Data processing method for improving customized tourism recommendation algorithm model

Abstract

The invention discloses a data processing method for improving a customized tourism recommendation algorithm model, relates to the technical field of big data processing, and effectively solves three core problems of a tourism recommendation system: for data sparsity, multi-source data fusion is realized through time sequence data set construction and cross-modal contrast learning of a time window segmentation interval 1h / 24h / 7d; for individual deficiency, an attention mechanism is adopted to calculate a user interest weight Ayh, and three types of interest attenuation models are established; and aiming at poor dynamic adaptability, an asynchronous data stream framework with a construction response smaller than or equal to a fixed time length and an environment perception mechanism, a dual matching degree threshold is set to realize real-time adjustment, and finally a closed-loop optimization system is formed through reinforcement learning, so that the recommendation accuracy and the user experience are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to a data processing method for improving a customized tourism recommendation algorithm model. Background Art

[0002] Travel recommendation algorithms originated from collaborative filtering technology in the 1990s. With the development of the Internet and big data, they have gradually integrated content recommendation, knowledge-based recommendation, and hybrid recommendation technologies. In recent years, advanced technologies such as deep learning, reinforcement learning, and graph neural networks have been introduced, further improving the personalization and accuracy of recommendations.

[0003] However, current tourism recommendation systems generally have the following problems:

[0004] 1. Data sparsity: The dimensions of behavior-related data and attraction feature data do not match, resulting in insufficient recommendation accuracy.

[0005] 2. Lack of personalization: Traditional collaborative filtering algorithms do not fully consider users' real-time preferences and scenario differences;

[0006] 3. Poor dynamic adaptability: Lack of response mechanism to real-time factors such as holidays and weather. Summary of the Invention

[0007] In view of the shortcomings of the existing technology, the present invention provides a data processing method for improving the customized tourism recommendation algorithm model, which solves the technical shortcomings mentioned in the background technology.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a data processing method for improving a customized travel recommendation algorithm model, applied to a user interaction system of a mobile device, comprising the following steps:

[0009] S1. Collect behavioral data and travel preference data of user groups and build a user preference data set; collect environmental information including user geographic location, timestamp, weather, and social media activity in the user's area through the positioning module of the mobile device and external data interface, and build an environmental information data set;

[0010] Implement time window segmentation on the environmental information data set, and perform synchronous time alignment processing on the user preference data set of the user group to construct a time-series data set;

[0011] S2. Build a time-series user profile based on a time-series dataset. Extract contextual features that influence recommendation accuracy based on the environmental information dataset. Use a multi-layer perceptron neural network combined with an attention mechanism to model the weighted impact of users on different recommendation factors, while dynamically adjusting the recommendation strategy.

[0012] S3. Establish an interest classification mechanism to divide user interests into short-term interests, long-term interests, and context-triggered interests. Then adjust the priority of recommended content and optimize the order of recommendations. Finally, optimize the weight of user interests based on historical user feedback.

[0013] S4. Prioritize matching the short-term interest recommendation list, and then optimize the recommendation order by combining long-term interests and context-triggered interests; then build and combine a real-time behavior analysis model to monitor user interactions. When the recommended content does not meet user needs, trigger the intelligent interaction adjustment mechanism.

[0014] Preferably, S2 specifically includes:

[0015] The time-series dataset includes: representing the behavioral time series characteristics of the user group as dimension d1; representing the environmental context characteristics of the user group as dimension d2; representing the fusion feature dimension after timestamp alignment as F = d1 + d2; then extracting the timestamp information from the time-series dataset and dividing it with the behavior-related data to obtain the time window;

[0016] Based on time-series datasets, we extract user preference data sets and extract user behavior patterns based on time series modeling methods. We perform cluster analysis on user behavior characteristics within different time windows to form time-series user feature vectors. We also combine user geographic location, social media activity, weather and other environmental information to build a user time-series profile model.

[0017] A feature fusion model based on the attention mechanism is used to analyze the correlation between user behavior and contextual features, and generate user interest preference vectors in different scenarios; finally, the user behavior temporal features, context-aware modeling results and interest preference vectors are integrated into a user portrait feature vector, where each time window corresponds to a comprehensive representation that includes user behavior features, contextual features and interest preference vectors.

[0018] Preferably, S2 specifically further includes:

[0019] Extracting environmental information data sets, using a feature selection algorithm based on information gain to perform feature engineering on environmental information such as the user's area timestamp, weather conditions, and social media activity, extracting context-related features; and performing correlation analysis on the user's preference data sets to generate context-aware feature vectors.

[0020] The user portrait feature vector and the context-aware feature vector are input into the MLP model to extract high-dimensional feature representations. The user interest weight Ayh of the recommendation factors is calculated through the attention mechanism. The recommendation factors include attraction type, activity type, and time preference. The calculation formula of the user interest weight Ayh is as follows:

[0021] ;

[0022] Where F is the fusion feature dimension, which is the fusion representation of user profile features and context-aware features, and depends on the dimensions of user profile features and context features;

[0023] W f is the weight matrix of the MLP model, used to map the input feature F to the hidden layer; its value is an h×(d1+d2) matrix, where h is the dimension of the hidden layer;

[0024] bf is the bias term of the MLP model, which is used to adjust the output of the hidden layer; its value is an h×1 vector;

[0025] W a is the weight matrix of the attention mechanism, which is used to calculate the user's attention score for each recommendation factor. Its value is a k×h matrix, where k is the number of recommendation factors;

[0026] The Softmax function is used to normalize the attention score into a probability distribution; its output range is (0, 1], and the sum of all output values is 1;

[0027] It is a hyperbolic tangent activation function, which is used to introduce nonlinear transformation and has an output range of (-1, 1].

[0028] Preferably, S2 specifically further includes:

[0029] Real-time collection and analysis of multimodal data, including historical user data, text reviews, click behaviors, and social interactions, are performed. A user interest weight threshold A is set for each recommendation factor. This is then compared with the user interest weight Ayh to analyze the user's interest in various recommendation factors, including attraction type, activity type, and time preference. The specific evaluation content is as follows:

[0030] When the user interest weight Ayh i ≥User interest weight threshold A i , indicating that the user's interest in the current recommendation factor is "qualified",

[0031] When the user interest weight Ayh i <User interest weight threshold A i , indicating that the user’s interest value for the current recommendation factor is “unqualified”;

[0032] Among them, user interest weight Ayh i represents the user interest weight Ayh of the i-th recommendation factor, and the user interest weight threshold A i A represents the user interest weight threshold of the i-th recommendation factor; i is the category index of the recommendation factor;

[0033] Finally, based on the comparative evaluation results, the priority of the recommended content is dynamically adjusted:

[0034] For recommendation factors that meet the “qualified” interest criteria, increase the recommendation priority;

[0035] For recommendation factors that are “unqualified” for interest, lower the recommendation priority or exclude them.

[0036] Preferably, S3 specifically includes:

[0037] Read the user's historical behavior data and extract user interest features, including clicks, browsing, favorites, and reservations, from the historical behavior data and time-series user profiles. Then, divide user interests into three categories: short-term interests, long-term interests, and context-triggered interests.

[0038] For short-term interests, we identify and model them based on the user's behavioral data, including clicks and browsing history in the past week.

[0039] For long-term interests, we identify and model them based on behavioral data, including the user's continuous attention records over the past three months;

[0040] For situation-triggered interests, identification and modeling are performed based on contextual feature data including weather, holidays, and geographic location. A decision tree-based rule engine is used to classify user interests and assign initial weights to each type of interest. The classified user interests and the corresponding initial weights are output.

[0041] Preferably, S3 specifically further includes:

[0042] An interest decay model is established for each user interest type, specifically including:

[0043] For short-term interest, an exponential decay function is used Modeling is performed, where I0 represents the initial interest intensity, λ represents the decay rate, and t represents the time interval;

[0044] For long-term interests, a linear decay model was used for modeling;

[0045] For context-triggered interest, a context-aware model based on the attention mechanism is used to capture the correlation between user behavior and contextual features;

[0046] Next, based on the behavior-related data and time-series user portraits, the interest decay coefficient Bsj of each type of interest is calculated;

[0047] For short-term interests, the initial interest intensity I0 and decay rate λ are calculated based on the user's short-term behavior data;

[0048] For long-term interests, the initial interest intensity I0 and decay rate λ are calculated based on the user's long-term behavior data;

[0049] For context-triggered interest, the dynamic decay rate λ is calculated in combination with the context feature data. The specific calculation formula of the interest decay coefficient Bsj is as follows:

[0050] ;

[0051] In the formula, for short-term interests, the larger the decay rate λ is, the faster the interest decays;

[0052] For long-term interests, the smaller the decay rate λ is, the slower the interest decays;

[0053] For context-triggered interest, the decay rate λ is adjusted according to the timeliness of contextual features;

[0054] t represents the time interval, which indicates the length of time from the generation of interest to the current moment; e represents the base of the natural logarithm;

[0055] Among them, the value range of the interest attenuation coefficient Bsj is obtained through experimental data fitting and context dynamic adjustment. The effective value range of the interest attenuation coefficient Bsj is set to 0<Bsj≤1. When the interest attenuation coefficient Bsj is closer to 1, the degree of interest attenuation is greater. When the interest attenuation coefficient Bsj is closer to 0, the degree of interest attenuation is smaller.

[0056] Preferably, S3 specifically further includes:

[0057] When interest has waned, we use deep reinforcement learning to build interest-matching rules and adjust the priority and order of recommended content based on user feedback. This involves building interest-matching rules, designing reward functions, and optimizing interest weights.

[0058] Secondly, a reward function is designed to optimize the user interest weight based on the user satisfaction index;

[0059] Finally, the optimized user interest weight is fed back to the recommendation system to generate the next round of recommendation lists and monitor user interaction behavior in real time. When the recommended content does not meet user needs, the intelligent interaction adjustment mechanism is triggered to further optimize the user interest weight.

[0060] Preferably, S4 specifically includes:

[0061] Based on the user's time-series user profile, short-term interest features are extracted and a short-term interest recommendation list is constructed. Short-term interest features include statistical information on the user's recent clicks, browsing, collections, and booking records, as well as the changing trends of the user's interest preferences in different time intervals.

[0062] Then, we model the user's long-term interest characteristics, including calculating the long-term preference distribution based on the user's historical behavior data, and optimizing the recommended content based on the user's stable preference characteristics;

[0063] Secondly, the user's contextual interest features are extracted, including information such as geographic location, weather, and social activity in the current environmental information data set, and the user's contextual interest weight is calculated. The ranking strategy of recommended content is dynamically adjusted based on the contextual interest weight;

[0064] Finally, in the recommendation process, according to the hierarchical optimization strategy, we first match the short-term interest recommendation list, and optimize the recommendation order based on the weight calculation results of long-term interests and situational triggered interests. Specifically, we calculate the user's long-term interest preference distribution, and optimize the recommended content based on stable preference characteristics. Secondly, we extract situational triggered interest features including geographic location, weather, and social activity, and calculate the situational interest weight; finally, we dynamically adjust the recommendation sorting strategy based on the comprehensive weight of long-term interests and situational interests.

[0065] Preferably, S4 specifically further includes:

[0066] Build a user interest matching model, input including user historical behavior data, time-series user profiles and contextual feature data; use a reinforcement learning framework including deep learning rules to train the user interest matching model, and then calculate the recommended content matching degree Cnr through the user interest matching model. The specific calculation formula is:

[0067] ;

[0068] Where n represents the total number of interest categories, j represents the index of the interest category, that is, the different classification numbers of the user's interests; wj is the weight coefficient of the j-th interest category, Sj is the similarity between the recommended content and the j-th interest category, and f(t) is the timeliness correction factor.

[0069] Preferably, S4 specifically further includes:

[0070] A first matching evaluation threshold C1 and a second matching evaluation threshold C2 are preset and compared with the recommended content matching degree Cnr to evaluate the real-time matching degree between the recommended content and the user's interests, and respond to changes in user needs in real time; and the first matching evaluation threshold C1 is greater than the second matching evaluation threshold C2. The specific evaluation content is as follows:

[0071] When the recommended content matching degree Cnr ≥ the first matching evaluation threshold C1: it indicates that the recommended content is highly matched with the user's interests. The system directly recommends it without adjusting the recommendation strategy and monitors user interaction feedback data.

[0072] When the second matching evaluation threshold C2 ≤ recommended content matching Cnr < first matching evaluation threshold C1, it indicates that the recommended content does not match the user's interests, but is not lower than the second matching evaluation threshold C2 within the minimum acceptable range. At this time, the recommended content is optimized. Specific adjustment methods include:

[0073] (1) Increase the weight of short-term interests and adjust the ranking of recommended content;

[0074] (2) Integrate environmental factors to increase the proportion of personalized recommendations;

[0075] When the recommended content matching degree Cnr is less than the second matching degree evaluation threshold C2: it indicates that the recommended content does not match the user's interests and is lower than the minimum acceptable range of the second matching degree evaluation threshold C2. At this time, the intelligent interactive adjustment mechanism is triggered to immediately replace the recommended content and recalculate the recommended content matching degree Cnr;

[0076] After the recommended content is pushed, the system continuously collects user feedback data, including clicks, length of stay, favorites, and reservations, and calculates the feedback score; it uses the reinforcement learning model to adjust the recommendation strategy so that the recommendation system can adapt to the changing trends of user interests.

[0077] The present invention provides a data processing method for improving a customized tourism recommendation algorithm model. It has the following beneficial effects:

[0078] (1) This is a data processing method for improving the customized tourism recommendation algorithm model. To address the data sparsity problem, this technical solution achieves deep fusion of multidimensional data by constructing a time-series data set and a cross-domain knowledge graph; uses sliding window statistics and time-series clustering methods to extract local change characteristics and global laws of user behavior; uses a heterogeneous data co-expression mapping network to uniformly represent structured data and unstructured data; enhances the relevance of different modal information through cross-modal comparative learning; and finally fuses the user portrait feature vector with the context-aware feature vector, significantly improving data utilization efficiency and recommendation accuracy.

[0079] (2) This is a data processing method to improve the customized tourism recommendation algorithm model. To address the problem of lack of personalization, this solution designs a user interest weight calculation model based on the attention mechanism; extracts high-dimensional feature representation through the MLP model; uses the weight matrix W of the attention mechanism a Calculate the user's interest weight Ayh for each recommendation factor; combined with the preset threshold A i Conduct dynamic evaluation; establish three types of interest decay models; and achieve personalized optimization of recommendation strategies through interest matching rules driven by reinforcement learning;

[0080] (3) This is a data processing method for improving the customized tourism recommendation algorithm model. To address the problem of poor dynamic adaptability, this solution constructs a multi-layer asynchronous feedback data stream processing framework; uses time window segmentation technology to capture real-time behavior changes; dynamically adjusts recommendation parameters through an environmental perception interest regulation mechanism; sets dual matching thresholds to achieve intelligent interactive adjustment; and ultimately forms a complete adaptive system that includes real-time behavior analysis, dynamic weight adjustment, and closed-loop feedback optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 The present invention is a flowchart of the steps of a data processing method for improving a customized travel recommendation algorithm model. DETAILED DESCRIPTION

[0082] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0083] Example 1

[0084] See also Figure 1 The present invention provides a data processing method for improving a customized travel recommendation algorithm model, which is applied to a user interaction system of a mobile device, comprising the following steps:

[0085] S1. Collect behavioral data and travel preference data of user groups and build a user preference data set; collect environmental information including user geographic location, timestamp, weather, and social media activity in the user's area through the positioning module of the mobile device and external data interface, and build an environmental information data set;

[0086] Implement time window segmentation on the environmental information data set, and perform synchronous time alignment processing on the user preference data set of the user group to construct a time-series data set;

[0087] S2. Build a time-series user profile based on a time-series dataset. Extract contextual features that influence recommendation accuracy based on the environmental information dataset. Use a multi-layer perceptron neural network combined with an attention mechanism to model the weighted impact of users on different recommendation factors, while dynamically adjusting the recommendation strategy.

[0088] S3. Establish an interest classification mechanism to divide user interests into short-term interests, long-term interests, and context-triggered interests. Then adjust the priority of recommended content and optimize the order of recommendations. Finally, optimize the weight of user interests based on historical user feedback.

[0089] S4. Prioritize matching the short-term interest recommendation list, and then optimize the recommendation order by combining long-term interests and context-triggered interests; then build and combine a real-time behavior analysis model to monitor user interactions. When the recommended content does not meet user needs, trigger the intelligent interaction adjustment mechanism.

[0090] In this embodiment, S1 specifically includes:

[0091] The extracted user behavior and travel preference data, including user click, browse, favorite, and booking records, are pre-processed to construct a user preference data set. At the same time, environmental information is collected through the positioning module of the mobile device and the external data interface, and pre-processed to construct an environmental information data set.

[0092] Furthermore, the user behavior and travel preference related data and environmental information are segmented according to preset time intervals. First, the user behavior and travel preference related data and environmental information are timestamp-aligned and segmented according to preset time intervals, including hourly, daily or weekly. Then, the data in each time interval are aggregated and feature extracted, including calculating the statistical values of the user's behavior frequency, preference changes and environmental characteristics in each time interval. Finally, the segmented data is stored as a time-series dataset.

[0093] Example 2

[0094] S2 specifically includes:

[0095] The time-series dataset includes: representing the behavioral time series characteristics of the user group as dimension d1; representing the environmental context characteristics of the user group as dimension d2; representing the fusion feature dimension after timestamp alignment as F = d1 + d2; then extracting the timestamp information from the time-series dataset and dividing it with the behavior-related data to obtain the time window;

[0096] Based on time-series datasets, we extract user preference data sets and extract user behavior patterns based on time series modeling methods. We perform cluster analysis on user behavior characteristics within different time windows to form time-series user feature vectors. We also combine user geographic location, social media activity, weather and other environmental information to build a user time-series profile model.

[0097] A feature fusion model based on the attention mechanism is used to analyze the correlation between user behavior and contextual features, and generate user interest preference vectors in different scenarios; finally, the user behavior temporal features, context-aware modeling results and interest preference vectors are integrated into a user portrait feature vector, where each time window corresponds to a comprehensive representation that includes user behavior features, contextual features and interest preference vectors.

[0098] S2 specifically includes:

[0099] Extracting environmental information data sets, using a feature selection algorithm based on information gain to perform feature engineering on environmental information such as the user's area timestamp, weather conditions, and social media activity, extracting context-related features; and performing correlation analysis on the user's preference data sets to generate context-aware feature vectors.

[0100] The user portrait feature vector and the context-aware feature vector are input into the MLP model to extract high-dimensional feature representations. The user interest weight Ayh of the recommendation factors is calculated through the attention mechanism. The recommendation factors include attraction type, activity type, and time preference. The calculation formula of the user interest weight Ayh is as follows:

[0101] ;

[0102] Where F is the fusion feature dimension, which is the fusion representation of user profile features and context-aware features, and depends on the dimensions of user profile features and context features;

[0103] W f is the weight matrix of the MLP model, used to map the input feature F to the hidden layer; its value is an h×(d1+d2) matrix, where h is the dimension of the hidden layer;

[0104] bf is the bias term of the MLP model, which is used to adjust the output of the hidden layer; its value is an h×1 vector;

[0105] W a is the weight matrix of the attention mechanism, which is used to calculate the user's attention score for each recommendation factor. Its value is a k×h matrix, where k is the number of recommendation factors;

[0106] The Softmax function is used to normalize the attention score into a probability distribution; its output range is (0, 1], and the sum of all output values is 1;

[0107] It is a hyperbolic tangent activation function, which is used to introduce nonlinear transformation and has an output range of (-1, 1].

[0108] S2 specifically includes:

[0109] Real-time collection and analysis of multimodal data, including historical user data, text reviews, click behaviors, and social interactions, are performed. A user interest weight threshold A is set for each recommendation factor. This is then compared with the user interest weight Ayh to analyze the user's interest in various recommendation factors, including attraction type, activity type, and time preference. The specific evaluation content is as follows:

[0110] When the user interest weight Ayh i≥User interest weight threshold A i , indicating that the user's interest in the current recommendation factor is "qualified",

[0111] When the user interest weight Ayh i <User interest weight threshold A i , indicating that the user’s interest value for the current recommendation factor is “unqualified”;

[0112] Among them, user interest weight Ayh i represents the user interest weight Ayh of the i-th recommendation factor, and the user interest weight threshold A i A represents the user interest weight threshold of the i-th recommendation factor; i is the category index of the recommendation factor;

[0113] Finally, based on the comparative evaluation results, the priority of the recommended content is dynamically adjusted:

[0114] For recommendation factors that meet the “qualified” interest criteria, increase the recommendation priority;

[0115] For recommendation factors that are “unqualified” for interest, lower the recommendation priority or exclude them.

[0116] In this embodiment, the intelligence level of the travel recommendation system is significantly improved through time series data set processing and multi-dimensional feature analysis. The local characteristics of user behavior are accurately captured based on time window segmentation technology (preset intervals are 1 hour / 24 hours / 7 days) and sliding window statistics (mean / variance / maximum / minimum). Global behavior patterns are extracted through time series clustering analysis (k-means cluster number n=5).

[0117] Combined with the environmental context features (dimension d2) to build the attention mechanism feature fusion model (weight matrix W a Size k×h); use the MLP model (hidden layer dimension h=64) to generate high-dimensional feature representation (input dimension d1+d2);

[0118] The user interest weight Ayh is calculated by the formula (including the weight matrix W f Size h×(d1+d2), bias term bf size h×1) quantify recommendation factor preference; set dynamic threshold Ai for interest qualification evaluation (when Ayh i ≥Ai); finally, the behavioral time series features, context perception vectors, and interest preference vectors are integrated to form a comprehensive user portrait, where the sliding window size of the behavioral time series features is w=7, the information gain threshold of the context perception vector is IG>0.3, and the activation output range of the interest preference vector tanh is (-1, 1]); this solution realizes a complete closed-loop optimization from raw data to personalized recommendations, enabling the system to have the core capabilities of capturing interest changes in real time, dynamically adjusting recommendation strategies, and accurately predicting user needs; among them, The calculation formula is: , where x represents the input value, that is, the value that needs to be nonlinearly transformed;

[0119] Example 3

[0120] S3 specifically includes:

[0121] Read the user's historical behavior data and extract user interest features, including clicks, browsing, favorites, and reservations, from the historical behavior data and time-series user profiles. Then, divide user interests into three categories: short-term interests, long-term interests, and context-triggered interests.

[0122] For short-term interests, we identify and model them based on the user's behavioral data, including clicks and browsing history in the past week.

[0123] For long-term interests, we identify and model them based on behavioral data, including the user's continuous attention records over the past three months;

[0124] For situation-triggered interests, identification and modeling are performed based on contextual feature data including weather, holidays, and geographic location. A decision tree-based rule engine is used to classify user interests and assign initial weights to each type of interest. The classified user interests and the corresponding initial weights are output.

[0125] S3 specifically includes:

[0126] An interest decay model is established for each user interest type, specifically including:

[0127] For short-term interest, an exponential decay function is used Modeling is performed, where I0 represents the initial interest intensity, λ represents the decay rate, and t represents the time interval;

[0128] For long-term interests, a linear decay model was used for modeling;

[0129] For context-triggered interest, a context-aware model based on the attention mechanism is used to capture the correlation between user behavior and contextual features;

[0130] Next, based on the behavior-related data and time-series user portraits, the interest decay coefficient Bsj of each type of interest is calculated;

[0131] For short-term interests, the initial interest intensity I0 and decay rate λ are calculated based on the user's short-term behavior data;

[0132] For long-term interests, the initial interest intensity I0 and decay rate λ are calculated based on the user's long-term behavior data;

[0133] For context-triggered interest, the dynamic decay rate λ is calculated in combination with the context feature data. The specific calculation formula of the interest decay coefficient Bsj is as follows:

[0134] ;

[0135] In the formula, for short-term interests, the larger the decay rate λ is, the faster the interest decays;

[0136] For long-term interests, the smaller the decay rate λ is, the slower the interest decays;

[0137] For context-triggered interest, the decay rate λ is adjusted according to the timeliness of contextual features;

[0138] t represents the time interval, which indicates the length of time from the generation of interest to the current moment; e represents the base of the natural logarithm;

[0139] Among them, the value range of the interest attenuation coefficient Bsj is obtained through experimental data fitting and context dynamic adjustment. The effective value range of the interest attenuation coefficient Bsj is set to 0<Bsj≤1. When the interest attenuation coefficient Bsj is closer to 1, the degree of interest attenuation is greater. When the interest attenuation coefficient Bsj is closer to 0, the degree of interest attenuation is smaller.

[0140] S3 specifically includes:

[0141] When interest has waned, we use deep reinforcement learning to build interest-matching rules and adjust the priority and order of recommended content based on user feedback. This involves building interest-matching rules, designing reward functions, and optimizing interest weights.

[0142] Secondly, a reward function is designed to optimize the user interest weight based on the user satisfaction index;

[0143] Finally, the optimized user interest weight is fed back to the recommendation system to generate the next round of recommendation lists and monitor user interaction behavior in real time. When the recommended content does not meet user needs, the intelligent interaction adjustment mechanism is triggered to further optimize the user interest weight.

[0144] In this embodiment, by reading the user's historical behavior data and time-series user portraits, user interest features are extracted and divided into three categories: short-term interests, long-term interests and situation-triggered interests, so as to achieve precise interest modeling; among them, short-term interests are identified and modeled through the user's click, browse and other behavioral data in the past week, reflecting the user's immediate interest changes; long-term interests are based on the user's continuous attention records in the past few months, describing the user's stable preference trends; situation-triggered interests combine contextual feature data such as weather, holidays and geographical locations to capture the user's dynamic interest changes affected by environmental factors; a rule engine is used to classify user interests and assign initial weights to optimize the rationality of interest weights; further, an interest decay model is established for each type of interest, wherein short-term interests are modeled using an exponential decay function, and the initial interest strength I0 and decay rate λ are used to calculate the interest decay trend over time; long-term interests are modeled using a linear decay model to ensure the smooth decay of users' long-term attention points; situation-triggered interests adopt a context-aware model based on an attention mechanism, The dynamic decay rate λ is calculated in combination with contextual feature data to reflect the impact of environmental changes on user interests; the interest decay coefficient Bsj is calculated based on behavior-related data and time-series user portraits to ensure that the recommendation system accurately perceives changes in interest, where the value range of the interest decay coefficient Bsj is 0<Bsj≤1. When Bsj=1, it means that the interest has not decayed, and when Bsj is close to 0, it means that the interest has decayed; when the user interest has decayed, a deep reinforcement learning algorithm is used to construct interest matching rules, and the priority and order of recommended content are dynamically adjusted according to user feedback to ensure the adaptability and accuracy of the recommended content; further, a reward function is designed to optimize the user interest weight based on the user satisfaction index, so that the system can adapt to changes in user needs; the optimized user interest weight will be fed back to the recommendation system to generate the next round of recommendation lists, and user interaction behavior will be monitored in real time. When the recommended content does not meet user needs, the intelligent interaction adjustment mechanism is triggered to further optimize the user interest weight, thereby improving the long-term intelligence level and user satisfaction of the recommendation system.

[0145] Example 4

[0146] S4 specifically includes:

[0147] Based on the user's time-series user profile, short-term interest features are extracted and a short-term interest recommendation list is constructed. Short-term interest features include statistical information on the user's recent clicks, browsing, collections, and booking records, as well as the changing trends of the user's interest preferences in different time intervals.

[0148] Then, we model the user's long-term interest characteristics, including calculating the long-term preference distribution based on the user's historical behavior data, and optimizing the recommended content based on the user's stable preference characteristics;

[0149] Secondly, the user's contextual interest features are extracted, including information such as geographic location, weather, and social activity in the current environmental information data set, and the user's contextual interest weight is calculated. The ranking strategy of recommended content is dynamically adjusted based on the contextual interest weight;

[0150] Finally, in the recommendation process, according to the hierarchical optimization strategy, we first match the short-term interest recommendation list, and optimize the recommendation order based on the weight calculation results of long-term interests and situational triggered interests. Specifically, we calculate the user's long-term interest preference distribution, and optimize the recommended content based on stable preference characteristics. Secondly, we extract situational triggered interest features including geographic location, weather, and social activity, and calculate the situational interest weight; finally, we dynamically adjust the recommendation sorting strategy based on the comprehensive weight of long-term interests and situational interests.

[0151] S4 specifically includes:

[0152] Build a user interest matching model, input including user historical behavior data, time-series user profiles and contextual feature data; use a reinforcement learning framework including deep learning rules to train the user interest matching model, and then calculate the recommended content matching degree Cnr through the user interest matching model. The specific calculation formula is:

[0153] ;

[0154] Where n represents the total number of interest categories, j represents the index of the interest category, that is, the different classification numbers of the user's interests; wj is the weight coefficient of the j-th interest category, Sj is the similarity between the recommended content and the j-th interest category, and f(t) is the timeliness correction factor.

[0155] S4 specifically includes:

[0156] A first matching evaluation threshold C1 and a second matching evaluation threshold C2 are preset and compared with the recommended content matching degree Cnr to evaluate the real-time matching degree between the recommended content and the user's interests, and respond to changes in user needs in real time; and the first matching evaluation threshold C1 is greater than the second matching evaluation threshold C2. The specific evaluation content is as follows:

[0157] When the recommended content matching degree Cnr ≥ the first matching evaluation threshold C1: it indicates that the recommended content is highly matched with the user's interests. The system directly recommends it without adjusting the recommendation strategy and monitors user interaction feedback data.

[0158] When the second matching evaluation threshold C2 ≤ recommended content matching Cnr < first matching evaluation threshold C1, it indicates that the recommended content does not match the user's interests, but is not lower than the second matching evaluation threshold C2 within the minimum acceptable range. At this time, the recommended content is optimized. Specific adjustment methods include:

[0159] (1) Increase the weight of short-term interests and adjust the ranking of recommended content;

[0160] (2) Integrate environmental factors to increase the proportion of personalized recommendations;

[0161] When the recommended content matching degree Cnr is less than the second matching degree evaluation threshold C2: it indicates that the recommended content does not match the user's interests and is lower than the minimum acceptable range of the second matching degree evaluation threshold C2. At this time, the intelligent interactive adjustment mechanism is triggered to immediately replace the recommended content and recalculate the recommended content matching degree Cnr;

[0162] After the recommended content is pushed, the system continuously collects user feedback data, including clicks, length of stay, favorites, and reservations, and calculates the feedback score; it uses the reinforcement learning model to adjust the recommendation strategy so that the recommendation system can adapt to the changing trends of user interests.

[0163] In this embodiment, this step realizes a comprehensive evaluation of the user's short-term interest characteristics, long-term interest characteristics and context-triggered interest characteristics by constructing a user interest matching model, thereby optimizing the recommended content matching degree and calculating the recommended content matching coefficient Cnr to ensure the accuracy of the recommended content; among them, the short-term interest characteristics are extracted through the statistical information of the user's recent click, browse, favorite and booking records, combined with the interest preference change trends in different time intervals, reflecting the user's recent interest fluctuations; the long-term interest characteristics calculate the long-term preference distribution based on the user's historical behavior data, and optimize the recommended content based on the user's stable preference characteristics to ensure the continued relevance of the recommendation; the context-triggered interest characteristics calculate the user's contextual interest weight based on the current environmental information data set, such as geographic location, weather, social activity, etc., and dynamically adjust the recommended content sorting strategy based on the contextual interest weight to enhance the real-time adaptability of the recommendation;

[0164] Subsequently, the user interest matching model is trained through a reinforcement learning framework, and a recommended content matching evaluation mechanism is used for real-time optimization. The first matching evaluation threshold C1 and the second matching evaluation threshold C2 are used to evaluate the rationality of the recommended content matching coefficient Cnr. C1 is set as the higher matching standard, and C2 is set as the lowest acceptable range.

[0165] When the recommended content matching coefficient Cnr ≥ the first matching evaluation threshold C1, direct recommendations are made and user interaction feedback data is monitored to improve the system's learning ability. When the second matching evaluation threshold C2 ≤ the recommended content matching coefficient Cnr < the first matching evaluation threshold C1, the short-term interest weight is increased, and the cosine similarity between the recommended content and the user's interest vector is recalculated. A new recommendation list is generated based on the updated scores in descending order, the recommended content ranking is adjusted, or the proportion of personalized recommended content is increased based on environmental factors to optimize the recommended content.

[0166] When the recommended content matching coefficient Cnr is less than the second matching evaluation threshold C2, the intelligent interactive adjustment mechanism is triggered, the recommended content is immediately replaced and the recommended content matching coefficient Cnr is recalculated to ensure the adaptability and accuracy of the recommended content; finally, the feedback score is calculated through user feedback data, including clicks, stay time, collections and bookings, and the reinforcement learning model is used to dynamically adjust the recommendation strategy, so that the system can adapt to the changing trends of user interests and improve the personalization ability and long-term effectiveness of the recommendation system.

[0167] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data processing method for improving a customized travel recommendation algorithm model, applied to a user interaction system of a mobile device, characterized by: The following steps are involved: S1. Collect behavioral data and travel preference data of user groups and build a user preference data set; collect environmental information including user geographic location, timestamp, weather, and social media activity in the user's area through the positioning module of the mobile device and external data interface, and build an environmental information data set; Implement time window segmentation on the environmental information data set, and perform synchronous time alignment processing on the user preference data set of the user group to construct a time-series data set; S2. Build a time-series user profile based on the time-series dataset; Based on the environmental information data set, we extract environmental context features that affect recommendation accuracy, and use a multi-layer perceptron neural network combined with an attention mechanism to model the weighted influence of users on different recommendation factors, while dynamically adjusting the recommendation strategy. S3. Establish an interest classification mechanism to divide user interests into short-term interests, long-term interests, and context-triggered interests. Then adjust the priority of recommended content and optimize the order of recommendations. Finally, optimize the weight of user interests based on historical user feedback. S4. Prioritize matching the short-term interest recommendation list, and then optimize the recommendation order by combining long-term interests and context-triggered interests; then build and combine a real-time behavior analysis model to monitor user interactions. When the recommended content does not meet user needs, trigger the intelligent interaction adjustment mechanism.

2. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S2 specifically includes: The time-series dataset includes: representing the behavioral time series characteristics of the user group as dimension d1; representing the environmental context characteristics of the user group as dimension d2; representing the fusion feature dimension after timestamp alignment as F = d1 + d2; then extracting the timestamp information from the time-series dataset and dividing it with the behavior-related data to obtain the time window; Based on time-series datasets, we extract user preference data sets and extract user behavior patterns based on time series modeling methods. We perform cluster analysis on user behavior characteristics within different time windows to form time-series user feature vectors. We also combine user geographic location, social media activity, weather and other environmental information to build a user time-series profile model. A feature fusion model based on the attention mechanism is used to analyze the correlation between user behavior and contextual features, and generate user interest preference vectors in different scenarios; finally, the user behavior temporal features, context-aware modeling results and interest preference vectors are integrated into a user portrait feature vector, where each time window corresponds to a comprehensive representation that includes user behavior features, contextual features and interest preference vectors.

3. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S2 specifically includes: Extracting environmental information data sets, using a feature selection algorithm based on information gain to perform feature engineering on environmental information such as the user's area timestamp, weather conditions, and social media activity, extracting context-related features; and performing correlation analysis on the user's preference data sets to generate context-aware feature vectors. The user portrait feature vector and the context-aware feature vector are input into the MLP model to extract high-dimensional feature representations. The user interest weight Ayh of the recommendation factors is calculated through the attention mechanism. The recommendation factors include attraction type, activity type, and time preference. The calculation formula of the user interest weight Ayh is as follows: ； Where F is the fusion feature dimension, which is the fusion representation of user profile features and context-aware features, and depends on the dimensions of user profile features and context features; W f is the weight matrix of the MLP model, used to map the input feature F to the hidden layer; its value is an h×(d1+d2) matrix, where h is the dimension of the hidden layer; bf is the bias term of the MLP model, which is used to adjust the output of the hidden layer; its value is an h×1 vector; W a is the weight matrix of the attention mechanism, which is used to calculate the user's attention score for each recommendation factor. Its value is a k×h matrix, where k is the number of recommendation factors; The Softmax function is used to normalize the attention score into a probability distribution; its output range is (0, 1], and the sum of all output values is 1; It is a hyperbolic tangent activation function, which is used to introduce nonlinear transformation and has an output range of (−1, 1].

4. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S2 specifically includes: Real-time collection and analysis of multimodal data, including historical user data, text reviews, click behaviors, and social interactions, are performed. A user interest weight threshold A is set for each recommendation factor. This is then compared with the user interest weight Ayh to analyze the user's interest in various recommendation factors, including attraction type, activity type, and time preference. The specific evaluation content is as follows: When the user interest weight Ayh i ≥User interest weight threshold A i , indicating that the user's interest in the current recommendation factor is "qualified", When the user interest weight Ayh i <User interest weight threshold A i , indicating that the user's interest value for the current recommendation factor is "unqualified"; Among them, user interest weight Ayh i represents the user interest weight Ayh of the i-th recommendation factor, and the user interest weight threshold A i A represents the user interest weight threshold of the i-th recommendation factor; i is the category index of the recommendation factor; Finally, based on the comparative evaluation results, the priority of the recommended content is dynamically adjusted: For recommendation factors that meet the "qualified" interest criteria, increase the recommendation priority; For recommendation factors that are "unqualified" for interest, lower the recommendation priority or exclude them.

5. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S3 specifically includes: Read the user's historical behavior data and extract user interest features, including clicks, browsing, favorites, and reservations, from the historical behavior data and time-series user profiles. Then, divide user interests into three categories: short-term interests, long-term interests, and context-triggered interests. For short-term interests, we identify and model them based on the user's behavioral data, including clicks and browsing history in the past week. For long-term interests, we identify and model them based on behavioral data, including the user's continuous attention records over the past three months; For situation-triggered interests, identification and modeling are performed based on contextual feature data including weather, holidays, and geographic location. A decision tree-based rule engine is used to classify user interests and assign initial weights to each type of interest. The classified user interests and the corresponding initial weights are output.

6. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S3 specifically includes: An interest decay model is established for each user interest type, specifically including: For short-term interest, an exponential decay function is used Modeling is performed, where I0 represents the initial interest intensity, λ represents the decay rate, and t represents the time interval; For long-term interests, a linear decay model was used for modeling; For context-triggered interest, a context-aware model based on the attention mechanism is used to capture the correlation between user behavior and contextual features; Next, based on the behavior-related data and time-series user portraits, the interest decay coefficient Bsj of each type of interest is calculated; For short-term interests, the initial interest intensity I0 and decay rate λ are calculated based on the user's short-term behavior data; For long-term interests, the initial interest intensity I0 and decay rate λ are calculated based on the user's long-term behavior data; For context-triggered interest, the dynamic decay rate λ is calculated in combination with the context feature data. The specific calculation formula of the interest decay coefficient Bsj is as follows: ； In the formula, for short-term interests, the larger the decay rate λ is, the faster the interest decays; For long-term interests, the smaller the decay rate λ is, the slower the interest decays; For context-triggered interest, the decay rate λ is adjusted according to the timeliness of contextual features; t represents the time interval, which indicates the length of time from the generation of interest to the current moment; e represents the base of the natural logarithm; Among them, the value range of the interest attenuation coefficient Bsj is obtained through experimental data fitting and context dynamic adjustment. The effective value range of the interest attenuation coefficient Bsj is set to 0<Bsj≤1. When the interest attenuation coefficient Bsj is closer to 1, the degree of interest attenuation is greater. When the interest attenuation coefficient Bsj is closer to 0, the degree of interest attenuation is smaller.

7. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S3 specifically includes: When interest has waned, we use deep reinforcement learning to build interest-matching rules and adjust the priority and order of recommended content based on user feedback. This involves building interest-matching rules, designing reward functions, and optimizing interest weights. Secondly, a reward function is designed to optimize the user interest weight based on the user satisfaction index; Finally, the optimized user interest weight is fed back to the recommendation system to generate the next round of recommendation lists and monitor user interaction behavior in real time. When the recommended content does not meet user needs, the intelligent interaction adjustment mechanism is triggered to further optimize the user interest weight.

8. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S4 specifically includes: Based on the user's time-series user profile, short-term interest features are extracted and a short-term interest recommendation list is constructed. Short-term interest features include statistical information on the user's recent clicks, browsing, collections, and booking records, as well as the changing trends of the user's interest preferences in different time intervals. Then, we model the user's long-term interest characteristics, including calculating the long-term preference distribution based on the user's historical behavior data, and optimizing the recommended content based on the user's stable preference characteristics; Secondly, the user's contextual interest features are extracted, including information such as geographic location, weather, and social activity in the current environmental information data set, and the user's contextual interest weight is calculated. The ranking strategy of recommended content is dynamically adjusted based on the contextual interest weight; Finally, in the recommendation process, according to the hierarchical optimization strategy, we first match the short-term interest recommendation list, and optimize the recommendation order based on the weight calculation results of long-term interests and situational triggered interests. Specifically, we calculate the user's long-term interest preference distribution, and optimize the recommended content based on stable preference characteristics. Secondly, we extract situational triggered interest features including geographic location, weather, and social activity, and calculate the situational interest weight; finally, we dynamically adjust the recommendation sorting strategy based on the comprehensive weight of long-term interests and situational interests.

9. The data processing method for improving the customized travel recommendation algorithm model according to claim 1, characterized in that: S4 specifically includes: Build a user interest matching model, input including user historical behavior data, time-series user profiles and contextual feature data; use a reinforcement learning framework including deep learning rules to train the user interest matching model, and then calculate the recommended content matching degree Cnr through the user interest matching model. The specific calculation formula is: ； Where n represents the total number of interest categories, j represents the index of the interest category, that is, the different classification numbers of the user's interests; wj is the weight coefficient of the j-th interest category, Sj is the similarity between the recommended content and the j-th interest category, and f(t) is the timeliness correction factor.

10. The data processing method for improving the customized tourism recommendation algorithm model according to claim 1, characterized in that: S4 specifically includes: A first matching evaluation threshold C1 and a second matching evaluation threshold C2 are preset and compared with the recommended content matching degree Cnr to evaluate the real-time matching degree between the recommended content and the user's interests, and respond to changes in user needs in real time; and the first matching evaluation threshold C1 is greater than the second matching evaluation threshold C2. The specific evaluation content is as follows: When the recommended content matching degree Cnr ≥ the first matching evaluation threshold C1: it indicates that the recommended content is highly matched with the user's interests. The system directly recommends it without adjusting the recommendation strategy and monitors user interaction feedback data. When the second matching evaluation threshold C2 ≤ recommended content matching Cnr < first matching evaluation threshold C1, it indicates that the recommended content does not match the user's interests, but is not lower than the second matching evaluation threshold C2 within the minimum acceptable range. At this time, the recommended content is optimized. Specific adjustment methods include: Increase the weight of short-term interests and adjust the ranking of recommended content; Combine environmental factors to increase the proportion of personalized recommendations; When the recommended content matching degree Cnr is less than the second matching degree evaluation threshold C2: it indicates that the recommended content does not match the user's interests and is lower than the minimum acceptable range of the second matching degree evaluation threshold C2. At this time, the intelligent interactive adjustment mechanism is triggered to immediately replace the recommended content and recalculate the recommended content matching degree Cnr; After the recommended content is pushed, the system continuously collects user feedback data, including clicks, length of stay, favorites, and reservations, and calculates the feedback score; it uses the reinforcement learning model to adjust the recommendation strategy so that the recommendation system can adapt to the changing trends of user interests.

Images