An unattended AI digital person automatic live broadcast method and system

Abstract

The application discloses an unattended AI digital person automatic live broadcast method and system, and relates to the technical field of e-commerce live broadcast.The method comprises the following steps: automatically starting broadcasting according to a pre-designed plan and driving the digital person to explain the goods; calculating an expected value and a comprehensive score according to the interaction data of the unexplained goods, and dynamically adjusting the order of the next goods; automatically putting the goods on the shelf when the remaining explanation time is less than the pre-shelf time; monitoring the change rate of the number of viewers, the moving average value and the interactive activity level, calculating the interactive activity level based on data such as comments, and automatically issuing coupons when the conditions are met; performing voice synthesis and rehearsal on the script before explanation, and adjusting the speech speed or modifying the script when the time deviation exceeds the threshold; and combining the smooth number of viewers and the minimum time when the broadcast is over to determine whether the explanation is overtime, and forcibly ending and stopping the broadcast when the explanation is overtime.The application realizes dynamic optimization of unattended explanation scheduling and intelligent field control, and improves the live broadcast efficiency and conversion effect.

Description

Technical Field

[0001] This invention relates to the field of e-commerce live streaming technology, specifically to an unattended AI digital human automatic live streaming method and system. Background Technology

[0002] With the rapid development of the e-commerce live streaming industry, utilizing artificial intelligence and AI digital humans for automated live streaming sales has become an important means to reduce operating costs and extend live streaming duration. Current AI digital human live streaming systems have achieved automated product explanations to a certain extent, but they still generally rely on human supervision and intervention, making it difficult to achieve truly unmanned operation. This is specifically reflected in the following aspects.

[0003] To the best of the inventors' knowledge, existing solutions typically explain products in a pre-arranged fixed order and for a fixed duration. For example, Chinese patent application CN119893152A describes an automated live-streaming method and system applied to an internet platform, which sequentially plays pre-cut scripts, completing the explanation of one product before controlling the robot to move to the next fixed location to explain the next product. This method cannot respond to the real-time interactive behavior of viewers in the live stream. When users express their preference for unexplained products through clicks, adding to cart, or comments, the system cannot adjust the explanation order in time, causing high-interest products to be delayed and reducing conversion efficiency. Furthermore, regarding the control of explanation duration, there is often a discrepancy between the preset duration and the actual broadcast duration after script synthesis. Existing methods cannot automatically adapt the speaking speed or content, easily leading to chaotic explanation rhythm.

[0004] More importantly, the operation of existing AI-powered digital human live streaming systems heavily relies on manual intervention. Key marketing actions such as product listing and coupon distribution require operators to continuously monitor live stream status parameters such as viewer numbers and interaction activity, and execute these actions manually based on experience. The lack of an intelligent mechanism that automatically detects viewer attrition trends based on real-time data and triggers intervention makes it difficult to maintain viewer engagement and interaction levels when live streams are unattended. Furthermore, live stream termination and termination still require manual operation or setting fixed times, lacking automated termination control that allows for flexible judgment based on live stream status. This prevents the system from intelligently ending live streams when traffic remains low or terminating them only after important product explanations have been completed.

[0005] These shortcomings result in relatively high labor costs and insufficient accuracy and real-time performance in the actual operation of existing AI digital human live streaming systems, which fail to fully realize the commercial value of unmanned live streaming.

[0006] Therefore, there is an urgent need for an unattended AI digital human automatic live streaming method and system that can dynamically adjust the lecture schedule based on real-time interactive data and automatically execute scene control and broadcast on / off based on live streaming status parameters. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an unattended AI digital human automatic live streaming method and system, which solves the problems mentioned in the background.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an unattended AI digital human automatic live streaming method, applied to an unattended AI digital human automatic live streaming system comprising a digital human driving module, a product scheduling engine, and a broadcast start / stop controller, comprising the following steps: Obtain the preset live streaming plan, which includes at least the sequence of products to be explained and the preset explanation duration for each product; When the preset broadcast time is reached, a live broadcast connection is automatically established and the virtual digital human is driven to start the live broadcast. For the current product, a virtual digital human is used to explain it. During the explanation of the current product, the interaction data of the audience with the unexplained products is collected in real time. Based on the interaction data, the next product to be explained is determined from the unexplained products. When the explanation time of the current product reaches its preset explanation time, the explanation of the current product ends and switches to the determined next product to be explained. During the explanation, the product listing operation is executed according to the preset scene control rules, and the live broadcast status parameters are monitored in real time. When the preset trigger conditions are met, the corresponding scene control operation is executed. Among them, the live broadcast status parameters include the rate of change of the number of viewers and / or the level of interaction. The scene control operation includes at least coupon distribution and / or product listing operation. During the live stream, the system checks whether the preset end-of-stream conditions are met and ends the live stream when the preset end-of-stream conditions are met.

[0009] As a further aspect of the present invention, obtaining the preset live streaming plan also includes obtaining the explanation script text corresponding to each product; before explaining the current product, it also includes: performing voice synthesis pre-rehearsal on the explanation script text and obtaining the pre-rehearsal duration; if the pre-rehearsal duration deviates from the preset explanation duration by more than a preset duration deviation threshold, the estimated explanation duration is matched with the preset explanation duration by adjusting the speech rate of the virtual digital human and / or adjusting the script content.

[0010] As a further aspect of the present invention, the preset end-of-broadcast conditions include: the duration of the live broadcast exceeds the preset minimum start-of-broadcast duration, and the current time reaches the preset end time; or, the duration of the live broadcast exceeds the preset minimum start-of-broadcast duration, and the smooth number of viewers is lower than the preset field view threshold in multiple consecutive sampling periods, wherein the smooth number of viewers is calculated by an exponentially weighted moving average.

[0011] As a further aspect of the present invention, the number of viewers is smoothed through... The calculation yields the result, where k is the sampling period number. Let the smoothed number of viewers be the number of viewers in the k-th period. Let be the number of viewers in the k-th period. It is the smoothing coefficient, and 0 < ≤1, the initial value of the smoothed-up viewer count is preset to 0; As a further aspect of the present invention, if the current explanation time of a product in the explanation state exceeds g times its preset explanation time and has not ended, the explanation of the product is forcibly ended and the broadcast is stopped, where g is a preset multiple value.

[0012] As a further aspect of the present invention, the field control rules include the timing of product listing: when the remaining explanation time of the current product is less than or equal to the preset advance listing time, the product is set to be available for purchase. The remaining explanation time is calculated by subtracting the current product's explanation time from the preset explanation time. The advance listing time is a value that is greater than zero and less than the product's preset explanation time.

[0013] As a further aspect of the present invention, during the explanation process, the order of explanations for unexplained products is dynamically adjusted based on audience interaction data during the live stream, as follows: Real-time acquisition of audience interaction data for unexplained products, and calculation of the expectation value for each unexplained product based on the interaction data; When the remaining time for explaining the current product reaches the preset advance notice, the next product to be explained is determined from the unexplained products based on the expected value. After the current product explanation is completed, switch to the next product to be explained.

[0014] As a further aspect of the present invention, the expected value of the unexplained product is achieved through... The calculation yields the result; where j is the index of the unexplained product number. Expectations for products that have not yet been explained. , , For the pre-set weighting coefficients, , and This includes the number of times a product is clicked, added to cart, or mentioned in comments within the interaction data.

[0015] As a further aspect of the present invention, the next product to be explained is determined from the unexplained products based on the expected value. The method is as follows: obtain the preset basic weight of each product, weight and combine the basic weight with the expected value to calculate the comprehensive score of the unexplained products, and select the product with the highest score as the next product to be explained based on the comprehensive score.

[0016] As a further aspect of the present invention, the weighted comprehensive calculation formula is as follows: The calculation shows that, in the formula, The overall score for the unexplained products. The product's default weights were not explained. This represents the expected value of the unexplained product, among which... This is the interaction weight coefficient, and its value range is... ,when When =1, the next product is determined entirely by real-time interactive data. When =0, the next item is determined according to the preset order.

[0017] As a further aspect of the present invention, the venue control operation includes coupon issuance, and the coupon issuance process includes: During the explanation, the rate of change of the number of viewers is calculated in real time. When the rate of change is lower than the preset decrease rate threshold, the remaining explanation time of the current product is greater than the preset minimum remaining time, and the number of times the product has been issued coupons has not reached the limit, coupons are automatically issued.

[0018] As a further aspect of the present invention, the rate of change in the number of viewers is calculated as follows: Let the current time be denoted as The number of viewers in real time is recorded as The previous preset time point The number of viewers is The rate of change is ; The rate of change in the number of viewers.

[0019] As a further aspect of the present invention, the coupon distribution process also includes: when the average number of real-time viewers in the sliding window is lower than a preset lower limit threshold for viewers and the remaining explanation time for the current product is greater than a preset minimum remaining time, the coupon distribution is directly triggered.

[0020] As a further aspect of the present invention, the site control operation also includes the distribution of coupons based on interaction activity: Define a sliding time window, count the total number of comments, likes and / or shares within the sliding time window, and calculate the interaction activity based on the average number of viewers within the sliding time window; When the interaction activity level is lower than the set threshold, the remaining explanation time for the current product is greater than the preset minimum remaining time, and the number of times coupons have been issued for the product has not reached the limit, coupons will be issued automatically.

[0021] As a further aspect of the present invention, the level of interaction activity is determined by... The calculation shows that, in the formula, , , and These represent the total number of comments, total number of likes, total number of shares, and average number of viewers counted within the sliding time window, respectively. , , These are the weighting coefficients for commenting, liking, and sharing behaviors, respectively.

[0022] As a further aspect of the present invention, in the product listing operation, each product has an independent advance listing time, and the method further includes: After a live stream ends, the pre-launch time of each product is adjusted individually based on the conversion data of each product during the live stream.

[0023] As a further aspect of the present invention, adjusting the advance listing time of each product individually includes: calculating the number of orders per unit exposure for the product in this live broadcast, comparing it with the historical average number of orders per unit exposure for the product, and increasing or decreasing the advance listing time of the product by a preset fixed step size based on the comparison result. The historical average number of orders per unit exposure is calculated by averaging the number of orders per unit exposure for the product over a preset number of K live streams.

[0024] An unattended AI digital human automated live streaming system is provided. This system is used to execute an unattended AI digital human automated live streaming method. The system includes: The digital human driving module is used to generate and control virtual digital humans to explain products; The product scheduling engine is used to obtain a preset live broadcast plan, which includes at least the sequence of products to be explained and the preset explanation duration for each product; when the explanation duration of the current product reaches its preset explanation duration, the explanation of the current product ends; and it is used to obtain the audience's interaction data on the unexplained products in real time, determine the next product to be explained from the unexplained products based on the interaction data, and switch to the determined next product to be explained after the explanation of the current product ends. The product scheduling engine is also used to execute product listing operations according to preset live streaming control rules, and automatically execute corresponding live streaming control operations when the real-time monitored live streaming status parameters meet preset trigger conditions. The live streaming status parameters include the rate of change of the number of viewers and / or the level of interaction. The live streaming control operations include at least coupon distribution and / or product listing operations. The live streaming controller is used to establish a live streaming connection when the preset start time is reached, and drive the virtual digital human to start the live stream; during the live stream, it detects whether the preset end-of-stream conditions are met, and ends the live stream when they are met.

[0025] This invention provides an unattended AI digital human automatic live streaming method and system. Compared with existing technologies, it has the following advantages: This invention utilizes a live streaming controller that works in conjunction with a product scheduling engine. Based on a preset live streaming plan, it automatically establishes a connection at the preset start time and drives a virtual digital avatar to explain the products. The live stream automatically ends when preset end conditions are met. This process requires no manual intervention, reducing reliance on human operators and enabling long-term automated operation of live streaming activities.

[0026] This invention acquires real-time interaction data from viewers regarding unexplained products, such as clicks, add-to-cart actions, and mentions, during the presentation process. It calculates the expectation value and overall score for each product and dynamically selects the next product to be presented. This allows the presentation order to respond to viewer interests, prioritizing products with higher engagement and improving the relevance and adaptability of the content arrangement.

[0027] This invention, based on live stream control rules, automatically sets a product to a purchaseable status when the remaining explanation time reaches the early listing time. Simultaneously, it monitors the rate of change in viewership and interaction activity in real time, automatically issuing coupons when the rate of change falls below a threshold or the activity level is insufficient. This automated intervention can respond instantly to changes in the live stream status, maintaining viewer retention and engagement.

[0028] This invention addresses the control of the pacing of the presentation by performing a pre-rehearsal of the script text using speech synthesis before the presentation. The pre-rehearsal duration is then compared to the preset presentation duration. If the deviation exceeds a threshold, the virtual digital human's speaking speed is automatically adjusted or adapted to the script content. This mechanism reduces the likelihood of the actual presentation duration deviating significantly from the preset value, resulting in a smoother transition in the live broadcast process.

[0029] This invention employs a combination of factors, such as a smoothing effect on the number of viewers and a minimum broadcast duration, to determine when to end a live stream. It also includes a forced termination protection mechanism: if the product explanation duration exceeds a preset multiple and is still ongoing, the stream will be forcibly terminated and ended. These measures effectively prevent the continuation of ineffective live streams and the excessive extension of single-product explanations, ensuring the stable execution of the live stream plan. Attached Figure Description

[0030] Figure 1 This is a system block diagram of an unattended AI digital human automatic live streaming system according to the present invention.

[0031] Figure 2 This is a flowchart illustrating an unattended AI digital human automatic live streaming method according to the present invention. Figure 1 .

[0032] Figure 3 This is a flowchart illustrating an unattended AI digital human automatic live streaming method according to the present invention. Figure 2 .

[0033] Figure 4 This is a flowchart illustrating an unattended AI digital human automatic live streaming method according to the present invention. Figure 3 . Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figures 1 to 4 As shown, the embodiments of the present invention provide the following technical solutions: As an embodiment of the present invention: See Figure 1 As shown, the present invention provides an unattended AI digital human automatic live streaming system, which includes a digital human driving module, a product scheduling engine, and a live streaming controller. The digital human driving module is used to generate and control virtual digital humans to explain products; The product scheduling engine is used to obtain a preset live streaming plan, which includes at least a sequence of products to be explained and a preset explanation duration for each product; when the explanation duration of the current product reaches its preset explanation duration, the explanation of the current product ends; and it is used to obtain the audience's interaction data on the unexplained products in real time, determine the next product to be explained from the unexplained products based on the interaction data, and switch to the determined next product to be explained after the explanation of the current product ends. The product scheduling engine is also used to perform product listing operations according to preset venue control rules, and automatically perform corresponding venue control operations when the live broadcast status parameters monitored in real time meet preset trigger conditions. The live broadcast status parameters include the rate of change of the number of viewers and / or the level of interaction activity; the venue control operations include at least coupon distribution and / or product listing.

[0036] The live streaming controller is used to automatically establish a live streaming connection when the preset live streaming start time is reached, and drive the virtual digital human to start live streaming; during the live streaming process, it detects whether the preset end-of-stream conditions are met, and ends the live streaming when they are met.

[0037] See Figure 2 As shown, the present invention also provides an unattended AI digital human automatic live streaming method. This system is implemented through an unattended AI digital human automatic live streaming system, and the method includes the following steps: Step S110: Obtain the preset live streaming plan; The live stream plan is a pre-configured data set by the operations team. This plan includes at least a sequence of products to be explained and the preset explanation duration for each product in the sequence.

[0038] Let i be the product to be explained, i = 1, 2, ..., N, where N is the total number of products to be explained. The preset explanation time for the i-th product is... .

[0039] The live streaming plan also includes the explanation script text for each product, product labels, preset start time, preset end time, etc.

[0040] Step S120: When the preset broadcast start time is reached, establish a live broadcast connection and drive the virtual digital human to start the live broadcast.

[0041] The system monitors the current system time in real time. When the preset start time in the live broadcast plan is reached, the on / off controller sends a push request to the live broadcast platform server to establish a live broadcast channel.

[0042] Meanwhile, the product scheduling engine identifies the first product in the product sequence to be explained as the current product, and calls the digital human driving module to load the virtual digital human and explanation resources corresponding to the product, and starts the live broadcast.

[0043] Step S130: Explain the current product using a virtual digital human, and end the explanation of the current product when the explained time reaches its preset explanation time.

[0044] After the explanation begins, a pre-configured timer starts recording the duration of the explanation for the current product.

[0045] In this embodiment, the explanation process is based on speech synthesis and lip-syncing of a preset script text, or generates explanation content by combining real-time interactive information.

[0046] When the already explained time reaches the preset explanation time When the product scheduling engine determines that the explanation of the current product is complete, it will then drive the virtual digital human to give a concluding remark or directly transition to a new section.

[0047] As a preferred implementation, this method also includes a pre-adaptation step for the presentation duration before starting the presentation of the current product, as follows: For product i to be explained, obtain its corresponding explanation script text, call the pre-configured speech synthesis engine to perform speech synthesis pre-playing on the explanation script text, and obtain the pre-playing duration. .

[0048] Rehearsal duration With preset explanation duration Comparison, If the rehearsal duration With preset explanation duration The absolute value of the corresponding duration deviation exceeds the preset duration deviation threshold. That is, satisfying Then, by adjusting the speaking speed of the virtual digital human and / or adjusting the content of the explanation script, the estimated explanation duration can be matched with the preset explanation duration.

[0049] When adjusting the speaking speed, a speaking speed adjustment factor is set. The adjusted estimated explanation time for ,make This resulted in the adjusted estimated explanation time. In an ideal situation, it equals .

[0050] If the calculated speech rate adjustment factor The speech speed exceeds the preset execution range of the virtual digital human. The execution range is then set to [[...]. , In this embodiment, for example... The value is 0.8. If the value is 1.5, then the speech rate will be set to the corresponding boundary value first, i.e., the minimum selected value. Maximum selection The script text is then truncated or expanded, and a rehearsal and evaluation are conducted until the rehearsal time reaches the specified duration. With preset explanation duration The corresponding duration deviation is less than or equal to the duration deviation threshold.

[0051] Step S140: During the current product presentation, perform the product listing operation according to the preset site control rules.

[0052] The rules for market control include the timing of product placement.

[0053] For product i, extract its pre-set early listing time. .

[0054] During the explanation, the remaining explanation time for the current product is calculated in real time. The remaining explanation time pass The calculation shows that, This indicates the duration of the current product's explanation.

[0055] when When the system is established, it calls the product listing interface provided by the e-commerce platform to change the product's status from "preview" or "unlisted" to "available for purchase," so that the product can be listed in a timely manner during the live broadcast.

[0056] Step S150: During the live broadcast, check whether the preset end-of-broadcast conditions are met.

[0057] The preset broadcast conditions include at least one of the following time conditions and field observation conditions: Live stream duration t live Exceeding the preset minimum broadcast duration T min And the current system time has reached the preset end time T in the live broadcast plan. end .

[0058] Live stream duration t live Exceeding the preset minimum broadcast duration T min Furthermore, the number of viewers remained below the preset field viewing threshold V for multiple consecutive sampling periods. th .

[0059] The smoothed viewership was calculated using an exponentially weighted moving average: Let k be the sampling period number, and let be the number of viewers in the kth period. , pass Calculate the smoothed number of viewers in the k-th period. ;in, It is the smoothing coefficient, and 0 < ≤1, the initial value of the smoothed-up viewer count is preset to 0; If all M consecutive sampling periods satisfy If the conditions for field viewing are met, a downcast signal will be generated.

[0060] Step S160: When the preset end-of-broadcast conditions are met, determine whether there is a product currently in the explanation state.

[0061] After receiving the end-of-broadcast signal, the on-broadcast controller checks the status of the product scheduling engine to determine whether a product is currently being explained.

[0062] If there is no product being explained at the moment, the live stream will end immediately.

[0063] If there is a product currently being explained, then determine the current duration of the explanation for that product. Has the preset explanation time been exceeded? The value is g times the value of g, where g is a preset multiple value, and in this embodiment g is 1.5.

[0064] If it exceeds, that is If the condition is met, the product demonstration will be forcibly ended and the live stream will be immediately stopped. If it does not exceed, that is If this is not the case, wait for the product explanation to finish naturally before proceeding with the end of the broadcast. This ensures the integrity of the explanation content and prevents the live stream from being abruptly cut off before the product introduction is completed, thus maintaining the viewer experience.

[0065] As a second embodiment of the present invention: See Figure 3 As shown, in specific implementation, compared to Embodiment 1, the only difference between the technical solution of this embodiment and Embodiment 1 is that in this embodiment, the next product to be explained is dynamically selected based on the audience's interaction data during the live broadcast, in order to better match the audience's interests and improve conversion rates. The specific steps are as follows: Step S210: Obtain the preset live streaming plan, which includes at least the sequence of products to be explained and the preset explanation duration T for each product. p,i .

[0066] In addition, the live streaming plan also includes a preset base weight W for each product. base,i This indicates the original priority specified in the operational strategy.

[0067] Step S220: When the preset broadcast time is reached, establish a live broadcast connection and drive the virtual digital human to start the live broadcast, and first explain the first product in the product sequence to be explained in a predetermined order.

[0068] Step S230: Explain the current product.

[0069] Step S240: During the explanation of the current product, obtain the audience's interaction data on the unexplained products in real time, and determine the next product to be explained based on the interaction data.

[0070] The set of goods U that are not currently explained.

[0071] For the j-th unexplained product in the product set U, collect one or more interaction data of the audience with the product within a preset time period.

[0072] In this embodiment, the preset time period is selected from the start of the live broadcast to the current time or the most recent preset time window; Interaction data includes, but is not limited to: number of product clicks Number of times added to cart Number of times the product is mentioned in the comments section .

[0073] pass Calculate the expected value of the product. ; in, , , In this embodiment, the pre-set weighting coefficients are determined based on the strength of each interaction's indication of purchase intention.

[0074] At the same time, obtain the product's preset base weight. .

[0075] In this embodiment, the base weights are pre-configured by operators based on factors such as gross profit margin and inventory.

[0076] pass Calculate the overall score for the unexplained product. ; in, This is the interaction weight coefficient, and its value range is... .when When =1, the next product is determined entirely by real-time interactive data; when When =0, the default order is followed completely.

[0077] Before calculating the overall score, it is necessary to... and Normalization is performed.

[0078] Expectations The normalization method is Among them, Emin and Emax are the highest values ​​among all currently unexplained products. The minimum and maximum values.

[0079] Basic weights The normalization method is Where Wmax is the maximum value of the basic weight among all the products to be explained.

[0080] Weighted composite calculation formula In and Take the normalized results respectively and To calculate the overall score.

[0081] As a preferred implementation method, the timing for determining the next product to be explained is set when the advance notice period is reached.

[0082] Let the lead time be . .

[0083] During the current product presentation, the remaining presentation time is calculated in real time. ,when At that time, based on the overall score Select the product with the highest score from the product set U as the next product to be discussed, denoted as j. next .

[0084] Step S250: After the current product explanation ends, switch to the selected next product j to be explained. next To explain.

[0085] After switching, repeat steps S230 to S250 until all products to be explained have been explained or the preset end-of-broadcast conditions are met.

[0086] Step S260: During the live broadcast, check whether the preset end-of-broadcast conditions are met. When the preset end-of-broadcast conditions are met, end the live broadcast.

[0087] This implementation method can automatically adjust the order of product presentations based on the real-time interest expressed by the audience, allowing high-interest products to be presented first, thereby achieving better conversion rates in unattended scenarios.

[0088] As an embodiment of the present invention: See Figure 4 As shown, in specific implementation, compared with Embodiment 1 and Embodiment 2, the technical solution of this embodiment is as follows: during the product explanation process, parameters representing the live broadcast status are monitored in real time, and when the parameters meet the triggering conditions, automatic scene control operations are performed, such as issuing coupons, to actively intervene in the atmosphere and retention of the live broadcast room. The steps are as follows: Step S310: Obtain a preset live streaming plan, which includes at least the sequence of products to be explained and the preset explanation duration for each product; When the preset broadcast time arrives, a live stream link is established and the virtual digital human is activated to start the live stream and explain the current product.

[0089] Step S320: During the explanation, monitor at least one live broadcast status parameter in real time.

[0090] Live streaming status parameters include the rate of change in the number of viewers and / or the level of interaction.

[0091] Rate of change in viewership The calculation method is as follows: Let the current time be denoted as The number of viewers in real time is recorded as The previous preset time point The number of viewers is The rate of change is ;in, This is a preset fixed time interval, which is less than the product explanation duration. The previous preset time point refers to the time point t counting backwards from the current time. The point in time.

[0092] Among them, the rate of change in the number of viewers A negative value indicates a decrease in the number of viewers; the larger the absolute value, the faster the viewers are lost.

[0093] Interaction activity The calculation method is as follows: Define a sliding time window of length W. In this embodiment, the sliding time window W is the time period 60 seconds before the current time. Count the total number of comments within this sliding time window. Total number of likes and / or share total quantity And calculate the average number of viewers within that window. .

[0094] pass The weighted interaction amount triggered by the audience within the sliding time window is calculated and denoted as the interaction activity level.

[0095] in, , , These are the weighting coefficients for commenting, liking, and sharing, respectively. This is used to eliminate the influence of the absolute number of viewers on the total number of interactions, making the interaction activity index comparable across different audience sizes.

[0096] Step S330: When the live broadcast status parameters meet the preset trigger conditions, the corresponding field control operation is automatically executed.

[0097] Market control operations include at least the issuance of coupons and / or the listing of products.

[0098] The preset trigger conditions for coupon issuance are as follows: Triggering condition based on the rate of decline in viewership: Set a preset decline rate threshold. ,in, In this embodiment, the descent rate threshold is a negative value. A value of -10 people / second indicates accelerated churn; a preset minimum remaining time is set. And the maximum number of times coupons can be issued per product session. In this embodiment, This is to avoid distributing coupons during meaningless time when the product presentation is about to end.

[0099] During the explanation of the current product i, if the following conditions are met simultaneously: , and If so, a coupon for that product will be automatically issued. This is the remaining explanation time for the current product. This indicates the number of times coupons have been issued for the current product during this live stream.

[0100] Triggering conditions based on audience lower limit protection: Set a preset audience lower limit threshold. .

[0101] During the explanation, if the average number of viewers of the sliding window is calculated in real time... satisfy and This will directly trigger the distribution of coupons.

[0102] Triggering conditions based on interaction activity: Set an interaction activity threshold. .

[0103] During the explanation, if the following conditions are met simultaneously , and This will directly trigger the distribution of coupons.

[0104] In this embodiment, coupons are issued by calling the marketing interface opened by the e-commerce platform. The type, face value, and usage threshold of the coupons can be pre-set by the operators in the live broadcast plan.

[0105] Step S340: During the live broadcast, check whether the preset end-of-broadcast conditions are met. When the preset end-of-broadcast conditions are met, end the live broadcast.

[0106] This implementation method allows for real-time perception of data changes in the live stream, similar to that of an experienced live stream operator. It also enables precise intervention decisions based on multiple dimensions such as churn rate, audience threshold, and interaction intensity, thereby effectively improving audience retention and conversion rates in unattended live streams.

[0107] As an embodiment of the present invention: As an embodiment of the present invention: In specific implementation, compared with Embodiment 1, Embodiment 2 and Embodiment 3, the only difference between this embodiment and Embodiment 1, Embodiment 2 and Embodiment 3 is that: after each live broadcast ends, the advance listing time of each product is adjusted differently based on the conversion data of that broadcast.

[0108] Step S410: After a live stream ends, obtain the conversion data for each product during the live stream. For product i, obtain its exposure volume. and attributable order numbers .

[0109] Exposure The number of unique viewers or views that entered the live stream during and after the product presentation is used to represent the number of views. Attributable order count This refers to the number of orders generated and completed due to exposure during the live stream.

[0110] In this embodiment, for each completed order, all valid interaction records of the user within a preset backtracking time before the transaction time are obtained. Valid interaction records include at least actions such as clicking the shopping cart and posting a comment with the product name. The latest valid interaction record is extracted from these valid interaction records. If the product corresponding to this interaction is product i, and the interaction occurs during the explanation period of product i, or within a preset attribution window after the explanation of product i ends, then this order is counted in the attribution order count for product i in this live broadcast session.

[0111] Step S420: Through Calculate the number of orders per unit exposure for this product during this live stream. : Step S430: Obtain the historical average number of orders per unit exposure for this product. .

[0112] Historical average number of orders per unit exposure This is calculated by averaging the number of orders per unit exposure for this product across K effective live streams in the past. The formula is: ,in, Let p be the number of orders per unit exposure for the product in the p-th live stream in the past, where p = 1, 2, ..., K.

[0113] Step S440: Increase the number of unit exposure orders for this live stream. Compared with historical average number of orders per unit exposure Compare the results and, based on the comparison, proceed with a preset fixed step size. Adjust the early listing time for this product. .

[0114] Extract the preset relative threshold coefficient ,and >0, the adjustment rules are as follows: like This indicates that the conversion rate of this product in this sale is significantly better than historical levels, suggesting that the advance listing time can be appropriately shortened to allow the product to become available for purchase sooner. ;in, This is the preset minimum allowable advance listing time to prevent the product from being listed too late after adjustments, or even before it's listed after the lecture.

[0115] like This indicates a significant decrease in conversion efficiency. It's advisable to appropriately extend the pre-listing time to increase suspense and build anticipation in product descriptions, thereby enhancing the product's appeal. ;in, This is the preset maximum allowable time for early listing, to avoid disrupting the live stream rhythm due to excessively long early listing times.

[0116] like Then let .

[0117] Adjusted early release time This will be written back into the live stream plan or product configuration.

[0118] This embodiment dynamically optimizes the best time to put each product on the market based on its historical performance and current feedback, enabling the control rules to have self-learning capabilities and further improving the precision of live streaming operations and overall revenue.

[0119] As an embodiment of the present invention: In specific implementation, compared with Embodiment 1, Embodiment 2, Embodiment 3 and Embodiment 4, the technical solution of this embodiment is to combine the solutions of Embodiment 1, Embodiment 2, Embodiment 3 and Embodiment 4.

[0120] This embodiment provides an unattended AI digital human automatic live streaming system, which also includes a processor and a memory storing a computer program. When the computer program is executed by the processor, an unattended AI digital human automatic live streaming method is implemented.

[0121] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0122] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0124] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for unattended AI digital human automatic live streaming, characterized in that, An unattended AI digital human automated live streaming system, which includes a digital human driving module, a product scheduling engine, and a broadcast controller, includes the following steps: Obtain the preset live streaming plan, which includes at least the sequence of products to be explained and the preset explanation duration for each product; When the preset broadcast time is reached, a live broadcast connection is automatically established and the virtual digital human is driven to start the live broadcast. For the current product, a virtual digital human is used to explain it. During the explanation of the current product, the interaction data of the audience with the unexplained products is collected in real time. Based on the interaction data, the next product to be explained is determined from the unexplained products. When the explanation time of the current product reaches its preset explanation time, the explanation of the current product ends and switches to the determined next product to be explained. During the explanation, the product listing operation is executed according to the preset scene control rules, and the live broadcast status parameters are monitored in real time. When the preset trigger conditions are met, the corresponding scene control operation is executed. Among them, the live broadcast status parameters include the rate of change of the number of viewers and / or the level of interaction. The scene control operation includes at least coupon distribution and / or product listing operation. During the live stream, the system checks whether the preset end-of-stream conditions are met and ends the live stream when the preset end-of-stream conditions are met.

2. The unattended AI digital human automatic live streaming method according to claim 1, characterized in that, Obtaining the preset live streaming plan also includes obtaining the explanation script text for each product; Before explaining the current product, the process also includes: performing a speech synthesis rehearsal on the explanation script text and obtaining the rehearsal duration. If the rehearsal duration deviates from the preset explanation duration by more than a preset duration deviation threshold, the estimated explanation duration is matched to the preset explanation duration by adjusting the speech rate of the virtual digital human and / or adjusting the script content.

3. The unattended AI digital human automatic live streaming method according to claim 1, characterized in that, The preset end conditions include: the live broadcast duration exceeds the preset minimum start time and the current time reaches the preset end time; or, the live broadcast duration exceeds the preset minimum start time and the smooth number of viewers is lower than the preset field view threshold in multiple consecutive sampling periods. The smooth number of viewers is calculated by exponential weighted moving average. If the current explanation time of a product in the explanation state exceeds g times its preset explanation time and has not ended, the explanation of the product will be forcibly ended and the broadcast will be stopped. g is the preset multiple value.

4. The unattended AI digital human automatic live streaming method according to claim 1, characterized in that, The rules for controlling the market include the timing of product listing. The method is as follows: when the remaining explanation time of the current product is less than or equal to the preset advance listing time, the product is set to be available for purchase. The remaining explanation time is obtained by subtracting the current product's explanation time from the preset explanation time. The advance listing time is a value that is greater than zero and less than the product's preset explanation time.

5. The unattended AI digital human automatic live streaming method according to claim 1, characterized in that, During the explanation, the order of explanations for products not yet explained is dynamically adjusted based on audience interaction data during the live stream, as follows: Real-time acquisition of audience interaction data for unexplained products, and calculation of the expectation value for each unexplained product based on the interaction data; When the remaining time for explaining the current product reaches the preset advance notice, the next product to be explained is determined from the unexplained products based on the expected value. After the current product explanation is completed, the system switches to the next product to be explained. The method for determining the next product to be explained is as follows: obtain the preset basic weight of each product, combine the basic weight with the expected value to calculate the comprehensive score of the unexplained products, and select the product with the highest comprehensive score as the next product to be explained.

6. The unattended AI digital human automatic live streaming method according to claim 1, characterized in that, The venue management operation includes coupon distribution, and the coupon distribution process includes: During the explanation, the rate of change of the number of viewers is calculated in real time. When the rate of change is lower than the preset decrease rate threshold, the remaining explanation time of the current product is greater than the preset minimum remaining time, and the number of times the product has been issued coupons has not reached the limit, coupons are automatically issued.

7. The unattended AI digital human automatic live streaming method according to claim 6, characterized in that, The coupon distribution process also includes: when the average number of real-time viewers in the sliding window is lower than the preset minimum audience threshold and the remaining explanation time for the current product is greater than the preset minimum remaining time, the coupon is directly triggered.

8. The unattended AI digital human automatic live streaming method according to claim 1, characterized in that, The event management also includes distributing coupons based on user engagement. Define a sliding time window, count the total number of comments, likes and / or shares within the sliding time window, and calculate the interaction activity based on the average number of viewers within the sliding time window; When the interaction activity level is lower than the set threshold, the remaining explanation time for the current product is greater than the preset minimum remaining time, and the number of times coupons have been issued for the product has not reached the limit, coupons will be issued automatically.

9. The unattended AI digital human automatic live streaming method according to claim 4, characterized in that, In the product listing process, each product has its own independent advance listing time, and the methods also include: After a live stream ends, the advance listing time of each product is adjusted individually based on the conversion data of each product during the live stream. The method is as follows: calculate the number of orders per unit exposure for the product during the live stream and compare it with the historical average number of orders per unit exposure for the product. Based on the comparison results, the advance listing time of the product is increased or decreased by a preset fixed step size. The historical average number of orders per unit exposure is calculated by averaging the number of orders per unit exposure for the product over a preset number of K live streams.

10. An unattended AI digital human automatic live streaming system, the system being used to execute the unattended AI digital human automatic live streaming method according to any one of claims 1-9, characterized in that, The system includes: The digital human driving module is used to generate and control virtual digital humans to explain products; The product scheduling engine is used to obtain a preset live broadcast plan, which includes at least the sequence of products to be explained and the preset explanation duration for each product; when the explanation duration of the current product reaches its preset explanation duration, the explanation of the current product ends; and it is used to obtain the audience's interaction data on the unexplained products in real time, determine the next product to be explained from the unexplained products based on the interaction data, and switch to the determined next product to be explained after the explanation of the current product ends. The product scheduling engine is also used to execute product listing operations according to preset live streaming control rules, and automatically execute corresponding live streaming control operations when the real-time monitored live streaming status parameters meet preset trigger conditions. The live streaming status parameters include the rate of change of the number of viewers and / or the level of interaction. The live streaming control operations include at least coupon distribution and / or product listing operations. The live streaming controller is used to establish a live streaming connection when the preset start time is reached, and drive the virtual digital human to start the live stream; during the live stream, it detects whether the preset end conditions are met, and ends the live stream when the preset end conditions are met.

Images