Information pushing method and device, equipment, medium and program product

Abstract

The invention provides an information pushing method and device, equipment, a medium and a program product, and relates to the technical field of big data, and the method comprises the steps: carrying out the clustering analysis of the historical travel information of at least one user, obtaining the travel preference characteristics of at least one user, and obtaining the travel score of at least one user based on the travel preference characteristics; based on the tourism beforehand behavior characteristics, obtaining a beforehand behavior score of at least one user; inputting the travel scores and the beforehand behavior scores of a plurality of users in the at least one user into a logistic regression model for prediction to obtain travel types of the plurality of users; the feature similarity between the travel types of the multiple users and each promotion short message template in the multiple promotion short message templates is obtained, the promotion short message template with the highest similarity is selected as a target template, push information corresponding to the target template is pushed to the multiple users, and each promotion short message template corresponds to at least one travel type. Through the method, the matching degree of the push result to the travel demand of the user can be improved.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to an information push method, apparatus, device, medium and program product. Background Technology

[0002] Recommendation models for tourism marketing content play a crucial role in today's tourism industry. These models can provide personalized tourism products and experiences based on users' historical behavior, preferences, and feedback, meeting their unique needs. By analyzing user feedback and preferences, tourism companies can innovate tourism products and services, meet emerging market demands, quickly respond to market changes and user needs, and adjust recommendation strategies in a timely manner. However, traditional recommendation models are mostly based on cluster analysis and collaborative filtering using users' historical preferences. This analysis focuses only on the similarity between users or features, neglecting the dynamic changes in user interests and characteristics, resulting in a low degree of matching between the recommended results and the user's current tourism needs. Summary of the Invention

[0003] This application provides an information push method and apparatus that can solve the problems of one-sided and low accuracy in information push results.

[0004] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide an information push method, the method comprising: Cluster analysis is performed on the historical travel information of at least one user to obtain the travel preference characteristics of the at least one user, and the travel score of the at least one user is obtained based on the travel preference characteristics; The pre-trip behavior score of at least one user is obtained based on pre-trip behavior characteristics. The travel scores and pre-trip behavior scores of multiple users among the at least one user are input into a logistic regression model for prediction to obtain the travel type of the multiple users. The similarity between the travel type of the multiple users and the features of each promotional SMS template among the multiple promotional SMS templates is obtained. The promotional SMS template with the highest similarity is selected as the target template, and the push information corresponding to the target template is pushed to the multiple users. Each of the promotional SMS templates corresponds to at least one travel type.

[0005] Secondly, embodiments of this application provide an information push device, the device comprising: The analysis module is used to perform cluster analysis on the historical travel information of at least one user to obtain the travel preference characteristics of the at least one user, and to obtain the travel score of the at least one user based on the travel preference characteristics; The acquisition module is used to acquire the pre-trip behavior score of the at least one user based on pre-trip behavior characteristics. The prediction module is used to input the travel scores and pre-trip behavior scores of multiple users among the at least one user into a logistic regression model for prediction, so as to obtain the travel type of the multiple users; The push module is used to obtain the feature similarity between the travel type of the multiple users and each of the multiple promotional SMS templates, select the promotional SMS template with the highest similarity as the target template, and push the push information corresponding to the target template to the multiple users, wherein each of the promotional SMS templates corresponds to at least one travel type.

[0006] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory, wherein the memory stores a program or instructions that can run on the processor, and when the program or instructions are executed by the processor, they implement the steps of the information push method as described in the first aspect.

[0007] Fourthly, embodiments of this application provide a readable storage medium storing a program or instructions, which, when executed by a processor, implement the steps of the information push method as described in the first aspect.

[0008] In this embodiment, by clustering analysis of the historical travel information of at least one user, the travel preference characteristics of the at least one user are obtained, and a travel score is obtained based on the travel preference characteristics. A pre-trip behavior score is obtained based on the pre-trip behavior characteristics of the at least one user. The travel scores and pre-trip behavior scores of multiple users are input into a logistic regression model for prediction to obtain the travel types of the multiple users. The feature similarity between the travel types of the multiple users and each of the multiple promotional SMS templates is obtained, and the promotional SMS template with the highest similarity is selected as the target template. The push information corresponding to the target template is pushed to the multiple users, wherein each promotional SMS template corresponds to at least one travel type. Thus, by extracting travel preference characteristics and calculating travel scores through clustering analysis of users' historical travel information, combining pre-trip behavior scores with pre-trip behavior characteristics, and then using a logistic regression model to predict travel types and match the optimal promotional SMS template for push information, a relatively accurate push of travel marketing information is achieved. This approach integrates historical preference and current behavior data to improve the matching degree of user needs, and enhances the targeting of pushes by relying on message-adapted template pushes, effectively improving the matching degree of push results with users' current travel needs. Attached Figure Description

[0009] Figure 1 One of the flowcharts illustrating the information push method provided in this application embodiment; Figure 2 A second schematic flowchart illustrating the information push method provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of the information push device provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The following description, in conjunction with the accompanying drawings, further illustrates the information push method, apparatus, device, and storage medium proposed in the embodiments of the application.

[0012] Please see Figure 1 , Figure 1 A flowchart illustrating an information push method provided in this application embodiment is shown in the figure. The method includes: Step 110: Perform cluster analysis on the historical travel information of at least one user to obtain the travel preference characteristics of the at least one user, and obtain the travel score of the at least one user based on the travel preference characteristics; The above-mentioned at least one user can be understood as a set of objects that need to be analyzed for information push. Its scope can be determined according to actual needs. It can be a group of users with travel records in a certain administrative region, or a group of users who have registered and authorized the provision of travel data on a certain service platform; it can be a single independent user, or a group of users with similar basic attributes (such as age and occupation).

[0013] The aforementioned historical travel information can be understood as various types of travel-related data of a user over a past period. These data can come from the user's past travel service booking records, such as hotel check-in records, attraction ticket purchase vouchers, and tour booking information; they can also come from the user's transportation data, such as high-speed rail tickets, plane tickets, driving navigation route trajectories, and public transportation card swipe records; and they can also come from travel-related content shared by the user on social media platforms, such as travel guide posting records and travel photo location information.

[0014] The clustering analysis described above can be understood as a data mining method that divides data objects into different groups based on data similarity. In this method, clustering analysis can group users with similar travel habits and preferences into the same category, so as to extract common travel preference features later. The algorithm used for clustering analysis can be K-means clustering, which achieves grouping by pre-setting the number of clusters and iteratively optimizing the cluster centers; or it can be hierarchical clustering, which constructs a hierarchical clustering structure by gradually merging or splitting data groups, and finally obtains the clustering results that meet the requirements.

[0015] The aforementioned travel preference characteristics can be understood as specific attributes that reflect a user's tendencies and preferences in travel activities. These characteristics can include the type of tour group, such as whether the user prefers to travel alone, with family, with friends, or in a group; the frequency of travel, such as the number of trips per month, per quarter, or per year; the preference for travel destinations, such as whether the user prefers natural scenic destinations like mountains and lakes, historical and cultural destinations like ancient cities and towns, entertainment and experience destinations like theme parks, or leisure and vacation destinations like beaches; in addition, the preference for travel duration, such as whether the user prefers short day trips, two-day weekend trips, multi-day trips during short holidays, or long-distance trips during long holidays.

[0016] The aforementioned travel score can be understood as a numerical value calculated based on the user's travel preference characteristics, used to quantify the degree of the user's travel preference. This value allows for a direct comparison of the differences in travel preferences among different users, providing data support for subsequent travel type prediction.

[0017] Step 120: Obtain the pre-trip behavior score of the at least one user based on pre-trip behavior characteristics; In this step, you can refer to Figure 2The aforementioned pre-trip behavioral characteristics can be understood as various behavioral attributes related to travel intentions exhibited by users before they may engage in travel behavior. These characteristics reflect users' potential travel needs and are important bases for judging the strength of users' travel intentions. Pre-trip behavioral characteristics can include users' online behavior, such as the frequency and duration of users browsing travel guides, travel product introductions, and destination weather and transportation information online, and records of searching for keywords such as "nearby travel recommendations" and "long holiday travel destinations"; it can also include users' commuting frequency, that is, the frequency of users' daily commute to and from work or school. A lower commuting frequency may mean that users have more free time for travel; it can also include users' number of working days, that is, the actual number of working days users have in a period of time. Fewer working days may indicate that users have more free time to plan travel; in addition, it can also include the timing of holidays, that is, the pattern of changes in users' behavior before and after holidays, such as whether users pay more attention to travel-related information before holidays.

[0018] The pre-trip behavior score can be understood as a numerical value calculated based on the user's pre-trip behavior characteristics, used to quantify the strength of the user's potential travel intention. The higher the pre-trip behavior score, the greater the likelihood that the user will engage in travel behavior, providing a reference for subsequent screening of users to be predicted and determining the type of travel.

[0019] Step 130: Input the travel scores and pre-trip behavior scores of multiple users among the at least one user into a logistic regression model for prediction to obtain the travel type of the multiple users; In this step, the aforementioned multiple users need to determine their travel type so that users can be matched with promotional SMS templates in the future. Multiple users can be selected from at least one user. The selection criteria can be set according to actual business needs. For example, users whose pre-trip behavior scores reach a preset threshold can be selected, indicating that the user has a strong potential travel intention. Users who have traveled once or more in the past 3 months and have recently had travel-related online behavior can also be selected. User groups in specific regions can also be selected.

[0020] The logistic regression model described above is a machine learning model used to solve classification problems. Its core principle is to use the sigmoid function to map the output of linear regression to a range of 0-1, thereby obtaining the probability that a sample belongs to a certain category. In this implementation, the logistic regression model can be pre-trained using training data, such as users' travel scores, pre-trip behavior scores, and corresponding actual tourism type labels, adjusting the model parameters until the model's prediction accuracy meets the requirements. After training, the model can receive multiple users' travel scores and pre-trip behavior scores as input, outputting the probability that the user belongs to different preset tourism types. These tourism types are specific categories based on users' travel preferences and potential travel needs. The classification criteria can be determined by combining common tourism product types in the market and users' travel purposes, such as natural scenery tourism, historical and cultural tourism, leisure and vacation tourism, entertainment and experience tourism, and adventure and challenge tourism.

[0021] Step 140: Obtain the feature similarity between the travel type of the multiple users and each of the multiple promotional SMS templates, select the promotional SMS template with the highest similarity as the target template, and push the push information corresponding to the target template to the multiple users, wherein each of the promotional SMS templates corresponds to at least one travel type.

[0022] The aforementioned promotional SMS templates can be pre-designed, standardized SMS content frameworks used to push travel-related information to users. Each template is designed around a specific travel theme or user needs, and typically includes a greeting, travel-type related recommendations such as attraction introductions, travel route recommendations, special offers, calls to action such as "Click the link to learn more" or "Book now for special offers," contact information or redirect links, and a closing statement. Each promotional SMS template corresponds to at least one travel type; for example, a promotional SMS template might correspond to both natural scenery travel and leisure vacation travel, with the template content focusing on recommending products and services related to these two types of travel.

[0023] The aforementioned feature similarity is an indicator used to measure the degree of matching between the travel type of the user to be predicted and the promotional SMS template. The higher the similarity, the more the content of the promotional SMS template matches the travel preferences of the user to be predicted, and the more likely the user is to accept the information corresponding to the template.

[0024] The aforementioned target template is the promotional SMS template selected from multiple templates that has the highest similarity to the travel type characteristics of the user to be predicted. This template forms the basis for generating subsequent push notifications, and its content has the highest match with the travel needs of the user to be predicted. The push notifications can be understood as the specific information content generated based on the target template and ultimately pushed to the user to be predicted. During generation, adjustments can be made to the target template based on the personalized information of the user to be predicted (such as user name, frequently traveled cities) to make the information more targeted; for example, if the greeting in the target template is "Dear user, hello!", the generated push notification can be adjusted to "Dear XX user, hello!" (XX is the user's name).

[0025] In the information push method implemented in this application, tourism preference features are extracted and travel scores are calculated by clustering analysis of users' historical travel information. Pre-trip behavior scores are then calculated by combining these with pre-trip behavior features. Finally, a logistic regression model is used to predict the tourism type and match the optimal promotional SMS template to push information, achieving relatively accurate delivery of tourism marketing information. This method integrates historical preference and current behavior data to improve the matching degree of user needs, and relies on message-adaptive template pushes to enhance the targeting of the push, effectively improving the matching degree of the push results to the user's current travel needs.

[0026] Optionally, obtaining the user's travel score based on the travel preference features includes: Each travel preference feature is assigned a weight, the total weight of which is 1. The travel preference feature includes at least one of the following: tour group type, tour frequency, tour destination, and tour duration. The raw data of each tourism preference characteristic are standardized to obtain standardized values; The weighted score for each tourism preference feature is determined to be the product of the weight of the corresponding tourism preference feature and the standardized value; The travel score is determined to be the sum of weighted scores for all travel preference features.

[0027] The weights mentioned above represent the degree of contribution of each travel preference feature to the travel score. The larger the weight value, the more significant the influence of the feature on the user's travel preferences. Furthermore, the sum of the weights of all selected travel preference features must be 1 to ensure the rationality and comparability of the travel score calculation.

[0028] The raw data mentioned above represents the initial recorded values ​​for each travel preference feature. The magnitude of the raw data varies significantly between different features. For example, the raw data for travel frequency might be "2 times / quarter," and the raw data for travel duration might be "3 days." Directly using these data for calculation would cause features with large magnitudes to excessively influence the travel score; therefore, standardization is necessary.

[0029] The standardization process described above involves transforming raw data of different magnitudes into a unified numerical range using mathematical transformations. The resulting value is the standardized value. Standardized values ​​eliminate differences in data magnitude, making the assessment of the contribution of each feature more equitable.

[0030] The weighted score represents the specific contribution of an individual travel preference feature after considering its importance (weight) and data standardization. It directly reflects the actual impact of the feature on the user's travel score. The higher the weight and the larger the standardized value, the higher the weighted score, indicating a stronger positive contribution of the feature to the user's travel preference. The aforementioned travel score is the final value obtained by summing the weighted scores of all travel preference features, used to quantify the strength of the user's overall travel preference. Its numerical range is usually consistent with the standardized value range (e.g., 0 to 1). A higher value indicates a higher degree of fit between the user's travel preference and the current promotional scenario, and vice versa.

[0031] For example, users' historical travel information can be collected, and the users' historical data can be tagged, mainly including tour group type, tour frequency, tour destination and tour duration. Through cluster analysis of historical data, travel users can be tagged with travel preferences.

[0032] Furthermore, the weight of each indicator can be determined. Let's assume the weights for these four indicators are assigned as follows: Group tour type weight: w Group Tourism frequency weight: w Freq Tourism Destinations Weight: w Dest Travel duration weight: w Dur The total weights are 1, i.e.: w Group +w Freq +w Dest +w Dur =1 Since each indicator may have different ranges and units, they can be standardized to the same scale, such as between 0 and 1. Mini-maximum normalization can be performed using the following formula: Normalized Value=

[0033] Where Value is the original data, and Min and Max are the minimum and maximum values ​​of the indicator, respectively.

[0034] The weighted score for each indicator can be calculated using the following formula: Group Score = w Group ×Normalized Group Type Frequency Score = w Freq ×Normalized Frequency Destinations Score = w Dest ×Normalized Destinations Duration Score = w Dur ×Normalized Duration Add up all the weighted scores to get the total travel score: Travel Score=Group Score + Frequency Score + Destinations Score +Duration Score Furthermore, specific metrics can be defined for each behavioral category. These metrics can be used to calculate weighted scores for characteristics such as a user's online behavior, communication frequency, number of working days, and holiday schedule before traveling.

[0035] (1) Internet behavior indicators: Content relevance: C Number of clicks: K Time Spent: T Duration of visit: D (2) Communication behavior indicators: Communication Frequency (Comm Frequency): F Communication Duration (Comm Duration): L (3) Work behavior indicators: Work Start / End Time: S,E Work Activity Record: A (4) Rest behavior indicators: Rest start / end time: R s ,R e Rest Activity Type: B For each behavioral category, you can calculate the score for that metric and then sum them up to get the total score for that category.

[0036] Given that data features often exhibit non-linear relationships, traditional linear normalization methods may not adequately capture these complex patterns. Therefore, hyperbolic tangent and logarithmic functions are introduced to better handle the non-linear variations of features. Furthermore, interactions may exist between different features; introducing interaction terms helps the model learn the complex relationships between features. Receive an internet behavior score:

[0037] Communication behavior score:

[0038] Work behavior score:

[0039] Rest behavior score:

[0040] The final pre-event behavior score is obtained by weighted summing of the scores for all behavior categories:

[0041] Among them, w web w comm w work w rest These are the weights for internet browsing behavior, communication behavior, work behavior, and rest behavior, respectively.

[0042] In the information push method of this application, by assigning weights to multi-dimensional preference features such as tour group type and tour frequency and standardizing them, and then calculating the travel score through weighted summation, the quantification and integration of user travel preferences are realized. This not only highlights the influence of key features through weighting to adapt to personalized needs, but also eliminates the interference of data heterogeneity through standardization. The resulting travel score is both objective and accurate, providing reliable data support for subsequent prediction of user travel intentions and precision marketing.

[0043] Optionally, obtaining the user's pre-trip behavior score based on pre-trip travel behavior characteristics includes: A weight is assigned to each pre-trip behavioral feature of tourism, which includes at least one of the following: internet usage, commuting frequency, number of working days, and holiday schedule. The raw data of each pre-trip behavior feature is processed using at least one of the hyperbolic tangent function, logarithmic function, and sine function to obtain a score for each pre-trip behavior feature. The score of the pre-trip behavior feature represents the strength of the correlation between the user's pre-trip behavior and travel intention. The weighted score of the pre-trip behavior feature is determined to be the product of the weight of the corresponding pre-trip behavior feature and the score. The pre-trip behavior score is determined to be the sum of the weighted scores of all pre-trip behavior characteristics of tourism.

[0044] The hyperbolic tangent, logarithmic, and sine functions are all mathematical functions used to process raw data. Their core function is to transform raw data of different magnitudes and distributions into values ​​that better meet the needs of "correlation assessment," eliminating the influence of extreme values ​​or distributional biases on the results. Specifically, the hyperbolic tangent (tanh) can map data to a range of -1 to 1, suitable for processing data with a large numerical range and extreme values, such as "travel-related page browsing time" in online behavior. The logarithmic function (log) can transform long-tailed distributed data—where a few data points are extremely large and most are small—into a flatter distribution, suitable for processing count data, such as "travel keyword search counts" in online behavior. The sine function (sin) can capture the periodic changes in data, suitable for processing data with time-period characteristics (such as the number of working days or holiday sequences). The aforementioned raw data are the initial recorded values ​​of various pre-travel behavioral characteristics.

[0045] The scores of the aforementioned pre-trip behavioral characteristics are numerical values ​​obtained after processing the raw data using a function. Their range is usually normalized to 0 to 1. The higher the value, the stronger the correlation between the user's behavior and travel intention. For example, an online behavior score of 0.9 means that the user's online operations are highly inclined towards travel planning and have a very strong correlation with travel intention; a score of 0.2 means that the correlation is relatively weak.

[0046] In the information push method of this application, weights are assigned to pre-trip behavioral features such as internet browsing behavior and commuting frequency. The raw data is then processed using functions such as hyperbolic tangent and logarithmic functions, and a weighted sum is used to obtain a pre-trip behavior score, thus achieving accurate quantification of users' potential travel intentions. This approach not only captures the non-linear correlation of feature data through function processing to improve scoring accuracy, but also distinguishes the degree of correlation between features and travel intentions using weights. The resulting pre-trip behavior score effectively reflects users' current travel preferences, providing a scientific basis for subsequent travel type prediction and user selection.

[0047] Optionally, the step of inputting the travel scores and pre-trip behavior scores of multiple users into a logistic regression model for prediction to obtain the travel types of the multiple users includes: The travel scores and pre-trip behavior scores of multiple users are input into a logistic regression model, and the probability of the multiple users belonging to each preset travel type is determined by the output of the logistic regression model. The travel type with the highest probability value is determined to be the travel type of the multiple users.

[0048] In this implementation, different tourism types can be defined, such as Type A, Type B, Type C, and Type D. Based on the user's travel score and pre-trip behavior score, a logistic regression model is trained to predict the tourism type based on the new user's feature values. For example, the logistic regression model can be as follows:

[0049] Where k is the type of tourism, K is the total number of tourism types, and β is the model parameter; Assuming a new user's travel score is 0.8, their pre-trip behavior score is 0.7, and there are four travel types, then a logistic regression model is used to predict the user's travel type:

[0050]

[0051]

[0052]

[0053] The type of travel most likely to be taken by the user is the one with the highest probability of the calculated result.

[0054] In the information push method of this application, by inputting the travel score and pre-trip behavior score of the user to be predicted into a logistic regression model, the probability of the user belonging to each preset travel type is output, and the type with the highest probability is selected, thus achieving accurate determination of the user's travel type. This method relies on both the dual scores—travel score reflecting historical preferences and pre-trip behavior score reflecting current tendencies—to provide comprehensive input to the model and improve the basic accuracy of prediction. Furthermore, by comparing probabilities, the most suitable travel type for the user is clearly identified, providing a basis for subsequent matching of promotional templates and achieving precise marketing.

[0055] Optionally, obtaining the similarity between the travel types of the multiple users and the features of each promotional SMS template in the multiple promotional SMS templates includes: Feature vectors are constructed for the travel types and promotional SMS templates of multiple users, resulting in the travel type feature vector and the promotional SMS template feature vector. Obtain the cosine value of the angle between the feature vector of the tourism type and the feature vector of the promotional SMS template. The cosine value represents the feature similarity between the tourism type and the promotional SMS template. The larger the value of the cosine value, the higher the feature matching degree between the corresponding tourism type and the promotional SMS template.

[0056] The aforementioned tourism type feature vector is a multi-dimensional numerical array used to describe the core attributes of the user's tourism type to be predicted. Each dimension corresponds to a feature that can characterize the tourism type, and the dimension value represents the importance of that feature in the tourism type. For example, if the tourism type is family-oriented, its core features may include parent-child interactive projects, child-friendly facilities, short trips, and safety guarantee services. Each feature corresponds to a numerical value, and these numerical values ​​together constitute the tourism type feature vector for family-oriented tourism.

[0057] The aforementioned promotional SMS template feature vector is a multi-dimensional numerical array used to describe the core content attributes of the promotional SMS template. Its dimensions can be consistent with the dimensions of the tourism type feature vector, and the value of each dimension represents the prominence of that feature in the template content. For example, a promotional SMS template with the content "Weekend family farm trip, including DIY crafts, children's playground, full safety escort, 1-hour drive away" has core features consistent with the features of the family-oriented tourism type. The prominence values ​​corresponding to each feature together constitute the promotional SMS template feature vector of this template.

[0058] For example, multiple SMS templates can be created, each corresponding to a different tourism type or marketing scenario. Features are defined for each tourism type, describing its key elements such as adventure, leisure, culture, and business. Features are extracted from each SMS template; these features can be keywords, themes, sentiment, etc. A similarity calculation method is used to evaluate the similarity between the tourism type feature vector and the SMS template feature vector.

[0059] Construct a message template containing the following fields:

[0060] Furthermore, suppose the feature vector of tourism type A is:

[0061] Assume the feature vector of SMS template T is:

[0062] The similarity between the tourism type characteristics and the SMS template characteristics is:

[0063] These represent the weights of tourism type A and SMS template T on features i and j, respectively.

[0064] For the user's travel type, calculate the similarity with all SMS templates and select the template with the highest feature similarity.

[0065] Furthermore, NLP techniques can be used to optimize the fluency of the message content generated based on the above steps.

[0066] In the information push method of this application, feature vectors are constructed for tourism types and promotional SMS templates respectively. The cosine similarity between the two features is calculated, and the optimal template is matched, thus achieving precise adaptation between tourism marketing content and user needs. This approach quantifies the core attributes of tourism types and the content attributes of templates through feature vectors, eliminating subjective matching bias. Furthermore, the cosine similarity directly reflects the degree of matching between the two, ensuring that the pushed templates align with user travel preferences and enhancing the acceptability and conversion potential of marketing information.

[0067] Optionally, before inputting the travel score and prior behavior score of the user to be predicted into the logistic regression model for prediction, the method further includes: Determine at least one user's preference type as a single travel type or a combination of multiple travel types; The user's pre-emptive behavior score for each preference type is split in chronological order to form multiple consecutive time-series score segments; The evolution trend of each time-series scoring segment is analyzed by linear fitting. The objective is to minimize the sum of squared errors between the actual score and the predicted score. Different weights are assigned to the errors according to the time distance, and an objective function including a time decay factor and a regularization term is constructed. The objective function is iteratively optimized using stochastic gradient descent to obtain the optimal interest feature coefficients; The user's travel intention is predicted by the optimal interest feature coefficient, and users whose travel intention meets the preset threshold are identified as the multiple users.

[0068] In this implementation, to determine the set of types that a user is interested in, each user Ui corresponding to each single type or combination of multiple single types can be regarded as a user preference, i.e., G. j A vector containing preference types can be represented as:

[0069] User travel behavior can be viewed as a time-series vector of events. To study the changing trends of users' travel type preferences, each user's type rating is divided according to time sequence, and a certain preference type G of user Ui is assigned. j The m rating data points divided according to time sequence are considered as a time-series data segment.

[0070] By analyzing the time series of each user type to identify its evolutionary trend characteristics, the optimal feature values ​​can be used to fit and predict trend extension changes. To further explore the changing trends of user interests, the convergence of the data is assessed by analyzing the changing trends of variance over time for different types of rating data.

[0071] Furthermore, linear relationships can be used to fit the overall trend changes of users. The basic idea is to use the least squares principle to fit the actual data. Compared with the prediction results The sum of squares of the error Et reaches its minimum, calculated using the following formula:

[0072] To describe the relationship between variables and prediction results, a trend characteristic equation can be constructed, and the formula can be as follows:

[0073] Where σ and δ are undetermined coefficients, which depend on the conditional extrema that satisfy the minimum value; ϴ is the time slice length; This represents the bias of user interest changes at time t; the function W is the variation function, primarily extracting the mean of rating data within a time segment. User interest in tourism types changes continuously over time; therefore, while considering the error between predicted and actual data, the impact of time on the error is also taken into account. Errors are weighted according to their temporal proximity. From the perspective of prediction accuracy, in data with a stable trend, errors from time segments closer to the current time are more significant than errors from more distant time segments. Therefore, rating errors from more recent to more recent times are assigned weight factors according to their temporal order. The error and calculation formula are shown below:

[0074] After adding weights and expanding, we get the formula:

[0075] Function Substituting the trend characteristic equation into the above formula, the simplified result is shown below:

[0076] Within the model framework, the transformed data can be divided into several segments. The convergence of the time slice score variance can be used to determine whether the data trend characteristics meet the stable extension trend. If they do, the trend prediction model can be trained. The model can predict the data that the user may provide at the next time point by initializing the data and calculate the error with the true value. New training data can be added to the previous results for prediction and verification. By adjusting the user's time series interest bias, the error between the trend prediction model and the actual data can be minimized.

[0077] Furthermore, to find the optimal interest feature coefficients, an objective function can be constructed using the error and time decay factor. To prevent overfitting, a regularization term is added to the objective function. The regularized objective function is as follows:

[0078] Where k represents the number of time sample points, It is the true value of the user data at time t. The interest feature coefficients to be optimized It is a regular expression parameter. It is a regularization term, a constant term in the objective function. It is not regularized, therefore It is not advisable to set the value too high, otherwise it will cause all parameters to be too small and the dependent variable to be close to a constant, resulting in underfitting. The weighting factor represents the weight of the error term at time t. It is a range in The constant is represented as follows, based on the decay error term from far to near and from fast to slow decay rate:

[0079] Since the objective function is a weighted sum of errors, the coefficients that minimize the error are needed. We choose stochastic gradient descent to optimize the objective function.

[0080]

[0081] Then, the stochastic gradient descent method is used for optimization. For each iteration of all sample points, the parameters are updated in the opposite direction of the gradient until convergence, as shown in the formula.

[0082]

[0083] in, The learning factor controls the learning rate and is used only for feature extraction during the training process; no activation function is set. Interest feature coefficients that meet the conditions are used to predict the user's rating for the corresponding type. Assuming the user's current rating action is at time t, and the next rating action will occur at time t+1, the prediction method for user U's rating of items containing type G at time t+1 is shown in the formula:

[0084]

[0085] 3. Set a threshold r for the feature vector of a new user and the prediction result R of the potential travel user. When R=1, the new user is a potential travel user and marketing information can be sent to the user.

[0086]

[0087] In the information push method of this application, by determining user preference types, splitting time-series rating segments, combining linear fitting with an objective function containing time decay factors and regularization terms, and then optimizing using stochastic gradient descent to obtain the optimal interest feature coefficients, users with high travel intentions are screened, thus achieving accurate identification of users' travel intentions. This method captures dynamic changes in user interests through time-series analysis, improves prediction accuracy through objective function optimization and parameter iteration, and focuses on high-value users through threshold screening, effectively reducing ineffective marketing and laying the foundation for subsequent accurate push notifications.

[0088] To more clearly describe the information push method provided in this application, please refer to [link / reference needed]. Figure 3 , Figure 3 This is a schematic diagram of the structure of an information push device 300 provided in an embodiment of this application. As shown in the figure, the information push device 300 includes: Analysis module 310 is used to perform cluster analysis on the historical travel information of at least one user to obtain the travel preference characteristics of the at least one user, and obtain the travel score of the at least one user based on the travel preference characteristics; The acquisition module 320 is used to acquire the pre-trip behavior score of the at least one user based on pre-trip behavior characteristics. Prediction module 330 is used to input the travel scores and pre-event behavior scores of multiple users among the at least one user into a logistic regression model for prediction, so as to obtain the travel type of the multiple users; The push module 340 is used to obtain the feature similarity between the travel type of the multiple users and each of the multiple promotional SMS templates, select the promotional SMS template with the highest similarity as the target template, and push the push information corresponding to the target template to the multiple users, wherein each of the promotional SMS templates corresponds to at least one travel type.

[0089] Optionally, the analysis module 310 can also be used for: Each travel preference feature is assigned a weight, the total weight of which is 1. The travel preference feature includes at least one of the following: tour group type, tour frequency, tour destination, and tour duration. The raw data of each tourism preference characteristic are standardized to obtain standardized values; The weighted score for each tourism preference feature is determined to be the product of the weight of the corresponding tourism preference feature and the standardized value; The travel score is determined to be the sum of weighted scores for all travel preference features.

[0090] Optionally, the acquisition module 320 can also be used for: A weight is assigned to each pre-trip behavioral feature of tourism, which includes at least one of the following: internet usage, commuting frequency, number of working days, and holiday schedule. The raw data of each pre-trip behavior feature is processed using at least one of the hyperbolic tangent function, logarithmic function, and sine function to obtain a score for each pre-trip behavior feature. The score of the pre-trip behavior feature represents the strength of the correlation between the user's pre-trip behavior and travel intention. The weighted score of the pre-trip behavior feature is determined to be the product of the weight of the corresponding pre-trip behavior feature and the score. The pre-trip behavior score is determined to be the sum of the weighted scores of all pre-trip behavior characteristics of tourism.

[0091] Optionally, the prediction module 330 can also be used for: The travel scores and pre-trip behavior scores of multiple users are input into a logistic regression model, and the probability of the multiple users belonging to each preset travel type is determined by the output of the logistic regression model. The travel type with the highest probability value is determined to be the travel type of the multiple users.

[0092] Optionally, the push module 340 can also be used for: Feature vectors are constructed for the travel types and promotional SMS templates of multiple users, resulting in the travel type feature vector and the promotional SMS template feature vector. Obtain the cosine value of the angle between the feature vector of the tourism type and the feature vector of the promotional SMS template. The cosine value represents the feature similarity between the tourism type and the promotional SMS template. The larger the value of the cosine value, the higher the feature matching degree between the corresponding tourism type and the promotional SMS template.

[0093] Optionally, the information push device 300 can also be used for: Determine at least one user's preference type as a single travel type or a combination of multiple travel types; The user's pre-emptive behavior score for each preference type is split in chronological order to form multiple consecutive time-series score segments; The evolution trend of each time-series scoring segment is analyzed by linear fitting. The objective is to minimize the sum of squared errors between the actual score and the predicted score. Different weights are assigned to the errors according to the time distance, and an objective function including a time decay factor and a regularization term is constructed. The objective function is iteratively optimized using stochastic gradient descent to obtain the optimal interest feature coefficients; The user's travel intention is predicted by the optimal interest feature coefficient, and users whose travel intention meets the preset threshold are identified as the multiple users.

[0094] The information push device in this application embodiment can be an electronic device, such as an electronic device with an operating system, or a component in an electronic device, such as an integrated circuit or a chip.

[0095] The information push device provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method embodiments shown achieve the same technical effects, and will not be described again here to avoid repetition.

[0096] For details, see Figure 4 As shown in the figure, this application embodiment also provides an electronic device, including a bus 401, a transceiver 402, an antenna 403, a bus interface 404, a processor 405, and a memory 406.

[0097] Processor 405, used for: Cluster analysis is performed on the historical travel information of at least one user to obtain the travel preference characteristics of the at least one user, and the travel score of the at least one user is obtained based on the travel preference characteristics; The pre-trip behavior score of at least one user is obtained based on pre-trip behavior characteristics. The travel scores and pre-trip behavior scores of multiple users among the at least one user are input into a logistic regression model for prediction to obtain the travel type of the multiple users. The similarity between the travel type of the multiple users and the features of each promotional SMS template among the multiple promotional SMS templates is obtained. The promotional SMS template with the highest similarity is selected as the target template, and the push information corresponding to the target template is pushed to the multiple users. Each of the promotional SMS templates corresponds to at least one travel type.

[0098] exist Figure 4 In this context, a bus architecture (represented by bus 401) is used. Bus 401 can include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 405 and memory represented by memory 406. Bus 401 can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 404 provides an interface between bus 401 and transceiver 402. Transceiver 402 can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 405 is transmitted over a wireless medium via antenna 403, which further receives data and transmits data to processor 405.

[0099] Processor 405 is responsible for managing bus 401 and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. Memory 406 can be used to store data used by processor 405 during operation.

[0100] Alternatively, the processor 405 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD).

[0101] Optionally, the processor 405 is specifically used for: Each travel preference feature is assigned a weight, the total weight of which is 1. The travel preference feature includes at least one of the following: tour group type, tour frequency, tour destination, and tour duration. The raw data of each tourism preference characteristic are standardized to obtain standardized values; The weighted score for each tourism preference feature is determined to be the product of the weight of the corresponding tourism preference feature and the standardized value; The travel score is determined to be the sum of weighted scores for all travel preference features.

[0102] Optionally, the processor 405 is specifically used for: A weight is assigned to each pre-trip behavioral feature of tourism, which includes at least one of the following: internet usage, commuting frequency, number of working days, and holiday schedule. The raw data of each pre-trip behavior feature is processed using at least one of the hyperbolic tangent function, logarithmic function, and sine function to obtain a score for each pre-trip behavior feature. The score of the pre-trip behavior feature represents the strength of the correlation between the user's pre-trip behavior and travel intention. The weighted score of the pre-trip behavior feature is determined to be the product of the weight of the corresponding pre-trip behavior feature and the score. The pre-trip behavior score is determined to be the sum of the weighted scores of all pre-trip behavior characteristics of tourism.

[0103] Optionally, the processor 405 is specifically used for: The travel scores and pre-trip behavior scores of multiple users are input into a logistic regression model, and the probability of the multiple users belonging to each preset travel type is determined by the output of the logistic regression model. The travel type with the highest probability value is determined to be the travel type of the multiple users.

[0104] Optionally, the processor 405 is specifically used for: Feature vectors are constructed for the travel types and promotional SMS templates of multiple users, resulting in the travel type feature vector and the promotional SMS template feature vector. Obtain the cosine value of the angle between the feature vector of the tourism type and the feature vector of the promotional SMS template. The cosine value represents the feature similarity between the tourism type and the promotional SMS template. The larger the value of the cosine value, the higher the feature matching degree between the corresponding tourism type and the promotional SMS template.

[0105] Optionally, the processor 405 can also be used for: Determine at least one user's preference type as a single travel type or a combination of multiple travel types; The user's pre-emptive behavior score for each preference type is split in chronological order to form multiple consecutive time-series score segments; The evolution trend of each time-series scoring segment is analyzed by linear fitting. The objective is to minimize the sum of squared errors between the actual score and the predicted score. Different weights are assigned to the errors according to the time distance, and an objective function including a time decay factor and a regularization term is constructed. The objective function is iteratively optimized using stochastic gradient descent to obtain the optimal interest feature coefficients; The user's travel intention is predicted by the optimal interest feature coefficient, and users whose travel intention meets the preset threshold are identified as the multiple users.

[0106] It should be noted that the electronic device provided in this application embodiment is a device capable of executing the above-described information push method. Therefore, all implementation methods in the above-described information push method embodiments are applicable to this electronic device and can achieve the same or similar beneficial effects. To avoid repetition, this embodiment will not elaborate further.

[0107] This invention also provides an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the various processes of the above-described information push method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0108] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described information push method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0109] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described information push method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0110] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0112] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An information push method characterized by comprising: The method comprises: performing cluster analysis on historical travel information of at least one user to obtain a travel preference feature of the at least one user, and obtaining a travel score of the at least one user based on the travel preference feature; obtaining a pre-travel behavior score of the at least one user based on a pre-travel behavior feature; inputting the travel score and the pre-travel behavior score of a plurality of users in the at least one user into a logistic regression model for prediction to obtain a travel type of the plurality of users; obtaining a feature similarity between the travel type of the plurality of users and each promotion short message template in a plurality of promotion short message templates, selecting a promotion short message template with the highest similarity as a target template, and pushing a push information corresponding to the target template to the plurality of users, wherein each promotion short message template corresponds to at least one travel type.

2. The method of claim 1, wherein, The obtaining of the travel score of the user based on the travel preference feature comprises: assigning a weight to each travel preference feature, wherein the sum of the weights is 1, and the travel preference feature comprises at least one of a travel group type, a travel frequency, a travel destination, and a travel duration; performing standardization processing on original data of each travel preference feature to obtain a standardized value; determining a weighted score of each travel preference feature as a product of the weight of the corresponding travel preference feature and the standardized value; and determining the travel score as a sum of the weighted scores of all travel preference features.

3. The method of claim 1, wherein, The obtaining of the pre-travel behavior score of the user based on the pre-travel behavior feature comprises: assigning a weight to each pre-travel behavior feature, wherein the pre-travel behavior feature comprises at least one of an online behavior, a commuting frequency, a working day number, and a holiday timing; processing original data of each pre-travel behavior feature by using at least one of a hyperbolic tangent function, a logarithmic function, and a sine function to obtain a score of each pre-travel behavior feature, wherein the score of the pre-travel behavior feature represents a strength of an association between a pre-travel behavior of the user and a travel intention; determining a weighted score of the pre-travel behavior feature as a product of the weight of the corresponding pre-travel behavior feature and the score; and determining the pre-travel behavior score as a sum of the weighted scores of all pre-travel behavior features.

4. The method according to any one of claims 1 to 3, characterized in that, The inputting of the travel score and the pre-travel behavior score of the plurality of users in the at least one user into the logistic regression model for prediction to obtain the travel type of the plurality of users comprises: inputting the travel score and the pre-travel behavior score of the plurality of users into the logistic regression model, determining a probability that the plurality of users belong to each preset travel type through an output of the logistic regression model, and determining a travel type with the highest probability value as the travel type of the plurality of users. The obtaining of the feature similarity between the travel type of the plurality of users and each promotion short message template in the plurality of promotion short message templates comprises:

5. The method according to any one of claims 1 to 3, characterized in that, constructing a feature vector for the travel type of the plurality of users and each promotion short message template respectively to obtain a travel type feature vector and a promotion short message template feature vector. ​ Obtaining a cosine value of an angle between the travel type feature vector and the promotion short message template feature vector, the cosine value representing a feature similarity between the travel type and the promotion short message template, and the greater the cosine value, the higher the feature matching degree between the corresponding travel type and the promotion short message template.

6. The method of claim 5, wherein, Before the travel score and the pre-travel behavior score of the user to be predicted are input into the logistic regression model for prediction, the method further includes: Determining that at least one user corresponds to a single travel type or a combination of multiple travel types as a preferred type of the user; Splitting the pre-travel behavior score of the user for each preferred type in chronological order to form a plurality of continuous time sequence score segments; By linear fitting analysis of the evolution trend of each time sequence score segment, taking the minimum sum of squared errors of actual score and predicted score as the target, and according to the time distance, the error is given different weights, a target function is constructed including time decay factor and regularization term; Using the stochastic gradient descent method to iteratively optimize the target function to obtain the optimal interest feature coefficient; Predicting the travel intention of the user through the optimal interest feature coefficient, and determining that the user whose travel intention meets the preset threshold is the plurality of users.

7. An information push apparatus characterized by comprising: Comprising: An analysis module configured to perform cluster analysis on historical travel information of at least one user to obtain a travel preference feature of the at least one user, and obtain a travel score of the at least one user based on the travel preference feature; An obtaining module configured to obtain a pre-travel behavior score of the at least one user based on a travel pre-travel behavior feature; A prediction module configured to input the travel score and the pre-travel behavior score of a plurality of users in the at least one user into a logistic regression model for prediction to obtain a travel type of the plurality of users; A pushing module configured to obtain a feature similarity between the travel type of the plurality of users and each promotion short message template in a plurality of promotion short message templates, select a promotion short message template with the highest similarity as a target template, and push a pushing information corresponding to the target template to the plurality of users, wherein each promotion short message template corresponds to at least one travel type.

8. An electronic device, comprising: Comprising: A processor, a memory, and a program stored on the memory and executable on the processor, the program being executed by the processor to implement the steps of the information pushing method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the information pushing method according to any one of claims 1 to 6.

10. A computer program product, characterised in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the information pushing method according to any one of claims 1 to 6.

Images