number of people flow N in each frame from the image data i , calculate the total flow N for each area R R , using the improved formula:

[0029]

[0030] Generate total crowd flow estimates N for each area R ;

[0031] Among them, N i represents the number of tourists in the i-th frame, c i represents the clarity weight, v i represents the field of view coverage weight, f vi represents the visibility score of the i-th frame;

[0032] The number of people flow in each area N R Normalize the area and calculate the crowd density index D R , using the formula:

[0033]

[0034] Generate the flow density D of each area R ;

[0035] Among them, N Rrepresents the total flow estimate of area R, A R represents the area of ​​region R, δ is the area adjustment coefficient;

[0036] The crowd density D R Match with geographic information system data, make positioning corrections, and calculate the formula:

[0037]

[0038] Refer to the square effect of regional distance to generate the location area flow data P location ;

[0039] Among them, L R represents the central geographic coordinates of each area, L0 represents the central geographic coordinates of the scenic spot, and k and μ are adjustment coefficients.

[0040] As a further solution of the present invention, the steps of establishing the behavior prediction model are specifically as follows:

[0041] Based on the flow of people in the positioning area and the historical visit records of users, frequent visitors and low-frequency visitors are screened, and the visit frequency of each user is calculated using the formula:

[0042]

[0043] Generate active user list U f ;

[0044] Among them, f i represents the number of visits by user i, when f i When it exceeds the average number of visits within time t, the value is 1, otherwise it is 0. θ(i, t) represents the threshold function, and t represents the time period;

[0045] For the active user list U f , group users according to their behavioral preferences, use weighted clustering algorithm to assign users to differentiated categories, and apply the formula:

[0046]

[0047] Generate user group C k ;

[0048] Among them, b j represents the behavior preference of user j, p j represents the preference probability, w j Represents the weight of the behavior;

[0049] For each user group C k , using multivariate regression analysis, using the formula:

[0050]

[0051] Generating a behavioral prediction model

[0052] Among them, α0, α n represents the model coefficient, v nk Is user category C k behavioral variables.

[0053] As a further solution of the present invention, the steps of obtaining the behavior trend prediction results are specifically as follows:

[0054] Collect multiple user categories C k Historical behavior data H k , calculate the behavior frequency and preference strength of each category, using the formula:

[0055]

[0056] Generate a behavior frequency set F for each category k ;

[0057] Among them, h ik Indicates user category C k The behavior data point of the i-th user in the historical period, w ik is the exponential weight of the behavior data point, n k For user category C k Number of users;

[0058] Using the behavior prediction model P model Analysis F k , determine the behavior patterns and preference trends in future time periods, and apply the improved prediction formula:

[0059]

[0060] Generate the behavior trend prediction result F for each user category trends ;

[0061] Among them, α k and γ k For user category C k The prediction coefficient and square root weight, F k Represents a set of behavior frequencies.

[0062] As a further solution of the present invention, the step of obtaining the congestion prediction result is specifically as follows:

[0063] Based on the behavioral trend prediction results, time series analysis is used to predict the flow of tourists in scenic spots in multiple time periods in the future. The formula is:

[0064]

[0065] Generate predicted traffic results P t ;

[0066] Among them, P t represents the predicted flow of people at time t, x t-i represents the flow of people in the past time period i, β0 is the model intercept, β i are the model coefficients;

[0067] The predicted flow result P t Compared with the capacity C of the scenic area, the formula is:

[0068]

[0069] Use the min function to maintain the congestion probability not exceeding 1 and calculate the congestion probability;

[0070] Among them, D t represents the congestion probability, P t represents the predicted flow of people, and C represents the capacity of the scenic spot;

[0071] The congestion probability D t Converted into congestion risk value to evaluate the congestion situation in different time periods. The formula is:

[0072]

[0073] Generate congestion prediction result F t ;

[0074] Where λ represents the sensitivity adjustment factor.

[0075] As a further solution of the present invention, the steps of establishing the recommended solution are specifically as follows:

[0076] The congestion prediction results are analyzed and combined with real-time passenger flow data. A weighted model is used to balance the impact of congestion data and passenger flow data. The formula is:

[0077] R init =ω·F congestion +(1-ω)·L real-time

[0078] Calculate the preliminary optimized travel route R init ;

[0079] Among them, F congestion represents the congestion prediction result, L real-time represents real-time passenger flow data, ω is a weight parameter that adjusts the relative importance of congestion data and real-time data;

[0080] Based on the preliminary optimized travel route, integrating user preference data, the formula is adopted:

[0081] R final =θ·R init +(1-θ)·U preferences

[0082] Adjust the recommendation model, refine the travel routes and visit times, and establish recommended routes;

[0083] Among them, U preferences represents user preference data, and θ is an adjustment factor used to balance the impact of historical recommendations and user current preferences.

[0084] Compared with the prior art, the advantages and positive effects of the present invention are:

[0085] In the present invention, through the integration of real-time image recognition and geographic information system, the rapid and accurate measurement of the density of tourist flow at scenic spots is achieved, and the speed and accuracy of data analysis are significantly improved. This processing logic optimizes the efficiency of crowd monitoring and prevents the occurrence of congestion, which is specifically reflected in the ability to quickly respond to dense crowds and adjust management strategies accordingly. Further, through the refined integration and analysis of user behavior data, the system can predict the behavior patterns of different user groups and provide accurate personalized travel suggestions. Through the application of time series analysis technology, the system not only predicts future trends in crowd flow changes, but also realizes the scientific adjustment of travel routes and visit times, optimizes resource allocation, and enhances the scientificity and rationality of operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 is a system flow chart of the present invention;

[0087] Figure 2 A flow chart of steps for obtaining the crowd density index of a scenic spot according to the present invention;

[0088] Figure 3 This is a flow chart of the steps for obtaining the human flow data in the positioning area of ​​the present invention;

[0089] Figure 4 A flow chart of steps for establishing a behavior prediction model of the present invention;

[0090] Figure 5 A flowchart of steps for obtaining behavior trend prediction results of the present invention;

[0091] Figure 6 A flow chart of the steps for obtaining congestion prediction results of the present invention;

[0092] Figure 7 Flow chart of the steps for establishing the recommended solution of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0094] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0095] Embodiment 1

[0096] See also Figure 1 , the AI-based personalized tourism recommendation system includes:

[0097] The video data analysis module analyzes the surveillance video data based on image recognition technology, detects the tourist-dense areas in the scenic area, converts the video image data into the density index of the scenic spot flow, and then matches the density index of the scenic spot flow with the actual map location through the geographic information system to generate the flow data of the positioning area;

[0098] The user behavior analysis module integrates the user's historical access records and behavior preferences based on the location area traffic data, performs user grouping, distinguishes multiple user categories, establishes a behavior prediction model for each category, predicts future behavior patterns and preference trends, and generates behavior trend prediction results;

[0099] The crowd flow prediction module uses time series analysis technology based on the behavioral trend prediction results to predict the crowd flow of scenic spots in multiple time periods in the future, compare the capacity of scenic spots, analyze potential congestion, predict the congestion probability of scenic spots in multiple time periods in the future, and generate congestion prediction results;

[0100] The recommendation strategy adjustment module optimizes and adjusts the travel routes and visit times based on the congestion prediction results, updates the recommendation model to match the real-time passenger flow and user preference data, updates the adjusted travel routes and times, and establishes a recommendation plan.

[0101] The passenger flow data in the positioning area specifically include the coordinates of densely populated areas, peak passenger flow periods, and map coverage granularity. The behavioral trend prediction results include user interest point analysis, activity participation prediction, and tourism product preferences. The congestion prediction results include congestion risk periods, key congestion nodes, and potential risk levels. The recommended plans include route optimization plans, estimated travel time, and recommendation system feedback efficiency.

[0102] See also Figure 2 ,The specific steps for obtaining the tourist flow density index of scenic spots are as follows:

[0103] Use image recognition technology to analyze the surveillance video data frame by frame and identify the number of tourists N in each frame i , for video frame i, use the formula:

[0104]

[0105] Calculate the total number of visitors N in all frames within the total number of video frames F total , generate the total number of visitors in the video;

[0106] Among them, N i represents the number of tourists in the i-th frame, c i is the clarity weight of the i-th frame, and the visitor count of each frame is adjusted to reflect the change of visual clarity;

[0107] The number of visitors per frame N i Converted into density index, refer to the screen area A and field of view coverage v i , using the formula:

[0108]

[0109] Calculate the density index D for each frame i , generate the density index D of all frames i ;

[0110] Calculate the density index D of all frames i The weighted average of i and stability factor s i , using the formula:

[0111]

[0112] Calculate the weighted average density and generate the weighted average traffic density D avg ;

[0113] Among them, D i Represents the density index of each frame, w i represents the temporal weight of the i-th frame, s i represents the stability factor, referring to the stability of time changes and frames;

[0114] Through the geographic information system, the weighted average flow density D avg Paired with the actual map location L, an innovative formula is used:

[0115]

[0116] Generate the tourist flow density index D location ;

[0117] Among them, L represents the geographic coordinates, L0 represents the central geographic coordinates, and k is the sensitivity adjustment parameter.

[0118] formula:

[0119]

[0120] N i : The number of tourists in the i-th frame.

[0121] F: total number of video frames.

[0122] c i : The clarity weight of the i-th frame, ranging from 0.5 to 1, represents the clarity of the picture.

[0123] Derivation process:

[0124] Assume that a video has 5 frames (F=5), and the number of tourists in each frame is N i They are [20, 15, 25, 10, 30] respectively, and the clarity weight c i They are [0.8, 0.9, 0.85, 0.7, 0.95] respectively. Then the total number of tourists N total The calculation of is as follows:

[0125] N total =(20×0.8)+(15×0.9)+(25×0.85)+(10×0.7)+(30×0.95)

[0126] =16+13.5+21.25+7+28.5=86.25

[0127] N total It is 86.25, which represents the estimated total number of tourists in the total frames of the video after referring to the picture clarity.

[0128] formula:

[0129]

[0130] N i : The number of tourists in the i-th frame.

[0131] vi : The coverage rate of the field of view of the i-th frame, ranging from 0.8 to 1, indicates the ratio of the area fully covered by the camera field of view.

[0132] A: Screen area (unit: square meters), assumed to be 50 square meters.

[0133] Derivation process:

[0134] Referring to the first frame, the number of tourists N i =20, visual coverage rate v i =0.8. Then the density index D i The calculation of is as follows:

[0135]

[0136] Income D i It is 0.32, which represents the tourist density per unit area in the first frame.

[0137] formula:

[0138]

[0139] w i : The temporal weight of the i-th frame, reflecting the importance of the frame period or the expected change of tourist flow.

[0140] s i : Stability factor, referring to the impact of frame stability on visitor number calculation, the value range is between 0.9 and 1.

[0141] Derivation process:

[0142] Assume that the time weight w i are [1,2,1.5,1,2] respectively, and the stability factor s i They are [0.95, 0.9, 0.92, 0.94, 0.98] respectively, and the density index D i They are [0.32, 0.27, 0.30, 0.25, 0.35] respectively. Then the weighted average density D avg The calculation of is as follows:

[0143]

[0144] Income D avg It is 0.236, which represents the average per-frame density index after considering time weight and stability.

[0145] formula:

[0146]

[0147] k: sensitivity adjustment parameter, which depends on the sensitivity of the geographical location to the crowd density and is set between 0.01 and 0.05.

[0148] L, L0: geographical coordinates and center geographical coordinates, for example, L=100, L0=95.

[0149] Derivation process:

[0150] Assume k = 0.03, D avg =0.236, L = 100, L0 = 95. Then D location The calculation of is as follows:

[0151]

[0152] Income D location It is 0.194, which represents the tourist density index after adjusting for geographical location and center deviation.

[0153] See also Figure 3 ,The specific steps for obtaining the crowd flow data in the positioning area are:

[0154] Extract the number of people flow N in each frame from the image data i , calculate the total flow N for each area R R , using the improved formula:

[0155]

[0156] Generate total crowd flow estimates N for each area R ;

[0157] Among them, N i represents the number of tourists in the i-th frame, c i represents the clarity weight, v i represents the field of view coverage weight, f vi represents the visibility score of the i-th frame;

[0158] The number of people flow in each area N R Normalize the area and calculate the crowd density index D R , using the formula:

[0159]

[0160] Generate the flow density D of each area R ;

[0161] Among them, N R represents the total flow estimate of area R, A R represents the area of ​​region R, δ is the area adjustment coefficient;

[0162] The crowd density DR Match with geographic information system data, make positioning corrections, and calculate the formula:

[0163]

[0164] Refer to the square effect of regional distance to generate the location area flow data P location ;

[0165] Among them, L R represents the central geographic coordinates of each area, L0 represents the central geographic coordinates of the scenic spot, and k and μ are adjustment coefficients.

[0166] formula:

[0167]

[0168] Parameter interpretation and derivation:

[0169] N i : The number of tourists in the i-th frame, assuming that 30 people are detected in the i-th frame.

[0170] c i : Weight based on clarity, assuming the clarity weight is 0.8.

[0171] v i : Field of view coverage weight, assumed to be 0.9.

[0172] f vi : Visibility function of the frame, assuming it is 1.2, indicating that the frame condition is relatively ideal.

[0173] Example: Assuming there are 10 frames (F=10), the number of people in each frame remains unchanged, calculate the total crowd flow estimate:

[0174]

[0175] This result indicates that, given the weight and number of frames, the estimated total crowd flow is 259 people.

[0176] formula:

[0177]

[0178] Parameter interpretation and derivation:

[0179] N R : The total estimated flow of people in area R is 259 people.

[0180] A R : The area of ​​region R, assumed to be 100 square meters.

[0181] δ: Area adjustment coefficient, assumed to be 0.05.

[0182] Calculation example:

[0183] Calculate the crowd density index:

[0184]

[0185] This result means that there are an average of about 10.56 people per square meter.

[0186] formula: Parameter interpretation and derivation:

[0187] D R : The crowd density index is 10.56.

[0188] k, μ: adjustment coefficients, assuming k=0.1, μ=0.02.

[0189] L R -L0: The distance difference between the center of the area and the center of the attraction, assumed to be 5 meters.

[0190] Calculation example:

[0191] Calculate the flow of people in the positioning area:

[0192]

[0193] This result indicates that, taking into account the influence of geographical location, the adjusted pedestrian density is approximately 4.46 people per square meter.

[0194] See also Figure 4 ,The specific steps of establishing the behavior prediction model are:

[0195] Based on the flow of people in the positioning area and the user's historical visit records, frequent visitors and low-frequency visitors are screened, and the visit frequency of each user is calculated using the formula:

[0196]

[0197] Generate active user list U f ;

[0198] Among them, f i represents the number of visits by user i, when f i When it exceeds the average number of visits within time t, the value is 1, otherwise it is 0. θ(i, t) represents the threshold function, and t represents the time period;

[0199] For active user list U f , group users according to their behavioral preferences, use weighted clustering algorithm to assign users to differentiated categories, and apply the formula:

[0200]

[0201] Generate user group C k ;

[0202] Among them, b j represents the behavior preference of user j, p j represents the preference probability, w j Represents the weight of the behavior;

[0203] For each user group C k , using multivariate regression analysis, using the formula:

[0204]

[0205] Generating a behavioral prediction model

[0206] Among them, α0, α n represents the model coefficient, v nk Is user category C k behavioral variables.

[0207] Formula explanation:

[0208]

[0209] f i : The number of visits by user i in period t.

[0210] θ(i, t): threshold function, when f i When a certain threshold τ is exceeded, the value is 1, otherwise it is 0. The threshold is set to the average number of visits in period t plus twice the standard deviation (for example, if the average number of visits is 10 times and the standard deviation is 3, the threshold τ will be set to 16).

[0211] Calculation example:

[0212] Assume there are 100 users, the average number of visits is 10, the standard deviation is 3, and the threshold τ is 16. If the number of visits by user A in period t is 20 (higher than 16), and the number of visits by user B is 15 (lower than 16):

[0213] f A =20, f B =15

[0214] θ(A, t) = 1, θ(B, t) = 0

[0215] but:

[0216] U f =20·1+15·0=20

[0217] This means that only user A is identified as an active user.

[0218] Formula explanation:

[0219]

[0220] b j : User j’s behavioral preference, such as “shopping”, “traveling”, etc., can be represented by digital codes.

[0221] p j : Preference probability, which indicates the probability that user j chooses a certain preference.

[0222] w j : Behavior weight, determined based on user j’s historical preference data.

[0223] Calculation example:

[0224] Assume that user A's preference is "shopping" (coded as 1), the preference probability is 0.8, and the behavior weight is 0.5:

[0225] b A =1, p A =0.8, w A =0.5

[0226] but:

[0227] C k =1·0.8·0.5=0.4

[0228] This result indicates that user A is assigned to the “shopping” preference group with an influence weight of 0.4.

[0229] Formula explanation:

[0230]

[0231] α0, α1, …, α n : coefficients of the regression model.

[0232] v nk : Behavioral variable, representing user category C k Specific behavioral data in.

[0233] Calculation example:

[0234] Assume that for user category “shopping” (C k ), with model coefficients α0 = 0.1, α1 = 0.2 and behavioral variables v 1k =5:

[0235]

[0236] This means that the behavior prediction model output for the "shopping" category is 1.1, which may represent some predicted shopping frequency or intensity.

[0237] See also Figure 5 , the steps for obtaining the behavior trend prediction results are as follows:

[0238] Collect multiple user categories C k Historical behavior data H k , calculate the behavior frequency and preference strength of each category, using the formula:

[0239]

[0240] Generate a behavior frequency set F for each category k ;

[0241] Among them, h ik Indicates user category C k The behavior data point of the i-th user in the historical period, w ik is the exponential weight of the behavior data point, n k For user category C k Number of users;

[0242] Using the behavior prediction model P model Analysis F k , determine the behavior patterns and preference trends in future time periods, and apply the improved prediction formula:

[0243]

[0244] Generate the behavior trend prediction result F for each user category trends ;

[0245] Among them, α k and γ k For user category C k The prediction coefficient and square root weight of F k Represents a set of behavioral frequencies.

[0246] Detailed explanation of formula and examples:

[0247]

[0248] h ik : Indicates user category C k The behavior data points of the i-th user in the historical period. For example, the number of times user i visited the website in a month.

[0249] w ik : is the exponential weight of the behavior data point, which may be adjusted based on the user's engagement or the importance of a specific behavior. For example, for a shopping website, the purchase behavior is estimated to have a higher weight than the browsing behavior.

[0250] Assumptions:

[0251] User Category C k There are 3 users.

[0252] Behavior data point ik They are 20, 15 and 10 respectively.

[0253] Weight w ik They are 0.5, 0.3 and 0.2 respectively.

[0254] calculate:

[0255] F k =20×e 0.5 +15×e 0.3 +10×e 0.2

[0256] ≈20×1.6487+15×1.3499+10×1.2214

[0257] ≈32.974+20.2485+12.214=65.4365

[0258] This means that F k For this user category, it is 65.4365, which is a weighted behavior frequency indicator that reflects the overall activity level of the user group in a specific period.

[0259] Detailed explanation of formula and examples:

[0260]

[0261] α k :For user category C k The prediction coefficient is determined based on historical data.

[0262] γ k : is the square root weight, which adjusts the sensitivity of the behavioral data of each category.

[0263] Assumptions:

[0264] There are 2 user categories, known as F k They are 65.4365 and 50 respectively.

[0265] Coefficient α k 2 and 1.5 respectively.

[0266] Weight γ k 2 and 3 respectively.

[0267] calculate:

[0268] F trends =2 65.4365 1 / 2 +1.5 501 / 3

[0269] ≈2·8.088+1.5·3.684≈16.176+5.526=21.702

[0270] This result F trends It is 21.702, which represents the comprehensive prediction trend value considering various weights and user behaviors, and is used to evaluate the future estimated behavior patterns and preference changes of these user groups.

[0271] See also Figure 6 ,The specific steps for obtaining the congestion prediction results are:

[0272] Based on the behavioral trend prediction results, time series analysis is used to predict the flow of tourists in scenic spots in multiple time periods in the future. The formula is:

[0273]

[0274] Generate predicted traffic results P t ;

[0275] Among them, P t represents the predicted flow of people at time t, x t-i represents the flow of people in the past time period i, β0 is the model intercept, β i are the model coefficients;

[0276] The predicted flow of people result P t Compared with the capacity C of the scenic area, the formula is:

[0277]

[0278] Use the min function to maintain the congestion probability not exceeding 1 and calculate the congestion probability;

[0279] Among them, D t represents the congestion probability, P t represents the predicted flow of people, and C represents the capacity of the scenic spot;

[0280] The congestion probability D t Converted into congestion risk value to evaluate the congestion situation in different time periods. The formula is:

[0281]

[0282] Generate congestion prediction result F t ;

[0283] Where λ represents the sensitivity adjustment factor.

[0284] Detailed explanation of formula and examples:

[0285]

[0286] β0: The intercept of the time series model, which can be set based on the average level of historical data.

[0287] β i : The coefficient at the i-th time point, affecting the historical data x t-i For the predicted value P t contribution.

[0288] x t-i : The flow of people in the i-th past time period.

[0289] Assumptions:

[0290] Historical traffic data (past three time periods): x t-1 =100,x t-2 =150,x t-3 =120

[0291] Model coefficients: β0 = 50, β1 = 0.5, β2 = 0.3, β3 = 0.2

[0292] calculate:

[0293] P t =50+0.5×100+0.3×150+0.2×120

[0294] P t =50+50+45+24=169

[0295] This indicates a predicted footfall of 169 people.

[0296] Detailed explanation of formula and examples:

[0297]

[0298] P t : Predicted flow of people.

[0299] C: The capacity of the scenic area, assumed to be 200 people.

[0300] calculate:

[0301]

[0302] This indicates that the probability of congestion during this time period is 84.5%.

[0303] Detailed explanation of formula and examples:

[0304]

[0305] λ: sensitivity adjustment factor, set to 0.5 to enhance the sensitivity of congestion probability.

[0306] D t : Congestion probability.

[0307] calculate:

[0308] F t =100×(1-e -0.5×0.845 )

[0309] F t =100×(1-e -0.4225 )

[0310] F t =100×(1-0.6557)

[0311] F t =100×0.3443=34.43

[0312] This indicates that the congestion risk value is 34.43, which means a moderate congestion risk.

[0313] See also Figure 7 , the specific steps for establishing the recommended solution are:

[0314] The congestion prediction results are analyzed and combined with real-time passenger flow data. A weighted model is used to balance the impact of congestion data and passenger flow data. The formula is:

[0315] R init =ω·F congestion +(1-ω)·L real-time

[0316] Calculate the preliminary optimized travel route R init ;

[0317] Among them, F congestion represents the congestion prediction result, L real-time represents real-time passenger flow data, ω is a weight parameter that adjusts the relative importance of congestion data and real-time data;

[0318] Based on the preliminary optimization of travel routes, integrating user preference data, the formula is adopted:

[0319] R final =θ·R init +(1-θ)·U preferences

[0320] Adjust the recommendation model, refine the travel routes and visit times, and establish recommended routes;

[0321] Among them, U preferences represents user preference data, and θ is an adjustment factor used to balance the impact of historical recommendations and user current preferences.

[0322] Formula explanation:

[0323] R init =ω·F congestion +(1-ω)·L real-time

[0324] R init : Initial adjustment of the tourist route.

[0325] F congestion： : Congestion prediction results, assumed to be obtained from a congestion prediction model.

[0326] L real-time : Real-time crowd flow data is assumed to be obtained from the real-time monitoring system of the scene.

[0327] ω: Weight parameter, which determines the relative importance of congestion prediction and real-time passenger flow data in route optimization. Assuming it is 0.6, this weight is determined based on previous optimization experience to ensure that congestion data has a higher consideration weight.

[0328] Assumptions:

[0329] F congestion =75 (congestion index, based on a 100-point scale, 75 indicates a higher likelihood of congestion)

[0330] L real-time =120 (real-time flow of people, in units of people)

[0331] ω=0.6

[0332] calculate:

[0333] R init =0.6·75+0.4·120

[0334] R init =45+48=93

[0335] This means that based on the current congestion situation and real-time passenger flow, our initial route optimization score is 93, which can be used to evaluate the optimization effect of the route.

[0336] Formula explanation:

[0337] R final =θ·R init +(1-θ)·U preferences

[0338] R final : Recommended travel routes and times.

[0339] R init : The initial recommended route obtained.

[0340] U preferences : User preference data, including user's historical access records and ratings.

[0341] θ: An adjustment factor to balance the impact of historical recommendations and the user’s current preferences. Assuming it is 0.7, this reflects a high degree of trust in the initial recommendation while still giving appropriate weight to the user’s preferences.

[0342] Calculation example:

[0343] Assumptions:

[0344] R init =93

[0345] U preferences =85 (user preference rating, 100 points)

[0346] θ=0.7

[0347] calculate:

[0348] R final =0.7·93+0.3·85

[0349] R final =65.1+25.5=90.6

[0350] This means that after integrating user preferences and congestion data, the recommended route customized for the user scores 90.6, which reflects the adaptability of the route and user satisfaction.

[0351] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. AI-based personalized tourism recommendation system, characterized by: The system comprises: The video data analysis module analyzes the surveillance video data based on image recognition technology, detects the tourist-dense areas in the scenic area, converts the video image data into the density index of the scenic spot flow, and then matches the density index of the scenic spot flow with the actual map location through the geographic information system to generate the flow data of the positioning area; The user behavior analysis module integrates the user's historical access records and behavior preferences based on the pedestrian flow data in the positioning area, performs user grouping, distinguishes multiple user categories, establishes a behavior prediction model for each category, predicts future behavior patterns and preference trends, and generates behavior trend prediction results; The crowd flow prediction module uses time series analysis technology based on the behavior trend prediction results to predict the crowd flow of the scenic spot in multiple time periods in the future, compare the capacity of the scenic spot, analyze the potential congestion, predict the congestion probability of the scenic spot in multiple time periods in the future, and generate congestion prediction results; The recommendation strategy adjustment module optimizes and adjusts the travel route and visit time based on the congestion prediction result, updates the recommendation model to match the real-time passenger flow and user preference data, updates the adjusted travel route and time, and establishes a recommendation plan.

2. The AI-based personalized tourism recommendation system according to claim 1, characterized in that: The steps for obtaining the tourist flow density index of the scenic spot are specifically as follows: Use image recognition technology to analyze the surveillance video data frame by frame and identify the number of tourists N in each frame i , for video frame i, use the formula: Calculate the total number of visitors N in all frames within the total number of video frames F total , generate the total number of visitors in the video; Among them, N i represents the number of tourists in the i-th frame, c i is the clarity weight of the i-th frame, and the visitor count of each frame is adjusted to reflect the change of visual clarity; The number of visitors per frame N i Converted into density index, refer to the screen area A and field of view coverage v i , using the formula: Calculate the density index D for each frame i , generate the density index D of all frames i ; Calculate the density index D of all frames i The weighted average of i and stability factor s i , using the formula: Calculate the weighted average density and generate the weighted average traffic density D avg ; Among them, D i Represents the density index of each frame, w i represents the temporal weight of the i-th frame, s i represents the stability factor, referring to the stability of time changes and frames; Through the geographic information system, the weighted average flow density D avg Paired with the actual map location L, an innovative formula is used: Generate the tourist flow density index D location ; Among them, L represents the geographic coordinates, L0 represents the central geographic coordinates, and k is the sensitivity adjustment parameter.

3. The AI-based personalized tourism recommendation system according to claim 2, characterized in that: The steps for acquiring the crowd flow data in the positioning area are specifically as follows: Extract the number of people flow N in each frame from the image data i , calculate the total flow N for each area R R , using the improved formula: Generate total crowd flow estimates N for each area R ; Among them, N i represents the number of tourists in the i-th frame, c i represents the clarity weight, v i represents the visual coverage weight, f vi represents the visibility score of the i-th frame; The number of people flow in each area N R Normalize the area and calculate the crowd density index D R , using the formula: Generate the flow density D of each area R ; Among them, N R represents the total flow estimate of area R, A R represents the area of ​​region R, δ is the area adjustment coefficient; The crowd density D R Match with geographic information system data, make positioning corrections, and calculate the formula: Refer to the square effect of regional distance to generate the location area flow data P location ; Among them, L R represents the central geographic coordinates of each area, L0 represents the central geographic coordinates of the scenic spot, and k and μ are adjustment coefficients.

4. The AI-based personalized tourism recommendation system according to claim 3, characterized in that: The steps for establishing the behavior prediction model are specifically as follows: Based on the flow of people in the positioning area and the historical visit records of users, frequent visitors and low-frequency visitors are screened, and the visit frequency of each user is calculated using the formula: Generate active user list U f ; Among them, f i represents the number of visits by user i, when f i When it exceeds the average number of visits within time t, the value is 1, otherwise it is 0. θ(i, t) represents the threshold function, and t represents the time period; For the active user list U f , group users according to their behavioral preferences, use weighted clustering algorithm to assign users to differentiated categories, and apply the formula: Generate user group C k ; Among them, b j represents the behavior preference of user j, p j represents the preference probability, w j Represents the weight of the behavior; For each user group C k , using multivariate regression analysis, using the formula: Generating a behavioral prediction model Among them, α0, α n represents the model coefficient, v nk Is user category C k behavioral variables.

5. The AI-based personalized tourism recommendation system according to claim 4, characterized in that: The steps for obtaining the behavior trend prediction result are specifically as follows: Collect multiple user categories C k Historical behavior data H k , calculate the behavior frequency and preference strength of each category, using the formula: Generate a behavior frequency set F for each category k ; Among them, h ik Indicates user category C k The behavior data point of the i-th user in the historical period, w ik is the exponential weight of the behavior data point, n k For user category C k Number of users; Using the behavior prediction model P model Analysis F k , determine the behavior patterns and preference trends in future time periods, and apply the improved prediction formula: Generate the behavior trend prediction result F for each user category trends ; Among them, α k and γ k For user category C k The prediction coefficient and square root weight, F k Represents a set of behavior frequencies.

6. The AI-based personalized tourism recommendation system according to claim 5, characterized in that: The steps for obtaining the congestion prediction result are specifically as follows: Based on the behavioral trend prediction results, time series analysis is used to predict the flow of tourists in scenic spots in multiple time periods in the future. The formula is: Generate predicted traffic results P t ; Among them, P t represents the predicted flow of people at time t, x t-i represents the flow of people in the past time period i, β0 is the model intercept, β i are the model coefficients; The predicted flow result P t Compared with the capacity C of the scenic area, the formula is: Use the min function to maintain the congestion probability not exceeding 1 and calculate the congestion probability; Among them, D t represents the congestion probability, P t represents the predicted flow of people, and C represents the capacity of the scenic spot; The congestion probability D t Converted into congestion risk value to evaluate the congestion situation in different time periods. The formula is: Generate congestion prediction result F t ; Where λ represents the sensitivity adjustment factor.

7. The AI-based personalized tourism recommendation system according to claim 6, characterized in that: The steps for establishing the recommended scheme are specifically as follows: The congestion prediction results are analyzed and combined with real-time passenger flow data. A weighted model is used to balance the impact of congestion data and passenger flow data. The formula is: R init =ω·F congestion +(1-ω)·L real-time Calculate the preliminary optimized travel route R init ; Among them, F congestion represents the congestion prediction result, L real-time represents real-time passenger flow data, ω is a weight parameter that adjusts the relative importance of congestion data and real-time data; Based on the preliminary optimized travel route, integrating user preference data, the formula is adopted: R final =θ·R init +(1-θ)·U preferences Adjust the recommendation model, refine the travel routes and visit times, and establish recommended routes; Among them, U preferences represents user preference data, and θ is an adjustment factor used to balance the impact of historical recommendations and user current preferences.

Images