A live car marketing traffic fluctuation prediction method and system

Abstract

The application provides a live broadcast automobile marketing traffic fluctuation prediction method and system, relates to the cross-field fusion technical field, and the method comprises the steps of synchronously collecting data in a live broadcast platform and cross-domain expansion data, generating a platform data set and a cross-domain data set; based on the cross-domain data set, an association space of social hot events and vehicle attributes is constructed through a mirror algorithm, and when the event characteristic intensity of the association space exceeds a preset threshold, an event influence characteristic value is generated; based on the event influence characteristic value and the platform data set, a basic traffic prediction curve is generated at a fixed period and a key point set is selected; and a convex polygon region is generated by processing the key point set through a convex hull algorithm. Through the whole-link closed-loop design of data collection-event association-traffic prediction-parameter optimization-promotion triggering-speech adjustment, the application realizes accurate control of live broadcast automobile marketing traffic fluctuation and improvement of marketing effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cross-domain fusion, and particularly relates to a live automobile marketing traffic fluctuation prediction method and system. BACKGROUND

[0002] With the deep integration of live e-commerce and the automobile industry, live streaming has become a core channel for automobile brand promotion, vehicle marketing and user conversion. According to industry data, the live streaming transaction volume of automobiles in 2024 increased by more than 150% year-on-year. The proportion of users obtaining vehicle information, participating in interaction and completing test drive reservations through live streaming has significantly increased. Precise grasp of live streaming traffic fluctuation rules has become a key to improving marketing efficiency. However, current automobile live streaming traffic prediction and marketing decision-making may have the following technical challenges:

[0003] Some existing traffic prediction methods rely on data within the live streaming platform (such as the number of viewers, comment interactions, etc.), ignoring the correlation value of cross-domain information. Vertical automobile forum public opinion dynamics (such as user discussions on privacy settings and price sensitivity), and the changing trend of offline 4S store consultation volume have strong correlation with live streaming traffic. However, due to data format heterogeneity and inconsistent time granularity, it may be difficult to effectively integrate and capture the linkage effect of online live streaming-cross-domain public opinion-offline behavior. Automobiles are high-value consumer goods, and their live streaming traffic may be easily affected by social hot events (such as privacy and security policies, and competitor promotion activities). However, existing methods may lack a quantitative mechanism for associating events with vehicle attributes. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a live automobile marketing traffic fluctuation prediction method and system. Through the construction of a full-link closed-loop design of data collection-event association-traffic prediction-parameter optimization-promotion triggering-speech adjustment, precise control of live automobile marketing traffic fluctuation and improvement of marketing effectiveness are achieved.

[0005] To solve the above technical problems, the technical solutions of the present application are as follows:

[0006] In a first aspect, a live automobile marketing traffic fluctuation prediction method is provided. The method comprises:

[0007] Step 1: Synchronously collecting live streaming platform data and cross-domain extended data to generate platform data sets and cross-domain data sets;

[0008] Step 2: Based on the cross-domain data set, an association space between social hot events and vehicle attributes is constructed through a mirror algorithm. When the event feature intensity of the association space exceeds a preset threshold, an event influence feature value is generated.

[0009] Step 3, based on the event influence characteristic value and the platform data set, a basic traffic prediction curve is generated at a fixed period and a key point set is selected; a convex polygon region is generated by processing the key point set through a convex hull algorithm;

[0010] Step 4, when the convex polygon region is segmented into grid cells, the traffic change gradient of each grid cell is calculated synchronously to generate a gradient distribution matrix, and the vehicle type is identified; if it is a high-end vehicle, a mirror algorithm is called based on a historical feature library, a privacy label weight mapping relationship is established according to the traffic change gradient, and an optimized weight parameter is generated;

[0011] Step 5, based on the optimized weight parameter and the gradient distribution matrix, a grid adjustment characteristic quantity is generated; a comprehensive correction parameter is generated by fusing the event influence characteristic value and the grid adjustment characteristic quantity, and the basic traffic prediction curve is corrected to generate a corrected traffic curve; when the corrected traffic curve shows a traffic downward trend and the optimized weight parameter meets a preset threshold, a promotion trigger instruction is pre-generated;

[0012] Step 6, based on the pre-generated promotion trigger instruction and the cross-domain data set, the promotion trigger instruction is executed; at the same time, the cross-domain data set and the public opinion correlation degree of the live broadcast room comments are processed through the mirror algorithm, and when the correlation degree exceeds a preset threshold, a speech optimization instruction is executed.

[0013] In a second aspect, a live automobile marketing traffic fluctuation prediction system comprises:

[0014] The acquisition module is configured to synchronously acquire data in the live broadcast platform and cross-domain extension data, and generate a platform data set and a cross-domain data set;

[0015] The correlation module is configured to construct an association space between social hot events and vehicle attributes based on the cross-domain data set through a mirror algorithm, and generate an event influence characteristic value when the event characteristic intensity of the association space exceeds a preset threshold;

[0016] The convex hull algorithm module is configured to generate a basic traffic prediction curve at a fixed period and select a key point set based on the event influence characteristic value and the platform data set; a convex polygon region is generated by processing the key point set through a convex hull algorithm;

[0017] The optimized weight module is configured to calculate the traffic change gradient of each grid cell synchronously to generate a gradient distribution matrix when the convex polygon region is segmented into grid cells, and identify the vehicle type; if it is a high-end vehicle, a mirror algorithm is called based on a historical feature library, a privacy label weight mapping relationship is established according to the traffic change gradient, and an optimized weight parameter is generated;

[0018] The correction module is configured to generate grid adjustment characteristic values based on the optimization weight parameter and the gradient distribution matrix; generate comprehensive correction parameters by fusing event influence characteristic values and the grid adjustment characteristic values, and correct the basic flow prediction curve to generate a corrected flow curve; and pre-generate a promotion trigger instruction when the corrected flow curve shows a flow decline trend and the optimization weight parameter meets a preset threshold value;

[0019] The optimization instruction module is configured to execute the promotion trigger instruction based on the pre-generated promotion trigger instruction and the cross-domain data set, and execute a script optimization instruction by processing the cross-domain data set and the public opinion correlation degree of the live streaming room comments through a mirror algorithm when the correlation degree exceeds a preset threshold value.

[0020] In a third aspect, a computing device includes:

[0021] one or more processors;

[0022] a memory device storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the method.

[0023] In a fourth aspect, a computer-readable storage medium stores a program, which, when executed by a processor, implements the method.

[0024] The above-mentioned scheme of the present application at least has the following beneficial effects:

[0025] By synchronously collecting data in the live broadcast platform and cross-domain expansion data, the limitation of single data in the platform is broken. At the same time, through time axis alignment and feature fusion, the consistency of multi-source data in the time and space dimensions is ensured. With the help of mirror algorithm, the correlation space of social hot events and vehicle attributes is constructed, the abstract social hot spots are converted into quantifiable event influence characteristic values, and the dynamic influence of events on the traffic of different vehicles is intuitively reflected. For the privacy label period of high-end vehicles, the convex polygon area is refined by adding auxiliary connection points, and the description accuracy of traffic fluctuation in the privacy configuration attention period is improved. For the price sensitive period of economic vehicles, the connection path is simplified to reduce the calculation amount and optimize the efficiency. This combination of accurate area and efficient path not only ensures the prediction depth of the core scene of high-end vehicles, but also avoids resource waste. Based on the traffic change gradient and the historical high-end vehicle feature library, the mirror algorithm is used to dynamically adjust the privacy label weight parameter: when the traffic rises, the weight of the high attention label is increased, and when the traffic falls, the weight is reduced, to respond to the user interest fluctuation in real time. By generating a corrected traffic curve through comprehensive correction parameters, when continuous traffic decline is detected and the user's attention to the core label is high, a promotion trigger instruction is automatically pre-generated, realizing the closed loop of traffic warning-accurate promotion. While executing the promotion instruction, the mirror algorithm is used to associate cross-domain public opinion and live broadcast room comments. When the public opinion correlation degree exceeds the threshold, a speech optimization instruction is automatically generated and pushed in real time through the anchor teleprompter, reducing the user information gap and improving the interaction satisfaction and conversion efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of a live automobile marketing traffic fluctuation prediction method provided by an embodiment of the present application.

[0027] Figure 2 is a schematic diagram of a live automobile marketing traffic fluctuation prediction system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0029] As Figure 1 shown, an embodiment of the present application proposes a live automobile marketing traffic fluctuation prediction method, which comprises the following steps:

[0030] Step 1, synchronously collecting data in the live broadcast platform and cross-domain expansion data, generating platform data set and cross-domain data set;

[0031] Step 2, based on the cross-domain data set, the association space of social hot events and vehicle attributes is constructed by a mirror algorithm, and when the event characteristic intensity of the association space exceeds a preset threshold, an event influence characteristic value is generated;

[0032] Step 3, based on the event influence characteristic value and the platform data set, a basic traffic prediction curve is generated and a key point set is selected according to a fixed period; and a convex polygon region is generated by processing the key point set through a convex hull algorithm;

[0033] Step 4, when the convex polygon region is segmented into grid cells, a gradient distribution matrix is generated by synchronously calculating the traffic change gradient of each grid cell, and a vehicle type is identified; if it is a high-end vehicle, the mirror algorithm is called based on the historical feature library, a privacy label weight mapping relationship is established according to the traffic change gradient, and an optimized weight parameter is generated;

[0034] Step 5, based on the optimized weight parameter and the gradient distribution matrix, a grid adjustment characteristic quantity is generated; a comprehensive correction parameter is generated by fusing the event influence characteristic value and the grid adjustment characteristic quantity, and the basic traffic prediction curve is corrected to generate a corrected traffic curve; when the corrected traffic curve shows a traffic downward trend and the optimized weight parameter meets a preset threshold, a promotion trigger instruction is pre-generated;

[0035] Step 6, based on the pre-generated promotion trigger instruction and the cross-domain data set, the promotion trigger instruction is executed; at the same time, the cross-domain data set and the public opinion correlation degree of the live room comments are processed through the mirror algorithm, and when the correlation degree exceeds a preset threshold, a speech optimization instruction is executed.

[0036] In the embodiment of the application, the data in the live platform and the cross-domain extended data are synchronously collected to ensure the integrity of the platform data set and the cross-domain data set; the association space of social hot events and vehicle attributes is constructed by a mirror algorithm, which can accurately capture the influence of events on vehicles, making the generation of event influence characteristic values more targeted; based on the event influence characteristic value and the platform data set, a basic traffic prediction curve is generated, and in combination with the processing of the key point set and the convex polygon region, the traffic fluctuation trend is finely described; the convex polygon region is segmented into grid cells and the traffic change gradient is calculated, which can accurately reflect the traffic dynamics of different time periods and different regions, and through vehicle type identification and weight parameter optimization, the accuracy of traffic prediction is further improved; the comprehensive correction parameter is generated by fusing the event influence characteristic value and the grid adjustment characteristic quantity, the basic traffic prediction curve is corrected, and when the corrected curve shows a traffic downward trend and meets the conditions, the promotion trigger instruction is pre-generated, so that the promotion decision can timely respond to the risk of traffic loss; at the same time of executing the promotion trigger instruction, the cross-domain data set and the public opinion correlation degree of the live room comments are processed through the mirror algorithm, and when the correlation degree exceeds a threshold, the speech optimization instruction is executed, which can improve the interaction quality with users and enhance the attractiveness and persuasiveness of the promotion activities.

[0037] In a preferred embodiment of the present application, the step 1 of synchronously collecting the data in the live broadcast platform and the cross-domain extended data to generate the platform data set and the cross-domain data set can include:

[0038] Step 11, real-time collection of platform dynamic data from the live broadcast warm-up period to the end of the live broadcast to obtain original platform data;

[0039] Step 12, based on the original platform data, synchronously triggering the cross-domain data collection of the public opinion keyword set of the target vehicle model in the vertical automobile forum and the consultation quantity change trend data of the offline sales channel to obtain original cross-domain data;

[0040] Step 13, based on the original cross-domain data and the original platform data, aligning the original platform data according to the live broadcast time axis to generate time-aligned platform data; based on the time-aligned platform data, extracting an interaction rate-conversion rate correlation matrix; synchronizing the original cross-domain data with the aligned platform time axis to generate time-aligned cross-domain data;

[0041] Step 14, performing feature encoding on the time-aligned platform data to generate a platform data set containing a viewing peak distribution matrix, an interaction conversion correlation tensor, and a competitive product traffic time sequence feature vector; performing semantic aggregation on the time-aligned cross-domain data to generate a cross-domain data set containing public opinion keyword weight distribution data and offline consultation quantity gradient sequence.

[0042] In the embodiment of the present application, through full-period real-time collection, the integrity and timeliness of the original platform data are ensured, and accurate basic data support is provided for subsequent analysis; based on the target vehicle model, cross-domain collection is triggered to realize the correlation of online platforms, offline channels, and public opinion data, avoid invalid data, and improve data pertinence; through time axis alignment, the time granularity difference of multi-source data is eliminated, laying a foundation for cross-domain data fusion and correlation analysis, and ensuring the effectiveness of data comparison; through feature encoding and semantic aggregation, the original data is converted into a structured data set, and the data processing efficiency is improved.

[0043] In the embodiment of the present application, when specifically applied, the following technical solutions can be used to realize, for example:

[0044] The above step 11 sets a real-time collection triggering mechanism from the live broadcast warm-up period (such as 72 hours before the live broadcast) to 24 hours after the end of the live broadcast; through the live broadcast platform background interface, the original data such as the number of viewers, the cumulative viewing time, the number of comments, the number of likes, the number of sharing times, and the product link click volume are collected at a second level frequency and stored as a time-stamped original platform data sequence.

[0045] The above step 12 automatically activates the cross-domain data collection unit when the target vehicle model name appears in the original platform data (such as recognized by keyword matching).

[0046] Collecting vertical car forum data: crawling the forum posts and comments containing the target vehicle model, extracting public opinion keywords (such as "range", "price", "configuration", etc.), counting the frequency of each keyword, and forming a public opinion keyword set.

[0047] Collecting offline channel data: connecting to the offline 4S store management platform, counting the number of target vehicle model visits and phone consultations per hour, and generating original trend data of consultation volume over time, and forming a correlation mapping with platform data.

[0048] Step 13 above, using the live time axis (such as the broadcast time is 0 o'clock, divided by minutes) as the basis, the original platform data is aggregated by the same time granularity (such as every minute), generating time-aligned platform data (such as "10th minute viewership = average value of real-time data in that minute").

[0049] Extracting the interaction rate-conversion rate correlation matrix, which is based on time-aligned platform data, counting the ratio of "interaction behavior number (comments + likes)" to "conversion behavior number (product clicks)" in each period, and constructing a matrix by period x behavior type dimension (rows represent periods, columns represent interaction-conversion correlation values).

[0050] Aligning the original cross-domain data (forum keyword frequency, offline consultation volume) to the same granularity (minute level) of the platform time axis, such as splitting the hourly offline consultation volume to every minute, generating time-aligned cross-domain data.

[0051] Step 14 above, platform data set feature encoding:

[0052] The viewing peak distribution matrix is a two-dimensional table form encoding of the time-aligned platform data, counting the number of viewers in each period, and marking the specific time point of the peak value.

[0053] The interaction-conversion correlation tensor is a combination of the interaction rate-conversion rate correlation matrix and the time dimension, integrated by "period x interaction type x conversion efficiency" three-dimensional structure, forming a tensor data.

[0054] Competitor traffic time series feature vector is the viewing number data of the same period competitor live room, after time axis alignment, generating a vector for comparison with the target vehicle traffic (such as "target vehicle traffic-competitor traffic" difference sequence).

[0055] Cross-domain data set semantic aggregation:

[0056] The public opinion keyword weight distribution data is a public opinion keyword in the time-aligned cross-domain data, and a weight is assigned according to the occurrence frequency (the higher the frequency, the greater the weight), the weight proportion of each keyword in each period is counted, and a weight distribution graph is formed.

[0057] The offline consultation quantity gradient sequence is a difference value (the consultation quantity of the current period - the consultation quantity of the last period) of the consultation quantity of adjacent periods, and a gradient sequence reflecting the increasing and decreasing trend of the consultation quantity is generated, and a positive value represents an increase and a negative value represents a decrease.

[0058] In a preferred embodiment of the present application, the step 2, based on the cross-domain data set, constructs the association space of the social hot event and the vehicle model attribute through the mirror algorithm, and when the event feature intensity of the association space exceeds the preset threshold, the event influence feature value is generated, which can include:

[0059] Step 21, based on the cross-domain data set, analyzing the public opinion keyword weight distribution data, extracting a high-frequency keyword set; standardizing the offline consultation quantity gradient sequence to generate a consultation quantity change vector to obtain a preprocessed cross-domain data set;

[0060] Step 22, based on the preprocessed cross-domain data set, mapping the high-frequency keyword set to the vehicle model attribute space through the mirror algorithm, and performing tensor product operation on the consultation quantity change vector and the vehicle model attribute space to obtain a social hot spot-vehicle model attribute association space;

[0061] Step 23, calculating the feature vector length of each social hot event in the social hot spot-vehicle model attribute association space, normalizing the length to generate an event feature intensity value sequence;

[0062] Step 24, when any intensity value in the event feature intensity value sequence exceeds the preset dynamic threshold, triggering the feature value generation and converting the intensity value exceeding the threshold to the event influence feature value.

[0063] In the embodiment of the present application, by screening high-frequency keywords and standardizing the consultation quantity gradient, low correlation degree data is removed and the data scale is unified, so that the preprocessed cross-domain data focuses on the core public opinion and trend characteristics; with the help of the mirror algorithm, the hot keywords and the vehicle model attribute are accurately mapped, and the association space is constructed by combining the tensor product operation, so that the dynamic association relationship between the two is directly quantified; through the feature vector length calculation and normalization processing, the abstract social hot spot influence is converted into a comparable quantitative intensity value, so that the influence degree of different events has horizontal comparability; based on the dynamic threshold triggering feature value generation, the influence event is accurately captured, and the threshold is dynamically adjusted to adapt to different live scenes, improving the pertinence and effectiveness of the event influence feature value.

[0064] In the embodiment of the present application, when specifically applied, the following technical solutions can be used to realize, for example:

[0065] Step 21: Extract the weight distribution data of public opinion keywords in the cross-domain dataset, and calculate the weight proportion of each keyword in each time period (e.g., the weight proportion of "price" in time period 1 is 30%). Filter out keywords with a weight proportion exceeding a preset threshold (e.g., 20%), and collect these high-frequency keywords (e.g., "range", "intelligent driving", "preferential policy") to form a high-frequency keyword set.

[0066] Step 22: Map high-frequency keywords to vehicle attribute space: Call the mirror algorithm to match the high-frequency keyword set generated in step 21 with the preset vehicle attribute space (including core attribute dimensions of vehicle models, such as price range, power type, configuration level, target user group, etc.). For example, "preferential policy" corresponds to "price attribute", and "intelligent driving" corresponds to "configuration level attribute". This allows each keyword to correspond to a specific vehicle attribute dimension, completing the mapping of keywords to attribute space.

[0067] Step 23: In the social hotspot-vehicle attribute association space, convert the association data corresponding to each social hotspot event (e.g., "a celebrity recommends the same car model") into a feature vector (including dimensions such as keywords involved in the event, associated vehicle attributes, and consultation quantity change trend). Calculate the length of this vector (i.e., the square root of the sum of the squares of the values of each dimension of the vector, reflecting the overall strength of the event's impact).

[0068] Step 24: Normalize the length of the feature vector of all social hotspot events by the maximum value (i.e., divide each length by the maximum length), obtaining event feature intensity values in the 0-1 interval. Arrange them in chronological order to generate a sequence of event feature intensity values.

[0069] Step 25: In the social hotspot-vehicle attribute association space, convert the association data corresponding to each social hotspot event (e.g., "a celebrity recommends the same car model") into a feature vector (including dimensions such as keywords involved in the event, associated vehicle attributes, and consultation quantity change trend). Calculate the length of this vector (i.e., the square root of the sum of the squares of the values of each dimension of the vector, reflecting the overall strength of the event's impact).

[0070] Step 26: Normalize the length of the feature vector of all social hotspot events by the maximum value (i.e., divide each length by the maximum length), obtaining event feature intensity values in the 0-1 interval. Arrange them in chronological order to generate a sequence of event feature intensity values.

[0071] Step 27: In the social hotspot-vehicle attribute association space, convert the association data corresponding to each social hotspot event (e.g., "a celebrity recommends the same car model") into a feature vector (including dimensions such as keywords involved in the event, associated vehicle attributes, and consultation quantity change trend). Calculate the length of this vector (i.e., the square root of the sum of the squares of the values of each dimension of the vector, reflecting the overall strength of the event's impact).

[0072] The step 24 sets a dynamic threshold according to the intensity value range of the historical data, that is, the event significantly changes the traffic trend. For example, the threshold is 0.6 in a non-promotion period, the threshold is 0.4 in a promotion period, and the threshold is automatically adjusted according to the live broadcast stage.

[0073] The real-time monitoring intensity value sequence is automatically triggered to generate a feature value when any intensity value exceeds the preset dynamic threshold of the corresponding period. The intensity value is directly converted into an event impact feature value (the normalized value is retained for subsequent traffic prediction correction).

[0074] In a preferred embodiment of the present application, the step 3 generates a basic traffic prediction curve and selects a key point set based on the event impact feature value and the platform data set at a fixed period. The convex polygon region is generated by processing the key point set through a convex hull algorithm, which can include:

[0075] Step 31 injects the event impact feature value into the interaction conversion correlation tensor of the platform data set, and generates a spatiotemporal enhanced data set based on the fused tensor;

[0076] Step 32 generates a basic traffic prediction curve based on the spatiotemporal enhanced data set and executes time series prediction at a fixed period. The basic traffic prediction curve is corrected based on the historical high-end vehicle feature library to automatically select a key point set. The peak point corresponds to a promotion sensitive period, the valley point corresponds to a traffic loss risk period, and the inflection point corresponds to a public opinion mutation response period.

[0077] Step 33 filters the boundary extreme points in the key point set through a convex hull algorithm, and generates an initial convex polygon region by connecting in time sequence. In the connection process, auxiliary connection points are added to the key point set corresponding to the high-end vehicle privacy label period to improve the region accuracy. The connection path is simplified for the key point set corresponding to the price sensitive period of the economy vehicle to optimize the calculation efficiency, and a convex polygon region covering the core traffic fluctuation area is generated.

[0078] In the embodiment of the present application, the fusion of event features and interaction conversion data enhances the spatiotemporal correlation of the data set, solves the problem of lack of correlation and incomplete information of single platform data, and provides more comprehensive feature support for accurate prediction. The fixed period prediction combined with historical feature correction improves the accuracy of the basic traffic curve. The classification and identification of the key point set make the key period of traffic fluctuation clearer. The differential region optimization based on the type of vehicle improves the accuracy of the high-end vehicle privacy period and improves the calculation efficiency of the economy vehicle. Finally, the generated convex polygon region can accurately cover the core traffic fluctuation area, laying a reliable foundation for grid segmentation and traffic analysis.

[0079] In the embodiment of the present application, when specifically applied, the following technical solutions can be used to achieve the technical effects, for example:

[0080] Step 31: Extract the interaction conversion correlation tensor of the platform dataset, which contains feature data of three dimensions: time period, interaction type, and conversion type. Align the event impact feature values by time dimension with the interaction conversion correlation tensor, and inject the tensor by dimension expansion method (such as adding an "event impact strength" dimension), so that the tensor contains both original interaction conversion features and event impact features.

[0081] Integrate the injected tensor in space-time dimensions, divide the data block according to the live broadcast time axis and user interaction scenarios (such as consultation scenarios and order scenarios), and generate a space-time enhanced dataset.

[0082] Step 32: Divide the entire live broadcast into equal-length time windows (such as every 30 minutes as a period), and each period as a prediction unit to form a continuous time series.

[0083] For each prediction period, use the data of the previous N periods (such as the previous 3 periods) as the history window to extract the following features:

[0084] Traffic statistics features: average number of viewers, maximum / minimum number of viewers, and standard deviation of the number of viewers (reflecting the fluctuation amplitude) within the window.

[0085] Interaction conversion features: change trend of interaction rate (number of comments / viewers) and conversion rate (clicks / viewers) within the window.

[0086] Event impact features: cumulative intensity and change slope (rising / falling trend) of event impact feature values within the window.

[0087] According to the extracted historical features and event impact feature values (such as the weight of public opinion keywords and offline consultation volume gradient) of the current period, generate a basic traffic prediction curve for the future period.

[0088] The historical high-end car model feature library is a structured feature set formed by mining, extracting, and verifying a large amount of historical high-end car model live broadcast data, which is used to support the traffic prediction correction of the current live broadcast; its establishment process is as follows:

[0089] Collect live broadcast records of all high-end vehicles (such as cars / SUVs priced ≥ 50 million yuan) in the past 1-2 years, covering different brands (such as Mercedes-Benz, BMW, Tesla high-end series), different live broadcast scenarios (such as new car launch, promotion special); real-time viewership, interaction data (comments / likes / shares), conversion data (click-to-try-drive / inquiry times), user dwell time, scroll keywords (such as "privacy glass" "rear row sound insulation"); corresponding live broadcast period vertical automobile forum public opinion (such as "Car Home" high-end vehicle board post keywords), offline 4S store consultation volume (such as "percentage of customers visiting the store to inquire about privacy configuration"), social hot events (such as "high-end vehicle privacy and security policy introduction").

[0090] From the collected historical data, the following core features (classified by business scenario) are extracted: traffic fluctuation features: traffic peak value in privacy label period (such as discussing "privacy configuration") compared to ordinary period (historical average is 1.3:1); traffic peak duration (high-end vehicles average 28 minutes, 10 minutes longer than economy vehicles); negative public opinion (such as "privacy leakage risk") appears when the traffic decreases at a rate of (historical average is -30 people / minute).

[0091] Time-sensitive features: historical data shows that high-end users watch 42% of the time between 20:00 and 21:30 on weekdays (after work); users average 19 minutes of dwell time, with 70% of dwell time concentrated in the "configuration explanation" segment.

[0092] Public opinion-related features: after the "privacy configuration" keyword appears, the traffic peak value appears on average 8 minutes later (users need time to digest the information); for every 10% increase in "smart privacy" keyword weight, traffic peak value increases by 12% (historical statistics).

[0093] Conversion features: privacy label period conversion rate (click-to-inquiry / viewership) is 17% higher than ordinary period; offline consultation volume and online traffic are linked: for every 5 additional offline consultations, online traffic peak value increases by an average of 8% (users prefer "online understanding + offline verification").

[0094] Remove extreme data (such as traffic drop due to live broadcast failure), identify and remove abnormal feature values outside "mean ± 3 standard deviations" using IQR (interquartile range) method; unify features into "feature name-time granularity-statistical value" format, such as "privacy period traffic peak value ratio-30 minute period-1.3" "keyword response delay-minutes-8", ensuring that data from different historical live broadcasts can be directly compared.

[0095] Select 3-5 typical historical live broadcasts (such as the launch of a high-end vehicle), use the extracted features to reverse the traffic curve, and verify the accuracy of the features; for example, if the feature library shows that "privacy keyword weight increases by 10% → traffic peak +12%", then backtrack the actual data of the live broadcast, if the deviation is ≤5%, the feature is valid; test the universality of the feature in different brands / scenarios (such as "high attention to privacy configuration" whether it applies to all high-end vehicles), and label brand-specific features (such as the traffic correlation of a brand's "star sky screen" configuration) separately.

[0096] Add nearly 3 months of high-end vehicle live broadcast data every quarter, repeat steps 2-4, and update the feature library (such as adding privacy features for "new energy high-end vehicles"); when new user behavior trends appear (such as the impact of "meta universe test drive" on traffic), supplement new feature dimensions to ensure the timeliness of the library.

[0097] Record the traffic curve patterns of high-end vehicle live broadcasts in history, such as the traffic peak during the privacy label period (such as discussing "privacy glass" and "soundproof configuration") is usually 20%-30% higher than that during ordinary periods, and the fluctuation frequency is lower (the duration is longer).

[0098] Traffic changes in specific periods, such as the traffic peak of high-end vehicles at 8-10 pm on weekdays is 50% higher than that during the day, while there is no significant difference for economy vehicles; record the response mode of the traffic curve when specific public opinion keywords related to high-end vehicles appear (such as "luxury feel" and "technology configuration") (such as the traffic peak appears 15 minutes later when the keyword weight increases by 10%); when competitors launch similar models, the anti-interference ability of high-end vehicle traffic curves (such as the decrease in high-end vehicle traffic is 15% lower than that of economy vehicles).

[0099] Compare the current generated basic traffic prediction curve with the typical curve in the historical high-end vehicle feature library, calculate the similarity (such as through the dynamic time warping DTW algorithm); if the current curve's peak in the privacy label period is lower than the historical average, adjust it by: increasing the predicted traffic value in the privacy label period by 15%-20%; reduce the derivative change rate of the curve in this period to make the peak closer to the historical typical pattern.

[0100] The first derivative of the curve is calculated, the points where the derivative is 0 are marked as peak points (the highest flow point) or trough points (the lowest flow point), and the points where the derivative suddenly changes are marked as inflection points (the flow trend changes), and are classified as "promotion sensitive period (peak point), flow loss risk period (trough point), public opinion mutation response period (inflection point)", forming a key point set; wherein the promotion sensitive period (peak point) is the highest flow point, which usually corresponds to the most active audience and the strongest purchase willingness period, which is suitable for launching promotion activities (such as time-limited discount); the flow loss risk period (trough point) is the lowest flow point, which may indicate that the audience interest is declining or the product is diverted, and the live broadcast content needs to be adjusted in advance; the public opinion mutation response period (inflection point) is the point where the flow trend suddenly changes, which is usually related to hot events or negative public opinion, and the language needs to be adjusted in time or the interaction link needs to be increased.

[0101] The above step 33 sorts the key point set according to the flow fluctuation amplitude, screens out the boundary feature points of fluctuation (such as the top 20% of peak points / trough points / inflection points), connects these points through a convex hull algorithm to form an initial convex polygon region covering the main flow fluctuation range.

[0102] For high-end vehicles: identify the key point set corresponding to the privacy label period (such as the peak points / inflection points associated with the "privacy configuration" public opinion), and insert interpolation auxiliary points in the time interval between adjacent key points to increase the vertex density of the region to refine the flow fluctuation description of the privacy period.

[0103] For economy vehicles: identify the key point set corresponding to the price sensitive period (such as the trough points / inflection points associated with "preferential activities"), and merge adjacent non-extreme original sampling points to simplify the connection path (such as replacing the polyline with a straight line) to reduce the calculation amount.

[0104] Integrate the adjusted key point set to reconnect to form the final convex polygon region, ensuring that it covers the core flow fluctuation area (such as containing more than 80% of the key point set).

[0105] In a preferred embodiment of the present application, the above step 4, when the convex polygon region is divided into grid cells, the flow change gradient of each grid cell is calculated to generate a gradient distribution matrix, and the vehicle type is identified; if it is a high-end vehicle, the mirror algorithm is called based on the historical feature library, the privacy label weight mapping relationship is established according to the flow change gradient, and the optimization weight parameter is generated, which can include:

[0106] Step 41, based on the convex polygon region, divide it into grid cells along the time axis, and extract the flow data and time period information contained in each grid cell to generate grid cell data; calculate the flow change rate of each grid cell to generate a gradient distribution matrix;

[0107] Step 42, based on the grid cell data and the gradient distribution matrix, the user portrait features in the grid cell are extracted, including the user query behavior on the vehicle configuration, the interaction frequency; the historical high-end vehicle feature library is called for feature comparison, the privacy configuration query frequency is counted, when the frequency is greater than the reference value in the historical high-end vehicle feature library, it is marked as a high-end vehicle, otherwise, it is marked as a mass vehicle, to obtain the vehicle identification;

[0108] Step 43, when the vehicle identification is a high-end vehicle, the gradient distribution matrix is called and based on the historical high-end vehicle feature library, the privacy label weight mapping relationship is established by the mirror algorithm, that is, the gradient change rate is taken as the independent variable, the historical weight of the privacy label in the feature library is taken as the base value, the weight value is dynamically adjusted according to the gradient change direction, the weight increase is increased when the gradient rises, and the weight decrease is applied when the gradient falls, and the optimized weight parameter is generated.

[0109] In the embodiment of the application, the continuous traffic fluctuation area is discretized by grid cell segmentation, so that the traffic change analysis is more precise; the gradient distribution matrix is calculated synchronously, the traffic change rate of each period is quantified intuitively, and data basis is provided for subsequent weight adjustment; the vehicle type is distinguished based on the user query behavior on the privacy configuration, the high attention feature of the high-end vehicle user to the privacy attribute is matched, the vehicle identification is more accurate; by comparing with the reference value of the historical feature library, the objectivity of classification is improved; the privacy label weight is dynamically adjusted combined with the traffic gradient, so that the weight parameter can adapt to the traffic change trend in real time.

[0110] In the embodiment of the application, when specifically applied, the following technical solutions can be used to realize, for example:

[0111] The above step 41, the time axis (horizontal axis) and the flow value axis (vertical axis) along the convex polygon area are equidistantly segmented to form a plurality of rectangular grid cells (such as each grid corresponds to a 5-minute time interval + 50-person flow interval), each grid cell corresponds to a specific space-time range.

[0112] The traffic data (such as the average number of viewers in the period, the maximum flow value) and the period information (such as the starting time, the duration) contained in each grid cell are counted, stored in the format of "grid number-flow data-period", and the grid cell data is generated.

[0113] For adjacent grid cells, the difference value of the flow value (current grid flow-previous grid flow) is calculated, and then divided by the time interval (such as 5 minutes) to obtain the flow change rate (positive value indicates rising, negative value indicates falling) of each grid cell; all the rates are arranged according to the position to form a gradient distribution matrix (row represents time sequence, column represents flow interval, and matrix element is the change rate of the corresponding grid).

[0114] Step 42 above extracts user behavior features from the grid cell data:

[0115] Configure query behavior: Count the frequency of vehicle configuration keywords (such as "privacy glass", "encrypted storage", "soundproof cotton", etc. privacy configuration, "price", "fuel consumption", etc. popular configuration) mentioned by users in comments and bullet screen in each grid cell.

[0116] Interaction frequency: Count the number of comments, likes, and shares of users in each grid cell, and calculate the interaction density (total number of interactions / number of viewers in the grid).

[0117] Call the historical high-end vehicle feature library and extract the "privacy configuration query benchmark value" (such as the average query frequency of privacy configuration in historical high-end vehicle live streaming is 15 times per grid).

[0118] Statistical total query frequency of privacy configuration in current grid cell data, if the frequency > benchmark value (such as current 20 times > 15 times), mark the vehicle as "high-end vehicle"; otherwise, mark it as "popular vehicle", and generate vehicle identification.

[0119] The above step 43 filters out two types of key grid cells from the gradient distribution matrix: one is the grid marked as "high-end vehicle", and the other is the time period when users frequently query privacy configurations (such as "privacy glass", "encrypted storage", "rear soundproofing", etc.) in this grid; For these grid cells, extract their specific features of traffic change gradient, i.e. traffic change rate per unit time (such as +8 people / minute indicates traffic increase, -5 people / minute indicates traffic decrease).

[0120] From the historical high-end vehicle feature library, match the privacy labels involved in the current live streaming (such as "privacy glass", "encrypted storage"), and extract the corresponding "historical base weight"; These weights are benchmark values calculated based on a large number of historical high-end vehicle live streaming data, representing the long-term attention of users to different privacy labels in the past; For example: "privacy glass" is frequently mentioned by high-end users in historical live streaming, and the base weight is set to 0.6 (i.e. the historical average influence degree of this label on traffic fluctuation is 60%); "encrypted storage" is a characteristic attention item of new energy high-end vehicles, and the base weight is set to 0.5 (historical average influence degree is 50%).

[0121] Based on the traffic change gradient of the grid cell, the historical base weight is real-time corrected, divided into two scenarios:

[0122] When the gradient rises (the rate is positive): if the flow rate of change is positive (such as +8 people / minute), it means that the current user's attention to the privacy label is increasing. At this time, the weight is increased by the "gradient rise amplitude": for every 2 people / minute rate, the weight is increased by 5% on the basis of history; for example, the historical basis weight of "privacy glass" is 0.6, and the current gradient rate is +8 people / minute (i.e. 4 "2 people / minute" units more than the basis rate), so the weight adjustment is: basis weight 0.6+(4x5% x 0.6)=0.6+0.12=0.72, which is actually a proportional cumulative increase, that is, every 2 people / minute corresponds to a 5% increase, 8 people / minute corresponds to 4 increase units, and the total increase is 4x5%=20%, so the adjusted weight is 0.6x(1+20%)=0.72.

[0123] When the gradient falls (the rate is negative): if the flow rate of change is negative (such as -6 people / minute), it means that the user's attention to the privacy label is decreasing. At this time, the weight is reduced by the "gradient decline amplitude": for every 2 people / minute rate, the weight is reduced by 3% on the basis of history; for example, the historical basis weight of "encryption storage" is 0.5, and the current gradient rate is -6 people / minute (i.e. 3 "2 people / minute" units of decline more than the basis rate), the total decline is 3x3%=9%, so the adjusted weight is 0.5x(1-9%)=0.455(approximately 0.46).

[0124] Integrate the weights of all privacy labels after gradient adjustment to form a set of optimized weight parameters that adapt to the current traffic trend; for example, the adjusted weight of "privacy glass" is 0.72, and the adjusted weight of "encryption storage" is 0.46, and other privacy labels (such as "rear row sound insulation") are adjusted according to the same logic to form the final optimized weight parameters.

[0125] In a preferred embodiment of the present application, step 5, based on the optimized weight parameters and the gradient distribution matrix, generates the grid adjustment feature quantity; fuse the event influence feature value and the grid adjustment feature quantity to generate the comprehensive correction parameter, and correct the basis flow prediction curve to generate the corrected flow curve; when the corrected flow curve shows a downward trend in traffic and the optimized weight parameter meets the preset threshold, a pre-generated promotion trigger instruction can be included:

[0126] Step 51, based on the optimized weight parameters and the gradient distribution matrix, extract the change rate feature through the spatiotemporal characteristics of the gradient distribution matrix, and aggregate the change rate feature based on the optimized weight parameters to obtain the grid adjustment feature quantity;

[0127] Step 52, based on the grid adjustment feature quantity and the event influence feature value, map the event influence feature value to a preset feature space, and perform tensor fusion with the grid adjustment feature quantity to obtain the comprehensive correction parameter;

[0128] Step 53, inject the comprehensive correction parameter into the fluctuation interval of the basic flow prediction curve to generate a corrected flow curve;

[0129] Step 54, when it is detected that the flow in the corrected flow curve decreases by more than a preset amplitude in two consecutive periods, and it is verified that the optimization weight parameter is greater than the vehicle model sensitivity threshold, a pre-generated promotion trigger instruction is generated.

[0130] In the embodiment of the application, the optimization weight parameter is combined with the gradient feature through weighted aggregation, so that the grid adjustment feature quantity is more focused on the high-attention privacy label related grid, and the feature is improved in pertinence to the high-end vehicle user behavior; the event influence and the grid feature are fused in a unified feature space, avoiding the limitation of a single feature, so that the comprehensive correction parameter covers the influence of external events and internal flow changes, and the comprehensiveness of correction is enhanced; the correction parameter is injected into the fluctuation interval of the curve to solve the problem that the basic prediction curve ignores the comprehensive influence of multiple factors, so that the corrected flow curve is more consistent with the actual flow change rule, and the prediction accuracy is improved; the promotion instruction is triggered based on the double conditions of flow trend and weight threshold, which not only ensures timely intervention when the flow decreases, but also filters out the high-end user group sensitive to the privacy label through the weight threshold, avoids invalid promotion, and improves the utilization rate of marketing resources.

[0131] In the embodiment of the application, when specifically applied, the following technical solutions can be used to achieve the above-mentioned technical effects, for example:

[0132] The above-mentioned step 51 extracts two types of change rate features from the gradient distribution matrix:

[0133] The time dimension feature: the average flow change rate (such as “the average gradient of the 2-3 period is +5 people / minute”) of the grid unit in each period is calculated according to the live broadcast time axis (such as every 30 minutes as a period), and the consistency of the change trend (such as the proportion of the consecutive 3 grids that all show an upward trend).

[0134] The space dimension feature: the gradient standard deviation (reflecting the stability of the flow fluctuation in the interval) in the same interval and the gradient difference (such as the gradient difference between grid A and grid B is 3 people / minute) of adjacent grids are calculated according to the flow interval (such as high flow interval, low flow interval) of the grid unit.

[0135] The optimization weight parameter (such as “privacy glass” weight 0.65, “encrypted storage” weight 0.48) is called to weight the extracted change rate features: the weight of the change rate feature of the grid associated with the high-weight privacy label (such as the “privacy glass” related grid) is increased (such as multiplied by 1.2 times); the weight of the change rate feature of the low-weight or irrelevant grid is reduced (such as multiplied by 0.8 times).

[0136] After weighting, the time and space dimension features are integrated into a multi-dimensional vector to form a grid adjustment feature quantity (including the weighted flow change features of each period and each interval).

[0137] The step 52 above presets a feature space containing three dimensions of time axis (aligned with the live broadcast period), event type (such as social hot spot, public opinion event), and influence intensity (0-100 points); the event influence feature value is mapped according to these three dimensions:

[0138] Time dimension: match the event occurrence time to the corresponding live broadcast period (such as “a certain privacy and security event occurs in the second live broadcast period, which is mapped to the second period dimension”);

[0139] Event type dimension: classified according to the association degree of the event and the vehicle property (such as “privacy-related events” are classified as A, and “price-related events” are classified as B);

[0140] Influence intensity dimension: standardize the feature value to 0-100 points (such as the original feature value is 80, and the mapped value is still 80 points).

[0141] Multi-dimensional fusion of the mapped event influence feature value and the grid adjustment feature quantity:

[0142] Time dimension alignment: ensure that the two are matched in the same live broadcast period (such as the event feature of the second period is fused with the grid feature of the second period);

[0143] Feature dimension merging: merge the event influence intensity and the weighted change rate feature of the grid adjustment feature quantity into a tensor (such as “the second period + A class event + grid weighted gradient” forms a three-dimensional feature);

[0144] Integrated into a comprehensive correction parameter: dimension compression is performed on the fused tensor to generate a comprehensive correction parameter containing event influence and grid flow change (such as the comprehensive correction coefficient of the second period is 1.15).

[0145] The step 53 above locates the fluctuation interval of the basic flow prediction curve (such as the peak interval, the trough interval, and the stable interval), and adjusts the curve shape according to the comprehensive correction parameter:

[0146] For the peak interval: if the comprehensive correction parameter is positive (indicating that the event and the grid feature jointly promote the flow to rise), the peak height is increased by the parameter value (such as the parameter 1.15 corresponds to a 15% increase in peak value);

[0147] For the trough interval: if the parameter is negative (indicating that there is a factor that suppresses the flow), the valley depth is reduced by the parameter value (such as the parameter 0.9 corresponds to a 10% reduction in valley value);

[0148] For the stationary interval: adjust the parameter according to the slope of the curve (e.g. parameter 1.05 corresponds to a 5% increase in slope, making the curve more consistent with the actual trend).

[0149] Integrate the adjustment results of all intervals to form a revised traffic prediction curve that reflects the combined effects of event impact and grid traffic changes.

[0150] Step 54 above compares the continuous two-period data of the revised traffic curve at a fixed period (e.g. every 30 minutes): calculate the difference between the traffic value of the next period and the previous period, if the difference is negative and the absolute value accounts for more than a preset amplitude (e.g. 10%) of the traffic of the previous period, it is determined as "continuous decline".

[0151] Call the vehicle type sensitive threshold (e.g. the privacy label weight threshold of high-end vehicles is 0.5), if the optimization weight parameter (e.g. the weight of "privacy glass" is 0.65) is greater than the threshold, it means that the user pays high attention to the privacy label, and the traffic decline needs to be intervened first.

[0152] When both "continuous two-period traffic decline > 10%" and "optimization weight parameter > 0.5" are met, automatically generate a pre-trigger instruction containing the target user group (high-end vehicle attention users) and the promotion strategy type (e.g. "privacy configuration exclusive gift package").

[0153] In a preferred embodiment of the present application, step 6 above, based on the pre-generated promotion trigger instruction and the cross-domain data set, executes the promotion trigger instruction; at the same time, through the mirror algorithm, the cross-domain data set and the public opinion correlation degree of the live room comments are processed, and when the correlation degree exceeds the preset threshold, the speech optimization instruction is executed, which can include:

[0154] Step 61, analyze the target user group and promotion strategy type in the pre-generated promotion trigger instruction; trigger the pop-up window push and time-limited gift issuance operation through the live platform interface to obtain the promotion execution state mark;

[0155] Step 62, when the promotion execution state mark detects that the promotion has been executed, activate the mirror processing flow of the cross-domain data set and the real-time comment stream; based on the cross-domain data set, construct a mapping space of public opinion keywords and comments through the mirror algorithm; calculate the real-time correlation value in the mapping space;

[0156] Step 63, when the real-time correlation value exceeds the preset dynamic threshold, locate the high-correlation public opinion keywords, generate a set of speech optimization instructions, including: keyword reinforcement explanation mark and parameter comparison speech template; execute the set of speech optimization instructions through the anchor teleprompter to complete the speech optimization closed loop.

[0157] In the embodiment of the application, by accurately analyzing instructions and triggering platform operations, it is ensured that the promotion information only reaches the target user group, reducing invalid push; the execution state is obtained in real time, providing an execution benchmark for subsequent public opinion analysis, improving the accuracy and controllability of the promotion; the mapping space constructed by the mirror algorithm realizes the association of cross-domain public opinion and live room comments, and real-time correlation calculation can timely capture the feedback focus of users on the promotion activity; based on the high-correlation keywords, targeted speech instructions are generated, the teleprompter is used to promote the anchor to dynamically adjust the explanation content, quickly respond to user questions, and improve the user interaction experience and promotion conversion rate.

[0158] In the embodiment of the application, when specifically applied, the following technical solutions can be used to achieve the above-mentioned technical effects, for example:

[0159] The above step 61 extracts the core information in the pre-generated promotion trigger instruction:

[0160] Target user group: filter the corresponding user ID list through the user tags (such as "high-end vehicle attention users" and "users with privacy configuration query frequency ≥ 5 times") included in the instruction;

[0161] Promotion strategy type: identify the strategy type (such as "privacy configuration exclusive limited-time gift package" and "try driving reservation discount offer") in the instruction, and clearly define the promotion content (the configuration included in the gift package, the discount amount, and the validity period).

[0162] Call the pop-up window push interface of the live platform, and push the promotion information (such as the pop-up window content "high-end user exclusive: privacy glass upgrade gift package for 30 minutes") to the target user group ID; call the platform's limited-time gift package distribution interface, set the gift package receiving rules (such as limited to 1 for each person, and bound to the user account), and monitor the distribution state (such as "pushed", "received", and "distribution failed") in real time; integrate the state information of push and distribution, and generate a promotion execution state marker (such as "successfully pushed to 80% of the target users, and 20% of the users have received the gift package").

[0163] The above step 62, when the promotion execution state marker shows "promotion has been executed" (such as pop-up window push is completed, and gift package distribution starts), automatically starts the association analysis process of cross-domain data set and live room real-time comment stream, and sets the processing period (such as every 5 minutes).

[0164] Extracting public opinion keywords (such as "privacy gift package cost-effective" and "test drive preferential intensity") from cross-domain data sets (such as vertical automobile forum data) as "source keywords" of the mapping space; collecting live streaming room real-time comment streams, extracting keywords (such as "what does the gift package contain" and "can the preferential treatment be stacked") in user comments as "target keywords" of the mapping space; constructing a corresponding relationship between source keywords and target keywords according to semantic similarity (such as "cost-effective" and "preferential intensity" are semantically related), forming a mapping space (such as an association network of "source keyword A-target keyword A1 / A2").

[0165] In the mapping space, the association strength is counted:

[0166] The co-occurrence frequency of source keywords and target keywords is calculated (such as "privacy gift package" appears 100 times in cross-domain public opinion, and related keywords appear 80 times in live streaming room comments); combined with semantic matching degree (such as 1 point for complete match and 0.6 points for partial match), the real-time correlation value is calculated (such as co-occurrence frequency x semantic matching degree, and the final score is 75 points).

[0167] In the above step 63, when the real-time correlation value exceeds a preset dynamic threshold (such as 60 points), the keywords with the highest correlation in the mapping space (such as "privacy gift package cost-effective" and "test drive preferential restriction") are screened out, and are marked as high-correlation public opinion keywords that need to be responded to.

[0168] For high-correlation keywords, a label that needs to be explained by the anchor is generated (such as "need to explain in detail the 3 items of privacy gift package upgrade configuration within 1 minute"); parameter comparison speech template: combined with user questions in cross-domain public opinion (such as "compared with the gift package of the competitor"), a structured speech template is generated (such as "our privacy gift package contains 1 item of sound insulation upgrade more than the competitor's, and the preferential intensity is 20% higher").

[0169] The speech optimization instruction set is synchronized to the anchor teleprompter in real time (such as the teleprompter displays "key explanation: privacy gift package cost-effective, reference template 3"), the anchor execution is monitored (such as whether it is explained within the specified time), and the optimization effect is confirmed according to the subsequent comment feedback (such as the reduction of user questions), and the closed loop from "public opinion correlation analysis" to "speech adjustment" is completed.

[0170] As shown in Figure 2 The embodiment of the present application also provides a live automobile marketing traffic fluctuation prediction system, which comprises:

[0171] The acquisition module is used for synchronously acquiring data in the live streaming platform and cross-domain extended data, and generating platform data sets and cross-domain data sets;

[0172] The association module is configured to construct an association space of social hot events and vehicle attributes based on cross-domain data sets by using a mirror algorithm, and generate an event influence characteristic value when an event characteristic intensity of the association space exceeds a preset threshold value.

[0173] The convex hull algorithm module is configured to generate a basic traffic prediction curve and select a key point set based on the event influence characteristic value and a platform data set at a fixed period, and generate a convex polygon region by processing the key point set by using a convex hull algorithm.

[0174] The optimization weight module is configured to calculate a traffic change gradient of each grid unit to generate a gradient distribution matrix and identify a vehicle type when the convex polygon region is segmented into grid units, and to call a mirror algorithm based on a historical feature library to establish a privacy label weight mapping relationship according to the traffic change gradient to generate an optimization weight parameter if the vehicle is a high-end vehicle.

[0175] The correction module is configured to generate a grid adjustment characteristic quantity based on the optimization weight parameter and the gradient distribution matrix, to generate a comprehensive correction parameter by fusing the event influence characteristic value and the grid adjustment characteristic quantity, and to correct the basic traffic prediction curve to generate a corrected traffic curve, and to pre-generate a promotion trigger instruction when the corrected traffic curve shows a traffic downward trend and the optimization weight parameter meets a preset threshold value.

[0176] The optimization instruction module is configured to execute the promotion trigger instruction based on the pre-generated promotion trigger instruction and the cross-domain data set, and to process a public opinion association degree of the cross-domain data set and live broadcast room comments by using a mirror algorithm, and to execute a speech optimization instruction when the association degree exceeds a preset threshold value.

[0177] The above describes preferred embodiments of the present application. It should be noted that those of ordinary skill in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements should also be considered within the scope of protection of the present application.

Claims

1. A method for predicting traffic fluctuations in live-stream car marketing, characterized in that, The method includes: Step 1: Synchronously collect data within the live streaming platform and cross-domain extended data to generate platform datasets and cross-domain datasets; Step 2: Based on the cross-domain dataset, construct an association space between social hot topics and vehicle attributes using a mirroring algorithm. When the event feature strength in the association space exceeds a preset threshold, generate event impact feature values, including: Step 21: Based on the cross-domain dataset, analyze the public opinion keyword weight distribution data and extract the high-frequency keyword set; standardize the offline consultation volume gradient sequence and generate a consultation volume change vector to obtain the preprocessed cross-domain dataset. Step 22: Based on the preprocessed cross-domain dataset, the high-frequency keyword set is mapped to the vehicle attribute space using the mirroring algorithm. The consultation volume change vector is then subjected to tensor product operation with the vehicle attribute space to obtain the social hotspot-vehicle attribute association space. Step 23: Calculate the feature vector magnitude of each social hotspot event in the social hotspot-vehicle attribute association space, normalize the magnitude, and generate a sequence of event feature intensity values. Step 24: When any intensity value in the event feature intensity value sequence exceeds a preset dynamic threshold, feature value generation is triggered and the intensity value exceeding the threshold is converted into an event impact feature value; Step 3: Based on the event impact feature values ​​and platform dataset, generate basic traffic prediction curves at fixed intervals and select key point sets; process the key point sets using the convex hull algorithm to generate convex polygon regions; Step 4: When segmenting the convex polygon region into grid cells, simultaneously calculate the traffic change gradient of each grid cell to generate a gradient distribution matrix and identify the vehicle type; if it is a high-end vehicle, call the mirror algorithm based on the historical feature library, establish a privacy label weight mapping relationship based on the traffic change gradient, and generate optimized weight parameters. Step 5: Generate grid adjustment feature quantity based on optimized weight parameters and gradient distribution matrix; merge event impact feature value and grid adjustment feature quantity to generate comprehensive correction parameter, and correct the basic traffic prediction curve to generate corrected traffic curve; when the corrected traffic curve shows a downward trend in traffic and the optimized weight parameter meets the preset threshold, pre-generate promotion trigger instruction; Step 6: Execute the promotion trigger instruction based on the pre-generated promotion trigger instruction and cross-domain dataset; at the same time, process the public opinion correlation between the cross-domain dataset and the live broadcast room comments through the mirroring algorithm, and execute the wording optimization instruction when the correlation exceeds the preset threshold.

2. The method for predicting traffic fluctuations in live-stream car marketing according to claim 1, characterized in that, Step 1: Synchronously collect data from the live streaming platform and cross-domain extended data to generate platform datasets and cross-domain datasets, including: Step 11: Collect platform dynamic data in real time from the pre-live broadcast period to the end of the live broadcast to obtain raw platform data; Step 12: Based on the original platform data, simultaneously trigger cross-domain data collection of the target model's public opinion keyword set in the vertical automotive forum and the consultation volume trend data of offline sales channels to obtain the original cross-domain data. Step 13: Based on the original cross-domain data and the original platform data, align the original platform data according to the live broadcast timeline to generate time-aligned platform data; based on the time-aligned platform data, extract the interaction rate-conversion rate correlation matrix; synchronize the original cross-domain data with the aligned platform timeline to generate time-aligned cross-domain data. Step 14: Perform feature encoding on time-aligned platform data to generate a platform dataset containing a peak viewing distribution matrix, an interaction conversion correlation tensor, and a competitor traffic time-series feature vector; perform semantic aggregation on time-aligned cross-domain data to generate a cross-domain dataset containing public opinion keyword weight distribution data and an offline consultation volume gradient sequence.

3. The method for predicting traffic fluctuations in live-stream car marketing according to claim 2, characterized in that, Step 3: Based on the event impact feature values ​​and platform dataset, generate a basic traffic prediction curve at a fixed period and select a key point set; process the key point set using the convex hull algorithm to generate a convex polygon region, including: Step 31: Inject the event impact feature values ​​into the interaction transformation association tensor of the platform dataset, and generate a spatiotemporal augmented dataset based on the fused tensor; Step 32: Based on the spatiotemporal augmentation dataset, perform time series prediction at fixed intervals to generate a basic traffic prediction curve; correct the shape characteristics of the basic traffic prediction curve based on the historical high-end vehicle feature library, and automatically select a set of key points, where: peak points correspond to promotional sensitive periods, trough points correspond to periods of traffic loss risk, and inflection points correspond to periods of public opinion change response. Step 33: Select boundary extreme points in the key point set using the convex hull algorithm, and connect them in chronological order to generate an initial convex polygon region; during the connection process, add auxiliary connection points to the key point set corresponding to the privacy label period of high-end models to improve the region accuracy, and simplify the connection path to optimize the calculation efficiency for the key point set corresponding to the price sensitive period of economy models, thereby generating a convex polygon region covering the core traffic fluctuation area.

4. The method for predicting traffic fluctuations in live-stream car marketing according to claim 3, characterized in that, Step 4: When dividing the convex polygon region into grid cells, simultaneously calculate the gradient distribution matrix of the flow change gradient of each grid cell and identify the vehicle type; For high-end models, the mirroring algorithm is invoked based on the historical feature database to establish a privacy label weight mapping relationship according to the traffic change gradient, generating optimized weight parameters, including: Step 41: Based on the convex polygon region, divide it into grid cells along the time axis, and extract the flow data and time period information contained in each grid cell to generate grid cell data; simultaneously calculate the flow change rate of each grid cell to generate a gradient distribution matrix; Step 42: Based on the grid cell data and gradient distribution matrix, extract user profile features within the grid cells, including user query behavior and interaction frequency regarding vehicle configuration; call the historical high-end vehicle feature library for feature comparison, count the frequency of privacy configuration queries, and when the frequency is greater than the benchmark value in the historical high-end vehicle feature library, mark it as a high-end vehicle; otherwise, mark it as a Volkswagen vehicle to obtain the vehicle identification. Step 43: When the vehicle model is identified as a high-end vehicle, the gradient distribution matrix is ​​invoked and a privacy label weight mapping relationship is established based on the historical high-end vehicle feature library through the mirroring algorithm. That is, the gradient change rate is used as the independent variable, the historical weight of the privacy label in the feature library is used as the base value, and the weight value is dynamically adjusted according to the gradient change direction. When the gradient increases, the weight increase is increased, and when the gradient decreases, the weight decrease is applied to generate optimized weight parameters.

5. The method for predicting traffic fluctuations in live-stream car marketing according to claim 4, characterized in that, Step 5: Generate grid adjustment feature quantities based on optimized weight parameters and gradient distribution matrix; fuse event impact feature values ​​and grid adjustment feature quantities to generate comprehensive correction parameters, and correct the basic flow prediction curve to generate the corrected flow curve; When the traffic curve shows a downward trend and the optimized weight parameters meet the preset threshold, a promotion trigger instruction is pre-generated, including: Step 51: Based on the optimized weight parameters and gradient distribution matrix, extract the rate of change feature through the spatiotemporal characteristics of the gradient distribution matrix, and perform weighted aggregation on the rate of change feature based on the optimized weight parameters to obtain the grid adjustment feature quantity. Step 52: Based on the mesh adjustment feature quantity and the event impact feature value, map the event impact feature value to the preset feature space, and perform tensor fusion with the mesh adjustment feature quantity to obtain the comprehensive correction parameter; Step 53: Inject the comprehensive correction parameters into the fluctuation range of the basic flow prediction curve to generate the corrected flow curve; Step 54: When it is detected that the flow rate decreases for two consecutive cycles in the corrected flow rate curve is greater than the preset range, and the verification optimization weight parameter is greater than the vehicle model sensitivity threshold, a promotion trigger instruction is pre-generated.

6. The method for predicting traffic fluctuations in live-stream car marketing according to claim 5, characterized in that, Step 6: Based on the pre-generated promotion trigger command and cross-domain dataset, execute the promotion trigger command; simultaneously, process the correlation between the cross-domain dataset and the live stream comments using a mirroring algorithm. When the correlation exceeds a preset threshold, execute the script optimization command, including: Step 61: Analyze the target user group and promotion strategy type in the pre-generated promotion trigger instruction; trigger pop-up push and limited-time gift pack distribution operations through the live streaming platform interface to obtain the promotion execution status marker; Step 62: When the promotion execution status flag detects that the promotion has been executed, activate the mirroring process of the cross-domain dataset and the real-time comment stream; based on the cross-domain dataset, construct a mapping space between public opinion keywords and comments through the mirroring algorithm; calculate the real-time correlation value within the mapping space; Step 63: When the real-time relevance value exceeds the preset dynamic threshold, locate highly relevant public opinion keywords, generate a set of speech optimization instructions, including: keyword reinforcement explanation markers and parameter comparison speech templates; execute the speech optimization instruction set through the anchor teleprompter to complete the speech optimization closed loop.

7. A live-streaming car marketing traffic fluctuation prediction system, wherein the system implements the method as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to simultaneously acquire data within the live streaming platform and cross-domain extended data, generating platform datasets and cross-domain datasets; The association module is used to construct an association space between social hot topics and vehicle attributes based on a cross-domain dataset using a mirroring algorithm. When the event feature intensity in the association space exceeds a preset threshold, event impact feature values ​​are generated. This includes: parsing the weight distribution data of public opinion keywords based on the cross-domain dataset and extracting a set of high-frequency keywords; standardizing the offline consultation volume gradient sequence and generating a consultation volume change vector to obtain a preprocessed cross-domain dataset; mapping the set of high-frequency keywords to the vehicle attribute space using a mirroring algorithm based on the preprocessed cross-domain dataset, and performing a tensor product operation between the consultation volume change vector and the vehicle attribute space to obtain the social hot topic-vehicle attribute association space; calculating the feature vector magnitude of each social hot topic event in the social hot topic-vehicle attribute association space, normalizing the magnitude, and generating a sequence of event feature intensity values; when any intensity value in the event feature intensity value sequence exceeds a preset dynamic threshold, feature value generation is triggered, and the intensity value exceeding the threshold is converted into an event impact feature value. The convex hull algorithm module is used to generate basic traffic prediction curves and select key point sets at fixed intervals based on event impact feature values ​​and platform datasets; and to generate convex polygon regions by processing the key point sets through the convex hull algorithm. The weight optimization module is used to simultaneously calculate the traffic change gradient of each grid cell when segmenting convex polygon regions into grid cells, generate a gradient distribution matrix, and identify vehicle type; if it is a high-end vehicle, it calls the mirror algorithm based on the historical feature library, establishes a privacy label weight mapping relationship based on the traffic change gradient, and generates optimized weight parameters. The correction module is used to generate grid adjustment feature quantities based on optimized weight parameters and gradient distribution matrix; it integrates event impact feature values ​​and grid adjustment feature quantities to generate comprehensive correction parameters, and corrects the basic traffic prediction curve to generate a corrected traffic curve; when the corrected traffic curve shows a downward trend in traffic and the optimized weight parameters meet the preset threshold, a promotion trigger instruction is pre-generated. The optimization instruction module is used to execute promotion trigger instructions based on pre-generated promotion trigger instructions and cross-domain datasets; at the same time, it uses a mirroring algorithm to process the correlation between cross-domain datasets and live stream comments, and executes the wording optimization instruction when the correlation exceeds a preset threshold.

8. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images