An online advertisement intelligent pushing method and system based on big data

Abstract

This invention discloses an intelligent online advertising push method and system based on big data, belonging to the field of advertising push technology. The invention first uses big data to predict the target scenarios a user may enter within a preset future time window. Based on the target scenarios, it filters suitable candidate ads from an ad library, pre-generating a candidate ad list. When a user enters the target scenario, it collects the user's current real-time environmental characteristics, selects the target ad, and obtains its preset delivery conditions and estimated consumption time. It also obtains the current weather duration prediction information and compares it with the estimated consumption time to generate a travel feasibility coefficient. The real-time environmental characteristics are matched with the preset delivery conditions to obtain a preliminary matching result, which is then corrected by the travel feasibility coefficient to obtain the final matching result, completing the ad push. This invention effectively reduces invalid ad pushes, improves ad conversion efficiency, and enhances user experience through dual verification of environmental adaptation and time feasibility.

Description

Technical Field

[0001] This invention belongs to the field of advertising push technology, specifically relating to an intelligent online advertising push method and system based on big data. Background Technology

[0002] With the widespread adoption of the internet and mobile devices, online advertising has become a crucial component of the internet business ecosystem, driving the development of various internet services through monetization. In the development of advertising delivery technology, contextualized advertising delivery based on big data has been widely applied. This technology can accurately deliver ads based on the user's current location, thereby improving ad conversion rates and user satisfaction. Existing solutions employ a static matching method, determining which ads to push based on the environmental characteristics of the user's current location.

[0003] However, this static matching method cannot consider the duration of environmental characteristics, nor can it assess the time cost required for user behavior corresponding to the advertisement. For example, an outdoor activity advertisement pushed when the weather forecast shows sunny weather often lacks actual conversion value because the sunny weather will not last long. Similarly, pushing advertisements for roadside businesses in congested areas, while users may easily notice this information in the current environment, is meaningless because users will quickly leave the congested area, failing to achieve effective ad conversion and wasting resources. Likewise, excessive or inappropriate ad pushes can lead to user impatience and even resistance, which, in the long run, will reduce user dependence on the platform, thus affecting the healthy development of the entire internet ecosystem.

[0004] Therefore, in order to address the aforementioned technical issues, it is necessary to provide a method and system for intelligent online advertising push based on big data.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for intelligent online advertising push based on big data, which can solve the technical problems mentioned in the background art.

[0007] To achieve the above objectives, a specific embodiment of the present invention provides the following technical solution: A big data-based intelligent online advertising push method includes: S100. Utilize big data to predict the scenarios that users may enter within a preset future time window to obtain the target scenario; S200: Based on the target scenario, select candidate ads from the ad library that are suitable for the target scenario and generate a candidate ad list; S300. When a user enters the target scene, the user's current real-time environmental characteristics are collected. The real-time environmental characteristics include one or more of the following: current weather data, current geographical location data, and current traffic status data. S400. Select one from the candidate advertisement list as the target advertisement, and obtain the preset delivery conditions and estimated consumption time corresponding to the target advertisement. The preset delivery conditions include one or more of the following: suitable weather type, suitable geographical location, and suitable traffic condition. S500: Obtain the duration prediction information of the current weather data, compare the estimated consumption time with the duration prediction information, and generate a travel feasibility coefficient based on the comparison result. S600. The real-time environmental features are matched and judged with the preset deployment conditions to obtain a preliminary matching result. The preliminary matching result is then corrected with the travel feasibility coefficient to obtain the final matching result. S700: If the final matching result is successful, the target advertisement is pushed to the user terminal; if the final matching result is unsuccessful, another advertisement is selected from the candidate advertisement list as the new target advertisement, and steps S500 to S700 are executed again.

[0008] In one or more embodiments of the present invention, the estimated time consumption is determined based on the type of the target advertisement, the historical average dwell time of the attraction or merchant corresponding to the target advertisement, and the user's historical behavior data.

[0009] In one or more embodiments of the present invention, the duration prediction information is obtained by predicting based on meteorological forecast data, the current speed and direction of the weather system.

[0010] In one or more embodiments of the present invention, step S500 involves generating a travel feasibility coefficient based on the comparison results, specifically including the following steps: Calculate the difference between the predicted duration information and the estimated consumption time. When the difference is greater than or equal to a first preset threshold, set the travel feasibility coefficient to a first value. When the difference is less than the first preset threshold and greater than the second preset threshold, the travel feasibility coefficient is set to the second value; When the difference is less than or equal to the second preset threshold, the travel feasibility coefficient is set to the third value; The first, second, and third values ​​decrease sequentially.

[0011] In one or more embodiments of the present invention, step S600 involves correcting the preliminary matching result using a travel feasibility coefficient, specifically including the following steps: When the travel feasibility coefficient is lower than the preset coefficient threshold, the preliminary matching result will be corrected to failure; When the travel feasibility coefficient is higher than or equal to the preset coefficient threshold, the initial matching result remains unchanged.

[0012] In one or more embodiments of the present invention, before the matching determination in step S600, the method further includes: The system acquires real-time feedback behavior data of the user before the target advertisement is pushed, including the user's click data, conversion data, and ignore data of historical advertisements; the real-time feedback behavior data is then used in a matching and judgment process in conjunction with real-time environmental features.

[0013] In one or more embodiments of the present invention, it further includes: S800: The successfully pushed advertisements and their corresponding real-time environmental features and travel feasibility coefficients are used as positive samples, and the advertisements that are ultimately skipped after the push fails and their corresponding real-time environmental features and travel feasibility coefficients are used as negative samples and stored in the sample library. S900. Periodically use samples in the sample library to perform offline training and optimization of the weight parameters for matching judgment, and synchronize the optimized weight parameters to the online matching judgment process.

[0014] In addition, a big data-based intelligent online advertising push system is also provided, including: The scene prediction module is used to use big data to predict the scenes that users may enter within a preset future time window, and obtain the target scene. The ad filtering module is used to filter candidate ads that are suitable for the target scenario from the ad library and generate a candidate ad list. The environment acquisition module is used to collect the user's current real-time environmental characteristics when the user enters the target scene. The real-time environmental characteristics include one or more of the following: current weather data, current geographical location data, and current traffic status data. The parameter acquisition module is used to select one as the target advertisement from the candidate advertisement list, and obtain the preset delivery conditions and estimated consumption time corresponding to the target advertisement. The preset delivery conditions include one or more of the following: suitable weather type, suitable geographical location, and suitable traffic status. The feasibility assessment module is used to obtain the duration prediction information of the current weather data, compare the estimated consumption time with the duration prediction information, and generate a travel feasibility coefficient based on the comparison result. The matching correction module is used to match and judge the real-time environmental features with the preset deployment conditions to obtain a preliminary matching result, and then correct the preliminary matching result with the travel feasibility coefficient to obtain a final matching result. The push determination module is used to push the target advertisement to the user terminal if the final matching result is successful; if the final matching result is unsuccessful, it selects another advertisement from the candidate advertisement list as the new target advertisement and re-executes the feasibility assessment, matching correction and push determination operations.

[0015] In one or more embodiments of the present invention, it further includes: The sample construction module is used to store successfully pushed advertisements and their corresponding real-time environmental features and travel feasibility coefficients as positive samples, and advertisements that are ultimately skipped after failed pushes and their corresponding real-time environmental features and travel feasibility coefficients as negative samples, and store them in the sample library. The iterative optimization module is used to periodically train and optimize the weight parameters for matching judgment offline using samples in the sample library, and synchronize the optimized weight parameters to the matching correction module.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: by using big data to predict in advance the target scene that the user may enter within a preset future time window, and generating a suitable candidate advertisement list in advance for the target scene, it not only reserves preparation time for advertisement screening, but also effectively disperses the real-time computing pressure of the system, and avoids push delay caused by temporary calculation after the user enters the scene. Based on the matching judgment of real-time environmental characteristics and preset advertising conditions, a new travel feasibility check has been added, which compares the predicted duration of weather with the estimated time of corresponding advertising behavior. The travel feasibility coefficient is used to make a secondary correction to the initial matching results, realizing dual verification of environmental adaptability and time feasibility. This fundamentally eliminates invalid pushes that seem to match but are actually caused by factors such as sudden weather changes or users quickly leaving the scene, resulting in no conversion value. This not only saves advertisers' advertising resources but also reduces unnecessary information disturbance to users, significantly improving the user experience. It generates an ordered list of candidate ads by combining the ad's suitability score with the scenario and historical click-through rate, taking into account both the scenario matching accuracy and conversion potential of the ads. It can flexibly adapt to different types and time-consuming ad push scenarios and has strong versatility. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall process of an online advertising intelligent push method based on big data in one embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the process of generating a travel feasibility coefficient and correcting matching results in a big data-based intelligent online advertising push method according to an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the process of generating a candidate ad list in a big data-based intelligent online ad push method according to an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the sample storage and model iteration optimization process in a big data-based intelligent online advertising push method according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the architecture of an online advertising intelligent push system based on big data in one embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0020] Example 1: An embodiment of the present invention provides an intelligent online advertising push method and system based on big data. By performing dual screening based on environmental adaptability and time feasibility, and dynamically correcting the matching results, it reduces invalid pushes, effectively improves advertising conversion efficiency, and enhances user experience.

[0021] like Figures 1-4 As shown, a method for intelligent online advertising push based on big data includes the following steps: Step S100: Use big data to predict the scenarios that users may enter within a preset future time window to obtain the target scenario.

[0022] In this step, we mainly rely on multi-dimensional data such as user historical behavior trajectory data, real-time location data, spatiotemporal geographic correlation data, and user scenario preference tags. Through conventional methods such as big data clustering analysis and trajectory prediction, we analyze the user's travel intentions and scenario switching patterns within a preset future time window, thereby predicting the scenario the user is about to enter and finally determining the target scenario. In order to reduce the system's computing pressure, the number of target scenarios is generally no more than 5. This step can be achieved based on existing mature big data prediction logic.

[0023] By determining the target scenario in advance through step S100, preparation time can be reserved for subsequent ad selection, while also alleviating the real-time calculation pressure on the system.

[0024] After step S100 is completed, proceed to step S200. In step S200, candidate ads that are suitable for the target scenario are selected from the ad library according to the target scenario, and a candidate ad list is generated. Multiple candidate ad lists are combined to form a candidate ad list set, where a candidate ad list set can hold up to 5 candidate ads for the target scenario.

[0025] Considering the potential for bias in predictions based on a single scenario, the target scenarios are limited to no more than five. By generating separate candidate ad lists for each scenario, we can cover users' potential travel intentions across multiple scenarios without overloading the system due to an excessive number of scenarios. This approach ensures a good match between individual ads and their corresponding scenarios while prioritizing high-performing ads with high historical click-through rates. This provides a more tailored pool of alternative ads for subsequent push notifications, addressing the issues of missed or incorrect pushes based on single-scenario predictions.

[0026] For each individual target scenario, generate a unique candidate ad list by following these steps: S211. Extract scene tags for the current target scene, such as outdoor scenic spots, commercial district restaurants, commuting routes, leisure parks, etc., traverse the ad library, filter out ads whose ad tags match the scene tags, remove completely irrelevant ads, and form the initial candidate ad pool for the scene. The formula for the matching score of a single ad tag is as follows:

[0027] In the formula, This represents the scene tag matching score of the k-th advertisement in the m-th target scene. The score directly reflects the degree of suitability of the advertisement with the current scene, with a maximum score of 100. The number of overlaps between the tags carried by the advertisement itself and the tags of the current target scene. For example, if the scene tags are outdoor scenic spot, family and children, daytime, and the advertisement tags are outdoor scenic spot and family play, then the number of matches is 2. The total number of tags preset for the current single target scene is a fixed parameter for scene labeling, such as the total number of tags for the scene mentioned above being 3.

[0028] This filter removes ads from the ad library that are irrelevant to the target scenario. Only ads that meet the score calculated by this formula can enter the subsequent sorting stage, serving as a threshold for the formation of a single candidate list. The preset screening threshold for a single ad tag matching score is 60 points. Ads with a score of ≥60 points are included in the initial candidate ad pool for that scenario, while those with a score below 60 points are directly excluded, preventing irrelevant ads from entering a single candidate list. By quantifying tag matching, it ensures that the ads in each single candidate ad list are highly relevant to the corresponding target scenario, controlling the quality of the list and preventing invalid ads from consuming push resources.

[0029] S212. The screening criteria adopt conventional tag matching rules to ensure that the advertising content is highly consistent with the scene attributes. For example, only cultural tourism and outdoor experience advertisements are screened in outdoor scenic spots, and only catering, retail and entertainment advertisements are screened in commercial districts.

[0030] S213. For each ad in the initial candidate ad pool, for each independent target scenario, calculate the comprehensive ranking score of the candidate ads in a single scenario by combining the scenario suitability score and the ad's historical click-through rate. Rank the ads according to their comprehensive scores within a single scenario. Sort the ads from highest to lowest quality, generating a separate list of candidate ads for each target scenario, ultimately forming a multi-scenario candidate ad list set. (Ad overall score) The calculation formula is as follows:

[0031] In the formula, This represents the overall ranking score of the k-th candidate ad in the m-th target scenario. The higher the value, the higher the ranking within the corresponding scenario. This indicates the scene suitability weight, which is preset manually. The specific preset value can be determined based on the ad popularity in the target scene to meet certain requirements. + =1; This represents the fit score between the k-th advertisement and the m-th target scenario, calculated using tag matching degree, with a value ranging from 0 to 100. This indicates historical click-through rate. This represents the historical average click-through rate of the k-th ad across all platforms, with a value ranging from 0 to 1, representing the number of clicks or impressions during the statistical period. This indicates the total number of target scenes, with a preset upper limit of 5. This indicates the candidate ad number within a single scenario, which is dynamically determined based on the ad library matching results.

[0032] By calculating the overall score of advertisements, we can solve the problem that relying solely on tags for filtering is not accurate enough. By adding the weight of historical click-through rate, we can prioritize advertisements with high user recognition and good conversion results, ensuring that advertisements are suitable for the scenario and improving the conversion probability of subsequent pushes. Scoring and sorting separately for each scenario is to ensure that each candidate list matches the user needs of the corresponding scenario and avoids push deviations caused by mixed advertisements from multiple scenarios.

[0033] S214. After all target scenarios have completed the above operations, the lists of each single scenario are summarized to obtain a multi-scenario candidate ad list set. In subsequent push notifications, the corresponding single candidate ad list can be called according to the scenario that the user actually enters.

[0034] In short, the candidate ad list is sorted based on the candidate ad's fit score with the target scenario and the historical click-through rate of each candidate ad, and the candidate ad list is generated according to the sorting results.

[0035] After the multi-scenario candidate ad list is generated, it waits for step S300 to be executed. Step S300 is executed when the user enters the target scenario. During execution, the user's current real-time environmental characteristics will be collected.

[0036] Specifically, step S300 relies on real-time positioning monitoring and third-party data interfaces to achieve fully automated execution of the entire process. First, it triggers the data collection action through scene access judgment: S311. Continuously acquire the real-time latitude and longitude coordinates uploaded by the user terminal, compare the coordinates with the preset geofence coordinates of each target scene predicted by S100, and if it is determined that the user's real-time coordinates fall within the geofence range of any target scene, it is determined that the user has entered the corresponding target scene and immediately issue an environmental collection command. S312. Acquire multi-dimensional real-time data through a layered acquisition method, including: Current weather data is collected in real time through the API of the national meteorological platform and the interface of commercial meteorological cloud services to obtain specific weather types such as sunny, rainy, cloudy, and foggy. Current geographic location data is collected through the fusion of user terminal GPS, Beidou satellite positioning or base station-assisted positioning, and accurately obtains geographic labels such as latitude and longitude, business district, road segment or region; Current traffic status data is collected in real time through connections with official traffic management platforms and internet map service providers, obtaining traffic information at four levels: smooth, slow, congested, and severely congested. Finally, feature output is completed through data validity judgment and normalization: the legality of the collected raw data is checked, invalid data that has expired or has abnormal values ​​is excluded, and then the data is normalized and normalized according to a unified format. One or more data from weather, geographical location, and traffic status are selected as needed and combined to form a standardized real-time environmental feature set.

[0037] After step S300 is completed, proceed to step S400, select one from the candidate ad list as the target ad, which is the first ad in the candidate ad list for the current region; then obtain the preset delivery conditions and estimated consumption time corresponding to the target ad. Estimated time consumption The calculation formula is:

[0038] In the formula, This represents the weight of the historical average dwell time, preset manually, and typically set to 0.7. This refers to the historical average length of stay at a tourist attraction or business; this data is calculated as an average based on big data statistics. For example, the historical dwell time of users in similar scenarios. The preset deployment conditions include one or more of the following: suitable weather type, suitable geographical location, and suitable traffic conditions.

[0039] Of course, the estimated time consumption can be determined not only by the historical average dwell time, but also by the type of target advertisement, the scenic spot corresponding to the target advertisement, and the user's historical behavior data. In this embodiment, it is determined by the historical average dwell time and the user's historical dwell time in similar scenarios.

[0040] After the estimated time is calculated, proceed to step S500 to obtain the duration prediction information of the current weather data, and obtain the weather duration prediction data. Weather duration forecast data The calculation formula is:

[0041] The raw data obtained through the API interface represents the duration of official meteorological forecasts. This is a weather system movement correction factor, a preset constant. The speed of weather system movement is obtained from the meteorological API interface; The weather system movement direction correction factor is 1 when approaching and 0.5 when moving away.

[0042] The formula uses the duration of large-scale static forecasts obtained from official meteorological platforms. As a base for calculation, the movement speed of weather systems is taken into account. Movement direction correction factor and manual preset correction coefficients The calculation of dynamic attenuation correction terms offsets the error of official forecasts ignoring the dynamic movement of weather systems by subtracting correction terms from the baseline value. This calibrates the large-scale macroscopic statistical values ​​to the true effective duration of weather in the user's current local scenario, accurately quantifying the duration of weather.

[0043] The primary purpose of calculating weather duration forecast data is to mitigate the risk of sudden weather changes in advance, preventing users from encountering abrupt weather changes while engaging in activities or making purchases related to the advertisement. This ensures users have sufficient good weather to complete their activities, significantly improving the actual conversion feasibility of ad pushes.

[0044] After obtaining the predicted duration of weather data, the estimated time is compared with the predicted duration to obtain the time difference. :

[0045] If the difference When the travel feasibility coefficient is greater than or equal to the first preset threshold, the travel feasibility coefficient is set to the first value. When the difference When the value is less than the first preset threshold and greater than the second preset threshold, the travel feasibility coefficient is set to the second value. When the difference When the travel feasibility coefficient is less than or equal to the second preset threshold, the travel feasibility coefficient will be set to the third value. In this embodiment, the first, second, and third values ​​decrease sequentially. After obtaining the comparison results, a travel feasibility coefficient is generated based on the comparison results. .

[0046] After obtaining the feasibility coefficient, step S600 is executed, which matches the real-time environmental features with the preset delivery conditions. This part compares the collected real-time environmental features with the pre-set delivery conditions of the target advertisement to determine whether the environment meets the advertising delivery requirements. In other words, a weighted comparison calculation is performed between the real-time environmental features and the preset delivery conditions of the target advertisement. Based on the calculation result, it is determined whether the preliminary matching is successful, and the preliminary matching result is obtained. Preliminary matching results The calculation formula is:

[0047] In the formula, It is a real-time environment matching score, with a value ranging from 0 to 100. The more the environment matches the delivery conditions, the higher the score. This is the environmental feature weight, which is preset manually with a default value of 0.8, and can be adjusted according to different usage scenarios; The system sets a threshold for initial matching; a match is considered successful when the score reaches this threshold.

[0048] After obtaining the initial matching results, the results are then corrected using a travel feasibility coefficient. Simply put, if environmental matching alone is insufficient to meet the push conditions, a second correction must be made based on the previously calculated travel feasibility coefficient. This way, even if the environment meets the requirements, if the good weather duration is insufficient or the user cannot complete the action corresponding to the advertisement, the push is directly judged as a failure, avoiding invalid pushes that appear to match but are actually unusable.

[0049] During the calculation process, a feasibility coefficient threshold is set in advance. ,like ≥ Final matching results Matching results Maintain consistency; if < Regardless of the initial matching results Whether it's a success or a failure, the final matching result All of them were failures.

[0050] Preferably, in step S600, real-time feedback behavior data of the user before the target advertisement is pushed can be obtained. The real-time feedback behavior data includes the user's click data, conversion data, and ignore data of historical advertisements. The real-time feedback behavior data and real-time environmental features are used together for matching and judgment.

[0051] After obtaining the final matching result, proceed to step S700. If the final matching result is successful, the target advertisement is pushed to the user terminal. If the final matching result is unsuccessful, another advertisement is selected from the candidate advertisement list as the new target advertisement, and steps S500 to S700 are executed again.

[0052] In the above implementation, steps S100 to S700 not only enable advance scenario prediction and ad selection, effectively distributing the system's real-time computing load and preventing push delays caused by temporary calculations after users enter the scenario, but also quantify the duration of weather conditions, avoiding the risk of users encountering sudden weather changes during their travels and consumption, thus effectively ensuring user experience. Furthermore, this solution employs a dual verification mechanism of environmental matching and time feasibility, fundamentally eliminating invalid pushes that appear to match but are actually unworkable. This saves advertisers' costs and reduces unnecessary information disturbance to users. It is adaptable to various advertising scenarios with different time consumption, such as scenic spots and fast food restaurants, significantly improving the adaptability and accuracy of ad pushes, ultimately effectively increasing ad click-through rates and conversion rates, while simultaneously enhancing user trust and stickiness with the platform.

[0053] As a further improvement to this embodiment, the following steps are also included: S800: The successfully pushed advertisements and their corresponding real-time environmental features and travel feasibility coefficients are used as positive samples, and the advertisements that are ultimately skipped after the push fails and their corresponding real-time environmental features and travel feasibility coefficients are used as negative samples and stored in the sample library. S900: Periodically use samples in the sample library to perform offline training and optimization of the weight parameters for matching judgment, and synchronize the optimized weight parameters to the online matching judgment process.

[0054] In simple terms, steps S800 and S900 involve marking successful ads that meet the environmental and time requirements, along with real-time environmental features and travel feasibility coefficients, as qualified positive samples. Similarly, ads that fail to be pushed or are ultimately skipped are marked as unqualified negative samples, along with corresponding environmental data and feasibility coefficients. All samples are stored in a dedicated sample library. These positive and negative samples are periodically retrieved from the sample library and trained using machine learning to optimize various weight parameters used in the matching process, such as scene adaptation weights, environment matching weights, and feasibility coefficient weights. This allows the model to continuously learn patterns from historical push data and understand the correlation between scenes, environments, weather durations, and user feedback. After training, the optimized new weight parameters are synchronized into the method, replacing the old parameters, making the entire push method increasingly accurate over time.

[0055] Steps S800 and S900 continuously absorb the successful and unsuccessful experiences of past push notifications, automatically revising the matching criteria. In the long run, this allows ad filtering to better align with real-world needs, reducing invalid push notifications, and continuously improving ad conversion rates. It establishes an operational logic of push notifications, data accumulation, optimization, and more precise push notifications, ensuring a steady and long-term improvement in the overall solution's push effectiveness. This achieves both increased ad conversion rates and saved advertisers' resources, while also reducing user disturbance and enhancing platform stickiness, realizing a dual advancement in technological effectiveness and commercial value.

[0056] Example 2: like Figure 5 As shown in one embodiment of the present invention, an online advertising intelligent push system based on big data reduces invalid pushes and effectively improves advertising conversion efficiency while improving user experience by dual screening of environmental adaptability and time feasibility and dynamically correcting the matching results.

[0057] One example is an online advertising intelligent push system based on big data, which includes a scene prediction module, an advertising filtering module, an environment acquisition module, a parameter acquisition module, a feasibility assessment module, a matching correction module, and a push determination module that are connected in sequence.

[0058] Specifically, the scene prediction module relies on multi-dimensional data such as user historical behavior trajectory data, real-time location data, spatiotemporal geographic correlation data, and user scene preference tags. Through big data clustering analysis and trajectory prediction methods, it analyzes the user's travel intentions and scene switching patterns within a preset future time window, predicts the scene the user is about to enter, and ultimately determines the target scene. To reduce the system's computational load, the number of target scenes is preset to a maximum of 5. The scene prediction module pre-determines potential user scenes, allowing preparation time for subsequent ad selection, alleviating the system's real-time computational pressure, and avoiding push delays caused by temporary calculations after the user enters the scene.

[0059] The ad filtering module receives the target scenarios output by the scenario prediction module. For each independent target scenario, it extracts scenario tags and iterates through the ad library. Ads matching the tags are then filtered to form an initial candidate ad pool. A comprehensive ranking score is calculated by combining the scenario suitability score with the ad's historical click-through rate. Ads are then sorted from highest to lowest score, generating an independent candidate ad list for each target scenario. Finally, these are aggregated into a multi-scenario candidate ad list set. By independently filtering and sorting ads by scenario, the ad filtering module covers users' travel intentions across multiple scenarios while ensuring ad suitability for each scenario. It prioritizes high-potential, high-quality ads, resolving the issues of missed or incorrect ad recommendations in single-scenario predictions.

[0060] The environmental data acquisition module continuously acquires real-time latitude and longitude coordinates uploaded by the user terminal, compares these coordinates with the preset geofence coordinates of the target scene, and immediately triggers the data acquisition action when it determines that the user has entered any target scene. By connecting to third-party data interfaces and the terminal positioning unit, it collects the user's current real-time environmental characteristics, including one or more of the following: current weather data, current geographical location data, and current traffic status data. It also performs legality verification on the collected raw data, removes invalid data that has timed out or has abnormal values, and outputs the data after standardization and normalization according to a unified format.

[0061] The parameter acquisition module is used to select the first advertisement as the target advertisement from the candidate advertisement list corresponding to the user's entry scene, retrieve the preset delivery conditions corresponding to the target advertisement, and calculate the estimated time required for the user to complete the corresponding behavior of the advertisement based on the type of the target advertisement, the historical average stay time of the corresponding scenic spot or merchant, and the user's historical behavior data.

[0062] The feasibility assessment module obtains the duration prediction information of current weather data. This duration prediction information is derived from meteorological forecast data and the movement speed and direction of the current weather system. It calculates the time difference between the predicted duration and the estimated time elapsed, and generates a travel feasibility coefficient based on the comparison of this difference with a preset threshold. This precise quantification of the effective duration of weather conditions helps mitigate the risk of sudden weather changes, ensuring users have sufficient time to complete their corresponding advertising actions and providing a core basis for subsequent push notification verification.

[0063] The matching correction module receives real-time environmental features from the environmental acquisition module and preset delivery conditions from the parameter acquisition module, performs a weighted comparison and matching judgment between the two to obtain a preliminary matching result. It then receives the travel feasibility coefficient from the feasibility assessment module and, using a preset coefficient threshold as a benchmark, performs a second correction on the preliminary matching result to obtain the final matching result. Optionally, this module can also access real-time feedback behavior data from users before the target advertisement is pushed, combining it with real-time environmental features to participate in the matching judgment, further improving matching accuracy.

[0064] The push decision module receives the final matching result output by the matching correction module. If the final matching result is successful, the target advertisement is pushed to the user terminal. If the final matching result is unsuccessful, the next advertisement is selected from the corresponding candidate advertisement list in sorted order as the new target advertisement. The parameter acquisition module, feasibility assessment module, and matching correction module are then triggered to re-execute the corresponding operations until an advertisement that meets the requirements is found or all candidate advertisements in the list have been traversed.

[0065] Preferably, an online advertising intelligent push system based on big data further includes a sample construction module and an iterative optimization module that are interconnected, and the iterative optimization module is also interconnected with the matching correction module. The sample construction module marks successfully pushed advertisements and their corresponding real-time environmental features and travel feasibility coefficients as positive samples, and marks advertisements that fail to push and are ultimately skipped by the user and their corresponding real-time environmental features and travel feasibility coefficients as negative samples, storing all positive and negative samples in a unified sample library. The iterative optimization module periodically retrieves positive and negative samples from the sample library, performs offline training and optimization of the weight parameters in the matching judgment process using machine learning, and synchronizes the optimized weight parameters to the matching correction module, completing the closed-loop iterative update of the model.

[0066] In this embodiment, the working principle of the online advertising intelligent push system based on big data is as follows: First, the scenario prediction module anticipates the user's potential target scenario and outputs it to the ad filtering module. The ad filtering module then filters and sorts the ads in advance, generating a corresponding candidate ad list. When the environment acquisition module determines that the user has entered the target scenario, it triggers the parameter acquisition module to retrieve the target ad's delivery conditions and estimated time consumption. Simultaneously, the feasibility assessment module generates a travel feasibility coefficient. After the matching and correction module completes environment matching and result correction, the push decision module executes ad push or cyclically changes the target ad. After the push process is completed, the sample construction module completes sample collection and storage, and the iterative optimization module periodically optimizes the model parameters.

[0067] In this embodiment, the beneficial effects of the big data-based online advertising intelligent push system include at least the following: By employing a dual verification mechanism that checks both environmental compatibility and time feasibility, ineffective ad pushes that appear to match but are actually unworkable are eliminated at the source. This not only improves ad conversion rates and saves advertisers' resources, but also reduces user disturbance and enhances platform stickiness, achieving a dual advancement in both technological effectiveness and commercial value.

[0068] It adopts an architecture that separates offline training and online push, which does not affect the real-time performance of online push. The modular design facilitates the independent upgrade and maintenance of each functional unit, and can flexibly adapt to different scenarios and types of advertising push needs, with strong versatility and scalability. During long-term operation, the push accuracy can be continuously optimized through a closed-loop iteration mechanism to ensure a long-term steady improvement in push effect.

[0069] It will be apparent to those skilled in the art that this disclosure is not limited to the details of the exemplary embodiments described above, and that this disclosure can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of this disclosure is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this disclosure. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0070] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for intelligent online advertising push based on big data, characterized in that, include: S100. Utilize big data to predict the scenarios that users may enter within a preset future time window to obtain the target scenario; S200: Based on the target scenario, select candidate ads that match the target scenario from the ad library and generate a candidate ad list; S300. When a user enters the target scene, the user's current real-time environmental characteristics are collected. The real-time environmental characteristics include one or more of the following: current weather data, current geographical location data, and current traffic status data. S400. Select one from the candidate advertisement list as the target advertisement, and obtain the preset delivery conditions and estimated consumption time corresponding to the target advertisement. The preset delivery conditions include one or more of the following: suitable weather type, suitable geographical location, and suitable traffic condition. S500: Obtain the duration prediction information of the current weather data, compare the estimated consumption time with the duration prediction information, and generate a travel feasibility coefficient based on the comparison result. S600. The real-time environmental features are matched and judged with the preset deployment conditions to obtain a preliminary matching result. The preliminary matching result is then corrected with the travel feasibility coefficient to obtain the final matching result. S700. If the final matching result is successful, the target advertisement is pushed to the user terminal. If the final matching result is unsuccessful, another advertisement is selected from the candidate advertisement list as the new target advertisement, and steps S500 to S700 are executed again.

2. The method for intelligent online advertising push based on big data according to claim 1, characterized in that, The estimated time consumption is determined based on the type of target advertisement, the historical average dwell time of the attraction or merchant corresponding to the target advertisement, and the user's historical behavior data.

3. The method for intelligent online advertising push based on big data according to claim 1, characterized in that, The duration prediction information is obtained based on meteorological forecast data, the current speed and direction of the weather system's movement.

4. The method for intelligent online advertising push based on big data according to claim 1, characterized in that, In step S500, a travel feasibility coefficient is generated based on the comparison results. Specific steps include: Calculate the difference between the predicted duration information and the estimated consumption time. When the difference is greater than or equal to a first preset threshold, set the travel feasibility coefficient to a first value. When the difference is less than the first preset threshold and greater than the second preset threshold, the travel feasibility coefficient is set to the second value; When the difference is less than or equal to the second preset threshold, the travel feasibility coefficient is set to the third value; The first, second, and third values ​​decrease sequentially.

5. The method for intelligent online advertising push based on big data according to claim 4, characterized in that, In step S600, the preliminary matching results are corrected using the travel feasibility coefficient. Specific steps include: When the travel feasibility coefficient is lower than the preset coefficient threshold, the preliminary matching result will be corrected to failure; When the travel feasibility coefficient is higher than or equal to the preset coefficient threshold, the initial matching result remains unchanged.

6. A method for intelligent online advertising push based on big data according to claim 1 or 5, characterized in that, Before performing the matching judgment in step S600, the following steps are also included: The system acquires real-time feedback behavior data of the user before the target advertisement is pushed, including the user's click data, conversion data, and ignore data of historical advertisements; the real-time feedback behavior data is then used in a matching and judgment process in conjunction with real-time environmental features.

7. The method for intelligent online advertising push based on big data according to claim 1, characterized in that, Also includes: S800: The successfully pushed advertisements and their corresponding real-time environmental features and travel feasibility coefficients are used as positive samples, and the advertisements that are ultimately skipped after the push fails and their corresponding real-time environmental features and travel feasibility coefficients are used as negative samples and stored in the sample library. S900. Periodically use samples in the sample library to perform offline training and optimization of the weight parameters for matching judgment, and synchronize the optimized weight parameters to the online matching judgment process.

8. The method for intelligent online advertising push based on big data according to claim 1, characterized in that, The candidate ad list is sorted according to the suitability score between the candidate ads and the target scene and the historical click-through rate of each candidate ad, and a candidate ad list is generated according to the sorting results.

9. A big data-based intelligent online advertising push system, used to execute the big data-based intelligent online advertising push method as described in any one of claims 1 to 8, characterized in that, include: The scene prediction module is used to use big data to predict the scenes that users may enter within a preset future time window, and obtain the target scene. The ad filtering module is used to filter candidate ads that are suitable for the target scenario from the ad library and generate a candidate ad list. The environment acquisition module is used to collect the user's current real-time environmental characteristics when the user enters the target scene. The real-time environmental characteristics include one or more of the following: current weather data, current geographical location data, and current traffic status data. The parameter acquisition module is used to select one as the target advertisement from the candidate advertisement list, and obtain the preset delivery conditions and estimated consumption time corresponding to the target advertisement. The preset delivery conditions include one or more of the following: suitable weather type, suitable geographical location, and suitable traffic status. The feasibility assessment module is used to obtain the duration prediction information of the current weather data, compare the estimated consumption time with the duration prediction information, and generate a travel feasibility coefficient based on the comparison result. The matching correction module is used to match and judge the real-time environmental features with the preset deployment conditions to obtain a preliminary matching result, and then correct the preliminary matching result with the travel feasibility coefficient to obtain a final matching result. The push determination module is used to push the target advertisement to the user terminal if the final matching result is successful. If the final matching result is unsuccessful, another advertisement is selected from the candidate advertisement list as the new target advertisement, and the feasibility assessment, matching correction and push determination operations are re-executed.

10. The intelligent online advertising push system based on big data according to claim 9, characterized in that, Also includes: The sample construction module is used to store successfully pushed advertisements and their corresponding real-time environmental features and travel feasibility coefficients as positive samples, and advertisements that are ultimately skipped after failed pushes and their corresponding real-time environmental features and travel feasibility coefficients as negative samples, and store them in the sample library. The iterative optimization module is used to periodically train and optimize the weight parameters for matching judgment offline using samples in the sample library, and synchronize the optimized weight parameters to the matching correction module.

Images