Live broadcast marketing traffic fluctuation prediction method and system based on data analysis

Abstract

The application discloses a live broadcast marketing traffic fluctuation prediction method and system based on data analysis and belongs to the technical field of network data analysis. The method comprises the following steps: extracting fluctuation shape features corresponding to various historical marketing activities, and constructing a traffic fluctuation mode set; determining a shape similarity score and an amplitude scaling ratio; generating a traffic burst trend description containing a predicted peak time; calculating a consistency strength index reflecting the degree of agreement; making the reserved resources declared in advance according to the burst trend description but not charged enter the formal operation charging state; and triggering early resource recovery when the actual value falls below half of the estimated peak value. The application solves the problem of false positive peak marking causing cloud resource mis-scheduling and reserved oscillation in the prior art by pre-constructing a traffic fluctuation mode set distinguished according to marketing event types.

Description

Technical Field

[0001] This invention relates to the field of network data analysis technology, and in particular to a method and system for predicting live marketing traffic fluctuations based on data analysis. Background Technology

[0002] As live-streaming marketing becomes the mainstream method for product promotion and user interaction, the traffic in live-streaming e-commerce rooms exhibits sudden and drastic fluctuations driven by marketing activities such as flash sales, red envelope giveaways, and surprise appearances by hosts. An event can generate several times the usual traffic surge within tens of seconds, only to rapidly subside afterward. To cope with these unpredictable traffic spikes, operators typically need to prepare sufficient computing and bandwidth resources in the cloud in advance to ensure smooth streaming and a real-time interactive experience for viewers, avoiding buffering or failed transactions due to congestion.

[0003] Existing live-streaming marketing traffic assurance systems continuously collect historical traffic data to train time-series prediction models and predict future traffic trends. Based on these predictions, they use numerical gradient change detection and local extrema identification to mark potential peak periods. Operators then use these peak markers to initiate batch expansion requests to the cloud resource management interface, starting server instances and reserving public network bandwidth in advance of the expected peak.

[0004] In existing technologies, the peak detection process does not differentiate between traffic patterns generated by different marketing activities, relying solely on general numerical fluctuations and rates of change for indiscriminate labeling. This results in a large number of false peaks caused by short-term random fluctuations or non-persistent minor disturbances being included in resource scheduling. These false peaks force the system to frequently perform unnecessary large-scale resource reservation and release operations. This not only causes reserved resources to be wasted during billing periods, but also triggers call frequency limits due to high-density calls to cloud interfaces in a short period, leading to the postponement or interception of expansion requests when real peaks arrive, thus weakening the system's timely response capability under real burst loads. Summary of the Invention

[0005] This application provides a data analysis-based method and system for predicting live marketing traffic fluctuations. It solves the problem in the prior art where indiscriminate detection of traffic patterns driven by different marketing activities leads to a large number of false positive spikes, causing frequent misscheduling of cloud resources, reservation oscillations, and obstructed expansion response under real sudden loads. It achieves resource activation control through event type pattern matching and phased delayed confirmation, thereby accurately completing resource readiness before the arrival of real traffic peaks and effectively suppressing resource idling and scaling cycles under false positive disturbances.

[0006] This application provides a data analysis-based method for predicting live marketing traffic fluctuations, including: extracting fluctuation pattern characteristics corresponding to various historical marketing activities and constructing a set of traffic fluctuation patterns; Extract local flow prediction segments from future continuous flow prediction sequences and determine the shape similarity score and amplitude scaling ratio; When the shape similarity score and amplitude scaling ratio meet the set conditions, a description of the sudden flow situation including the expected peak time is generated. Extract the real-time traffic growth rate before the expected peak time and calculate the consistency strength index, which reflects the degree of agreement. When the consistency strength index meets the continuous verification condition, the reserved resources that were declared in advance according to the emergency situation description but not billed are put into the formal operation and billing state. After the system is officially launched and billed, real-time traffic is monitored. When the actual traffic value drops to less than half of the estimated peak, early resource reclamation is triggered, and new resource occupancy of the same event type is rejected during the specified cooldown period.

[0007] Furthermore, the steps for constructing a set of traffic fluctuation patterns include: The collected historical traffic change sequences are categorized according to the nature of the marketing campaigns in the historical records; For each category label, the complete change process of traffic from deviating from the baseline level to restabilizing is captured as a combination of sample segments; The entire sample fragment combination of the same type is stretched and transformed to extract representative standard fluctuation features; The standard fluctuation characteristics include the rate of change of slope from the starting point to the peak, the length of time the slope is maintained at a high level, and the rate of change of slope during the descent. The deformation tolerance allowed by the standard fluctuation characteristics of each category is combined with the difference distribution pattern between historical actual values ​​and early predicted values ​​to form a set of flow fluctuation patterns.

[0008] Further steps in determining the shape similarity score and amplitude scaling ratio include: Obtain a continuous flow prediction sequence and set a fixed-length observation interval as a sliding window; The sliding window is controlled to move across the prediction sequence at fixed time intervals to extract local flow prediction segments; The local flow prediction segments are dynamically time-axis scaling and alignment comparisond with the pattern records stored in the flow fluctuation pattern set to find the best correspondence path between the local flow prediction segments and the standard fluctuation characteristics. Based on the found corresponding paths, a shape similarity score reflecting the degree of closeness of the overall undulating contours is calculated; Simultaneously, by comparing the elevation difference of the local flow prediction segment with the elevation difference of the standard form in the pattern record, the amplitude scaling ratio of the local flow prediction segment relative to the standard form is calculated for subsequent steps.

[0009] Furthermore, the steps for generating a description of the traffic surge situation, including the expected peak time, include: Determine whether the shape similarity score exceeds the preset first judgment threshold, and check whether the amplitude scaling ratio falls within the defined normal floating limit; If all conditions are met, the activity category tags of the matching pattern records are extracted, and combined with the expected peak time and the estimated peak amplitude, a description of the traffic burst situation with indications is generated. If the shape similarity score does not exceed the first judgment threshold, compare the fluctuation range of the local flow prediction segment with the preset second judgment threshold; If the fluctuation range is lower than the second judgment threshold, the segment will be discarded without further processing. When the fluctuation range exceeds the second judgment threshold, a provisional traffic surge description is generated with an empty activity category indicator and a low accuracy rating.

[0010] Furthermore, the process for generating reserved resources that are declared in advance according to contingency plans but are not charged includes: Continuously monitor the time span between the current time and the expected peak time recorded in the generated traffic surge situation description; When the time span narrows and falls within the planned preparation period, the server computing power level and number of servers required to support the traffic volume are calculated based on the estimated peak amplitude. Send a resource reservation request to the cloud resource scheduling center, which includes the aforementioned computing power level and number of servers; Add a special instruction symbol to pause billing in the resource placeholder request sent; After receiving the application, the dispatch center only locks the corresponding specifications and quantity of computing equipment in the background to ensure ownership, without issuing actual power-on instructions, so that the equipment is in a waiting state without consuming funds and becomes a reserved resource.

[0011] Furthermore, the steps for calculating the consistency strength index, which reflects the degree of fit, include: Before the expected peak time arrives, select a time node with a lead time as the starting point for observation; Starting from the observation start point, continuously acquire real-time traffic information actually generated in the current network; The actual rate of increase in traffic can be analyzed from real-time traffic information, and the geographical distribution of the originating locations of the current traffic can be analyzed simultaneously. Retrieve the standard rate of change of the upward slope and the corresponding common geographical location distribution characteristics under the original activity category label; By comparing the actual rate of ascent with the standard rate of change of the slope, and by verifying the overlap between the geographical distribution and common geographical features, a consistent intensity index is obtained.

[0012] Furthermore, the steps to bring reserved resources that were declared in advance as contingency scenarios but not yet billed into the formal operational and billing state include: Set a verification clearance threshold to measure the degree of agreement, and delineate multiple consecutive verification collection points on the timeline; Calculate the latest consistency strength index at each data collection time and observe whether the index crosses the verification clearance threshold; When the consistency intensity index calculated at multiple consecutive time points all crosses the permissible boundary, it is determined that the actual flow pattern matches the forecast and meets the continuous verification condition. At this point, a formal start command to unlock is sent to the cloud dispatch center via the network channel; The center, based on the activation command, grants operational permissions to the computing devices that were in a waiting state, transforming them into officially operational and billing devices that provide services to the outside world and begin consuming funds.

[0013] Furthermore, the steps that trigger early resource reclamation include: After the reserved resources enter the formal operation and billing state, maintain a tracking record of real-time traffic fluctuations; Retrieve the estimated duration of the high-level situation from the description of the emergency, and monitor whether the current time does not exceed the coverage of the estimated duration. During the period within the coverage area, continuously observe whether the real-time traffic value has dropped back to below half of the estimated peak value; It was confirmed that the downward trend remained stable over multiple consecutive monitoring periods and did not experience a significant rebound. Once confirmed, it is determined that the traffic peak has ended ahead of schedule, and a resource release notification is sent to the cloud dispatch center to stop the device's operating permissions, cut off the financial consumption, and trigger early resource recovery.

[0014] Furthermore, the steps for rejecting new resource preemptions of the same event type during the specified cooldown period include: After triggering the early resource reclamation action, a specified observation and waiting period is initiated to prevent frequent scaling fluctuations as a cooldown period. The duration of the cooling-off period is set to exceed the preset minimum limit to ensure that the backend system has time to recover smoothly. If, during the specified cooling-off period, the predictive analysis module generates a description of a sudden situation that belongs to the same activity category as the just-ended traffic peak; And it attempts to resubmit a new resource placeholder request to the cloud based on the newly generated description; Placement requests belonging to the same event type will be directly blocked, without response or forwarding to the dispatch center.

[0015] This application provides a data analysis-based live streaming marketing traffic fluctuation prediction system, used to implement a data analysis-based live streaming marketing traffic fluctuation prediction method, including: The system includes a traffic fluctuation pattern set construction module, a traffic prediction segment extraction module, a situation description generation module, a consistency strength index calculation module, a billing status entry module, and a resource reclamation trigger module. The traffic fluctuation pattern set construction module is used to extract the fluctuation pattern characteristics corresponding to various historical marketing activities and construct a traffic fluctuation pattern set. The flow prediction segment extraction module is used to extract local flow prediction segments from future continuous flow prediction sequences and determine the shape similarity score and amplitude scaling ratio. The situation description generation module is used to generate a traffic burst situation description including the expected peak time when the shape similarity score and amplitude scaling ratio meet the set conditions. The consistency strength index calculation module is used to extract the real-time traffic growth rate before the expected peak time and calculate the consistency strength index, which reflects the degree of consistency. The billing status entry module is used to enable reserved resources that were declared in advance according to the emergency situation description but not billed to enter the formal operation billing status when the consistency strength index meets the continuous verification condition. The resource reclamation trigger module is used to monitor real-time traffic after formal operation and billing. When the actual value drops to less than half of the estimated peak, it triggers early resource reclamation and rejects new resource occupancy of the same event type during the specified cooling period.

[0016] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By pre-extracting standard fluctuation features of historical traffic segments according to marketing event types and constructing a set of fluctuation patterns, the pattern records are matched in a dynamic alignment manner on the prediction sequence to generate a sudden situation description with event type tags. This makes the identification of traffic peaks no longer dependent on single numerical fluctuations, but has the ability to distinguish different activity rhythms, thereby reducing the situation of misjudging random fluctuations as significant peaks and providing a reference basis that is closer to the actual activity form for subsequent resource preparation.

[0017] Furthermore, when preparing resources based on the description of an emergency situation, a placeholder request is first placed in the cloud to lock computing devices without actually starting them. Then, just before the peak, a final decision is made based on the consistency between the real-time traffic increase rate and source distribution and the standard characteristics of the corresponding event type. Only after continuous confirmation of a match is the system switched to operational and billing status. This delayed confirmation process postpones the actual deployment of resources until real-time verification is obtained, avoiding the premature occupation of a large number of idle devices due to the uncertainty of the prediction results themselves, and reducing the capital consumption and resource idleness ratio after reservation.

[0018] Furthermore, after resources are put into regular operation, the estimated peak duration and actual traffic changes recorded in the emergency situation description are continuously referenced. If the actual value drops to less than half of the estimated peak and remains stable before the end of the expected duration, resource reclamation is initiated in advance, and new preemptive requests for the same event type are blocked during the subsequent cooling-off observation period. This approach can release equipment in a timely manner when traffic rapidly subsides and prevent high-frequency scaling cycles caused by immediate re-preemption after release, maintaining a relatively stable resource scheduling rhythm without affecting the actual peak. Attached Figure Description

[0019] Figure 1 Flowchart of a live-stream marketing traffic fluctuation prediction method based on data analysis provided in this application embodiment; Figure 2 A schematic diagram of the structure of a live marketing traffic fluctuation prediction system based on data analysis provided in this application embodiment. Detailed Implementation

[0020] This application provides a data analysis-based method and system for predicting live-stream marketing traffic fluctuations. This solves the problem in existing technologies where indiscriminate detection of traffic patterns driven by different marketing activities leads to numerous false positive spikes, causing frequent misscheduling of cloud resources, reserved oscillations, and hindered expansion response under real burst loads. By matching predicted traffic segments with pre-constructed fluctuation patterns based on event types to obtain a type-labeled description of traffic bursts, and performing phased resource activation—first locking in positions, then confirming the match based on real-time traffic patterns before transitioning to formal operation—this achieves accurate capture of real traffic spikes and timely resource readiness, while eliminating resource waste and high-frequency scaling cycles caused by false positive spikes.

[0021] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0022] like Figure 1The diagram shows a flowchart of a live marketing traffic fluctuation prediction method based on data analysis provided in this application embodiment. This method is applied to a live marketing traffic fluctuation prediction system based on data analysis. The method includes the following steps: extracting the fluctuation pattern characteristics of traffic deviation from the recovery baseline corresponding to various historical marketing activities, and constructing a set of traffic fluctuation patterns containing different event types. Extract local flow prediction segments from the future continuous flow prediction sequence, compare them with the entries in the flow fluctuation pattern set, and determine the shape similarity score and amplitude scaling ratio. When the shape similarity score and amplitude scaling ratio meet the set conditions, a description of the sudden flow situation, including the expected peak time, is generated according to the corresponding event type. The real-time traffic growth rate is extracted before the expected peak time and compared with the morphological features corresponding to the description of the traffic surge situation to calculate the consistency intensity index reflecting the degree of agreement. When the consistency strength index meets the continuous verification condition, the reserved resources that were declared in advance according to the emergency situation description but not billed are put into the formal operation and billing state. After the system is officially launched and billed, real-time traffic is monitored. When the actual traffic value drops to less than half of the estimated peak, early resource reclamation is triggered, and new resource occupancy of the same event type is rejected during the specified cooldown period.

[0023] Furthermore, the steps for constructing a set of traffic fluctuation patterns include: The collected historical traffic change sequences are categorized according to the nature of the marketing campaigns in the historical records; For each category label, the complete change process of traffic from deviating from the baseline level to restabilizing is captured as a combination of sample segments; The entire sample fragment combination of the same type is stretched and transformed to extract representative standard fluctuation features; The standard fluctuation characteristics include the rate of change of slope from the starting point to the peak, the length of time the slope is maintained at a high level, and the rate of change of slope during the descent. The deformation tolerance allowed by the standard fluctuation characteristics of each category is combined with the difference distribution pattern between historical actual values ​​and early predicted values ​​to form a set of flow fluctuation patterns.

[0024] In this embodiment, the first step is to establish an experience database, namely a set of traffic fluctuation patterns. The computer system extracts historical data from multiple live streams held by the company over the past year from the database. This data includes access traffic for each minute (e.g., requests per second).

[0025] In practice, technicians will label each segment of traffic data based on operational records, such as "the flash sale on May 20th" or "the Double Eleven red envelope rain". Then, the system will automatically locate the moment when traffic begins to rise significantly (deviating from the normal baseline) and the moment when traffic falls back to a stable level, capturing this complete fluctuation process to form sample segments.

[0026] To ensure comparability of features across live streams of different scales, these segments were normalized (i.e., their numerical proportions were stretched to a uniform range). Three key parameters were extracted: Slope: The gradient from the starting point to the highest peak, reflecting the speed of the flow burst.

[0027] Duration of high traffic volume: The period during which traffic remains near the peak (e.g., above 90% of the peak).

[0028] Descending slope: How quickly the flow rate recedes from its peak.

[0029] Furthermore, for multi-peak scenarios like "red envelope rain," the time intervals between peaks are also recorded. Finally, these features, along with their allowed range of deformation (e.g., slope fluctuation of 15%), are stored as pattern entries. This creates a morphological dictionary capable of identifying different marketing tactics.

[0030] Further steps in determining the shape similarity score and amplitude scaling ratio include: Obtain a continuous flow prediction sequence and set a fixed-length observation interval as a sliding window; The sliding window is controlled to move across the prediction sequence at fixed time intervals to extract local flow prediction segments; The local flow prediction segments are dynamically time-axis scaling and alignment comparisond with the pattern records stored in the flow fluctuation pattern set to find the best correspondence path between the local flow prediction segments and the standard fluctuation characteristics. Based on the found corresponding paths, a shape similarity score reflecting the degree of closeness of the overall undulating contours is calculated; Simultaneously, by comparing the elevation difference of the local flow prediction segment with the elevation difference of the standard form in the pattern record, the amplitude scaling ratio of the local flow prediction segment relative to the standard form is calculated for subsequent steps.

[0031] Furthermore, the steps for generating a description of the traffic surge situation, including the expected peak time, include: Determine whether the shape similarity score exceeds the preset first judgment threshold, and check whether the amplitude scaling ratio falls within the defined normal floating limit; If all conditions are met, the activity category tags of the matching pattern records are extracted, and combined with the expected peak time and the estimated peak amplitude, a description of the traffic burst situation with indications is generated. If the shape similarity score does not exceed the first judgment threshold, compare the fluctuation range of the local flow prediction segment with the preset second judgment threshold; If the fluctuation range is lower than the second judgment threshold, the segment will be discarded without further processing. When the fluctuation range exceeds the second judgment threshold, a provisional traffic surge description is generated with an empty activity category indicator and a low accuracy rating.

[0032] In this embodiment, during the live broadcast, a predicted traffic sequence for the next 30 minutes is continuously output. At this time, a sliding window (e.g., with a length of 15 minutes) is used to shift across the predicted sequence.

[0033] With each step, the predicted waveform in the window is compared with the standard pattern in the morphology dictionary. A time axis scaling alignment technique (i.e., dynamic time warping) is used, which means that even if the predicted flow burst is slightly slower or faster than the historical data, as long as the overall fluctuation pattern is similar, the system can recognize it.

[0034] The comparison will yield two core values: Shape similarity score: The higher the value, the more closely the contours match.

[0035] Amplitude scaling factor: This represents how many times the predicted peak value is compared to the standard model (e.g., 1.5 times).

[0036] If the shape similarity score exceeds the set first threshold (e.g., 0.85) and the amplitude scaling ratio is within a reasonable range, it will be determined as: "The forecast indicates a flash sale-like traffic surge in 5 minutes, with an estimated peak of 100,000 QPS, lasting for 10 minutes." This is the generated description of the traffic surge. If the similarity is very low but the traffic value is very large, it will be marked as an unknown surge, alerting operations and maintenance personnel to pay attention, thereby reducing false alarms caused by random small fluctuations.

[0037] Furthermore, the process for generating reserved resources that are declared in advance according to contingency plans but are not charged includes: Continuously monitor the time span between the current time and the expected peak time recorded in the generated traffic surge situation description; When the time span narrows and falls within the planned preparation period, the server computing power level and number of servers required to support the traffic volume are calculated based on the estimated peak amplitude. Send a resource reservation request to the cloud resource scheduling center, which includes the aforementioned computing power level and number of servers; Add a special instruction symbol to pause billing in the resource placeholder request sent; After receiving the application, the dispatch center only locks the corresponding specifications and quantity of computing equipment in the background to ensure ownership, without issuing actual power-on instructions, so that the equipment is in a waiting state without consuming funds and becomes a reserved resource.

[0038] Furthermore, the steps for calculating the consistency strength index, which reflects the degree of fit, include: Before the expected peak time arrives, select a time node with a lead time as the starting point for observation; Starting from the observation start point, real-time traffic information actually generated in the current network is continuously acquired through the data acquisition channel; The actual rate of increase in traffic can be analyzed from real-time traffic information, and the geographical distribution of the originating locations of the current traffic can be analyzed simultaneously. Retrieve the standard rate of change of the upward slope and the corresponding common geographical location distribution characteristics under the original activity category label; By comparing the actual rate of ascent with the standard rate of change of the slope, and by verifying the overlap between the geographical distribution and common geographical features, a consistent intensity index is obtained.

[0039] Furthermore, the steps to bring reserved resources that were declared in advance as contingency scenarios but not yet billed into the formal operational and billing state include: Set a verification clearance threshold to measure the degree of agreement, and delineate multiple consecutive verification collection points on the timeline; Calculate the latest consistency strength index at each data collection time and observe whether the index crosses the verification clearance threshold; When the consistency intensity index calculated at multiple consecutive time points all crosses the permissible boundary, it is determined that the actual flow pattern matches the forecast and meets the continuous verification condition. At this point, a formal start command to unlock is sent to the cloud dispatch center via the network channel; The center, based on the activation command, grants operational permissions to the computing devices that were in a waiting state, transforming them into officially operational and billing devices that provide services to the outside world and begin consuming funds.

[0040] In this embodiment, resources are not immediately purchased upon detecting a warning; instead, the process is divided into three steps: Phase 1: Placeholder Request. When the predicted peak time is approaching (e.g., 3 minutes in advance), a placeholder is first sent to the cloud provider. This request specifies the required server specifications and quantity, but includes a special flag indicating that it will not be started immediately. The cloud provider will reserve these resources in the background to ensure that others cannot take them away, but at this time the servers are not actually powered on or billed, and are in a standby state.

[0041] Phase Two: Lag Confirmation. In the tens of seconds before the expected traffic surge, begin monitoring the live feed. Extract the real-time traffic increase rate and observe the geographical locations of visitors.

[0042] Consistency strength index ( The calculation logic is as follows: ; in, It is the real-time ascent speed. It's the standard speed from the pattern library. Regional overlap weighting and It refers to adjusting parameters.

[0043] Phase Three: Delivery and Execution. Only when the consistency strength index... The power-on command is only issued when five consecutive sampling points consistently exceed the acceptable threshold. At this point, the resource, which was in a pre-emptive state, instantly becomes operational, begins billing, and starts handling traffic. This "pre-emptive, then verified, then powered on" logic ensures that charges are only incurred after traffic actually arrives.

[0044] Furthermore, the steps that trigger early resource reclamation include: After the reserved resources enter the formal operation and billing state, maintain a tracking record of real-time traffic fluctuations; Retrieve the estimated duration of the high-level situation from the description of the emergency, and monitor whether the current time does not exceed the coverage of the estimated duration. During the period within the coverage area, continuously observe whether the real-time traffic value has dropped back to below half of the estimated peak value; It was confirmed that the downward trend remained stable over multiple consecutive monitoring periods and did not experience a significant rebound. Once confirmed, it is determined that the traffic peak has ended ahead of schedule, and a resource release notification is sent to the cloud dispatch center to stop the device's operating permissions, cut off the financial consumption, and trigger early resource recovery.

[0045] Furthermore, the steps for rejecting new resource preemptions of the same event type during the specified cooldown period include: After triggering the early resource reclamation action, a specified observation and waiting period is initiated to prevent frequent scaling fluctuations as a cooldown period. The duration of the cooling-off period is set to exceed the preset minimum limit to ensure that the backend system has time to recover smoothly. If, during the specified cooling-off period, the predictive analysis module generates a description of a sudden situation that belongs to the same activity category as the just-ended traffic peak; And it attempts to resubmit a new resource placeholder request to the cloud based on the newly generated description; Placement requests belonging to the same event type are directly blocked, without response or forwarding to the dispatch center, thereby completely breaking the vicious cycle of repeated requests and releases.

[0046] In this embodiment, after the traffic peak has passed, real-time data is continuously monitored to prevent resource idleness. If the current traffic has dropped to less than half of the estimated peak and this situation continues for one minute, the activity is determined to have ended.

[0047] At this point, a release command is immediately issued to stop billing. However, to prevent drastic fluctuations such as immediate release followed by immediate re-application (which is detrimental to system stability), a "cooling-off period" (e.g., 10 minutes) is set.

[0048] During the cooldown period, if another flash sale alert of the same type is issued, the blocking system will filter it out directly. This is based on common business sense: the same live stream room usually doesn't hold two completely identical flash sales within a few minutes. This blocking mechanism protects the system from crashing due to frequent request-release operations and also avoids being restricted by cloud vendors on the frequency of API calls.

[0049] like Figure 2 The diagram shown is a structural schematic of the live marketing traffic fluctuation prediction system based on data analysis provided in this application embodiment. The live marketing traffic fluctuation prediction system based on data analysis provided in this application embodiment includes: a traffic fluctuation pattern set construction module, a traffic prediction segment extraction module, a situation description generation module, a consistency strength index calculation module, a billing status entry module, and a resource reclamation trigger module. The traffic fluctuation pattern set construction module is used to extract the fluctuation pattern characteristics corresponding to various historical marketing activities and construct a traffic fluctuation pattern set. The flow prediction segment extraction module is used to extract local flow prediction segments from future continuous flow prediction sequences and determine the shape similarity score and amplitude scaling ratio. The situation description generation module is used to generate a traffic burst situation description including the expected peak time when the shape similarity score and amplitude scaling ratio meet the set conditions. The consistency strength index calculation module is used to extract the real-time traffic growth rate before the expected peak time and calculate the consistency strength index, which reflects the degree of consistency. The billing status entry module is used to enable reserved resources that were declared in advance according to the emergency situation description but not billed to enter the formal operation billing status when the consistency strength index meets the continuous verification condition. The resource reclamation trigger module is used to monitor real-time traffic after formal operation and billing. When the actual value drops to less than half of the estimated peak, it triggers early resource reclamation and rejects new resource occupancy of the same event type during the specified cooling period.

[0050] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0051] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0052] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0053] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0054] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0055] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting live-stream marketing traffic fluctuations based on data analysis, characterized in that, Includes the following steps: Extract the fluctuation pattern characteristics corresponding to various historical marketing activities to construct a set of traffic fluctuation patterns; Extract local flow prediction segments from future continuous flow prediction sequences and determine the shape similarity score and amplitude scaling ratio; When the shape similarity score and amplitude scaling ratio meet the set conditions, a description of the sudden flow situation including the expected peak time is generated. Extract the real-time traffic growth rate before the expected peak time and calculate the consistency strength index, which reflects the degree of agreement. When the consistency strength index meets the continuous verification condition, the reserved resources that were declared in advance according to the emergency situation description but not billed are put into the formal operation and billing state. After the system is officially launched and billed, real-time traffic is monitored. When the actual traffic value drops to less than half of the estimated peak, early resource reclamation is triggered, and new resource occupancy of the same event type is rejected during the specified cooldown period.

2. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps to construct a set of traffic fluctuation patterns include: The collected historical traffic change sequences are categorized according to the nature of the marketing campaigns in the historical records; For each category label, the complete change process of traffic from deviating from the baseline level to restabilizing is captured as a combination of sample segments; The entire sample fragment combination of the same type is stretched and transformed to extract representative standard fluctuation features; The standard fluctuation characteristics include the rate of change of slope from the starting point to the peak, the length of time the slope is maintained at a high level, and the rate of change of slope during the descent. The deformation tolerance allowed by the standard fluctuation characteristics of each category is combined with the difference distribution pattern between historical actual values ​​and early predicted values ​​to form a set of flow fluctuation patterns.

3. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps for determining the shape similarity score and amplitude scaling ratio include: Obtain a continuous flow prediction sequence and set a fixed-length observation interval as a sliding window; The sliding window is controlled to move across the prediction sequence at fixed time intervals to extract local flow prediction segments; The local flow prediction segments are dynamically time-axis scaling and alignment comparisond with the pattern records stored in the flow fluctuation pattern set to find the best correspondence path between the local flow prediction segments and the standard fluctuation characteristics. Based on the found corresponding paths, a shape similarity score reflecting the degree of closeness of the overall undulating contours is calculated; Simultaneously, by comparing the elevation difference of the local flow prediction segment with the elevation difference of the standard form in the pattern record, the amplitude scaling ratio of the local flow prediction segment relative to the standard form is calculated for subsequent steps.

4. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps to generate a description of a traffic surge that includes the expected peak time are as follows: Determine whether the shape similarity score exceeds the preset first judgment threshold, and check whether the amplitude scaling ratio falls within the defined normal floating limit; If all conditions are met, the activity category tags of the matching pattern records are extracted, and combined with the expected peak time and the estimated peak amplitude, a description of the traffic burst situation with indications is generated. If the shape similarity score does not exceed the first judgment threshold, compare the fluctuation range of the local flow prediction segment with the preset second judgment threshold; If the fluctuation range is lower than the second judgment threshold, the segment will be discarded without further processing. When the fluctuation range exceeds the second judgment threshold, a provisional traffic surge description is generated with an empty activity category indicator and a low accuracy rating.

5. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The process of generating reserved resources that are declared in advance according to the contingency plan but are not charged includes: Continuously monitor the time span between the current time and the expected peak time recorded in the generated traffic surge situation description; When the time span narrows and falls within the planned preparation period, the server computing power level and number of servers required to support the traffic volume are calculated based on the estimated peak amplitude. Send a resource reservation request to the cloud resource scheduling center, which includes the aforementioned computing power level and number of servers; Add a special instruction symbol to pause billing in the resource placeholder request sent; After receiving the application, the dispatch center only locks the corresponding specifications and quantity of computing equipment in the background to ensure ownership, without issuing actual power-on instructions, so that the equipment is in a waiting state without consuming funds and becomes a reserved resource.

6. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps for calculating the consistency strength index, which reflects the degree of fit, include: Before the expected peak time arrives, select a time node with a lead time as the starting point for observation; Starting from the observation start point, continuously acquire real-time traffic information actually generated in the current network; The actual rate of increase in traffic can be analyzed from real-time traffic information, and the geographical distribution of the originating locations of the current traffic can be analyzed simultaneously. Retrieve the standard rate of change of the upward slope and the corresponding common geographical location distribution characteristics under the original activity category label; By comparing the actual rate of ascent with the standard rate of change of the slope, and by verifying the overlap between the geographical distribution and common geographical features, a consistent intensity index is obtained.

7. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps to bring reserved resources that were declared in advance as contingency scenarios but not yet billed into a formal operational and billed state include: Set a verification clearance threshold to measure the degree of agreement, and delineate multiple consecutive verification collection points on the timeline; Calculate the latest consistency strength index at each data collection time and observe whether the index crosses the verification clearance threshold; When the consistency intensity index calculated at multiple consecutive time points all crosses the permissible boundary, it is determined that the actual flow pattern matches the forecast and meets the continuous verification condition. At this point, a formal start command to unlock is sent to the cloud dispatch center via the network channel; The center, based on the activation command, grants operational permissions to the computing devices that were in a waiting state, transforming them into officially operational and billing devices that provide services to the outside world and begin consuming funds.

8. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps that trigger early resource reclamation include: After the reserved resources enter the formal operation and billing state, maintain a tracking record of real-time traffic fluctuations; Retrieve the estimated duration of the high-level situation from the description of the emergency, and monitor whether the current time does not exceed the coverage of the estimated duration. During the period within the coverage area, continuously observe whether the real-time traffic value has dropped back to below half of the estimated peak value; It was confirmed that the downward trend remained stable over multiple consecutive monitoring periods and did not experience a significant rebound. Once confirmed, it is determined that the traffic peak has ended ahead of schedule, and a resource release notification is sent to the cloud dispatch center to stop the device's operating permissions, cut off the financial consumption, and trigger early resource recovery.

9. The live-streaming marketing traffic fluctuation prediction method based on data analysis as described in claim 1, characterized in that, The steps for rejecting new resource preemptions of the same event type during the specified cooling-off period include: After triggering the early resource reclamation action, a specified observation and waiting period is initiated to prevent frequent scaling fluctuations as a cooldown period. The duration of the cooling-off period is set to exceed the preset minimum limit to ensure that the backend system has time to recover smoothly. If, during the specified cooling-off period, the predictive analysis module generates a description of a sudden situation that belongs to the same activity category as the just-ended traffic peak; And it attempts to resubmit a new resource placeholder request to the cloud based on the newly generated description; Placement requests belonging to the same event type will be directly blocked, without response or forwarding to the dispatch center.

10. A live-streaming marketing traffic fluctuation prediction system based on data analysis, used to implement the live-streaming marketing traffic fluctuation prediction method based on data analysis as described in any one of claims 1-9, characterized in that, include: The system includes a traffic fluctuation pattern set construction module, a traffic prediction segment extraction module, a situation description generation module, a consistency strength index calculation module, a billing status entry module, and a resource reclamation trigger module. The traffic fluctuation pattern set construction module is used to extract the fluctuation pattern characteristics corresponding to various historical marketing activities and construct a traffic fluctuation pattern set. The flow prediction segment extraction module is used to extract local flow prediction segments from future continuous flow prediction sequences and determine the shape similarity score and amplitude scaling ratio. The situation description generation module is used to generate a traffic burst situation description including the expected peak time when the shape similarity score and amplitude scaling ratio meet the set conditions. The consistency strength index calculation module is used to extract the real-time traffic growth rate before the expected peak time and calculate the consistency strength index, which reflects the degree of consistency. The billing status entry module is used to enable reserved resources that were declared in advance according to the emergency situation description but not billed to enter the formal operation billing status when the consistency strength index meets the continuous verification condition. The resource reclamation trigger module is used to monitor real-time traffic after formal operation and billing. When the actual value drops to less than half of the estimated peak, it triggers early resource reclamation and rejects new resource occupancy of the same event type during the specified cooling period.

Images