E-commerce live broadcast real-time interaction quality evaluation system based on edge computing

Abstract

This invention discloses a real-time interactive quality assessment system for e-commerce live streaming based on edge computing, relating to the field of e-commerce live streaming technology. It includes a multimodal interactive data acquisition module: through a distributed edge computing node cluster, it collects multimodal interactive data from the live stream in real time and constructs a multidimensional quality feature vector. The multimodal interactive data includes video encoding parameters, audio quality indicators, user interaction behavior data, and network transmission status data. This invention achieves real-time, accurate, and comprehensive evaluation of the interactive quality of e-commerce live streaming by using a distributed edge computing node cluster to collect multimodal interactive data from the live stream, employing a lightweight spatiotemporal attention neural network model for data fusion processing, a deep reinforcement learning network for generating quality optimization schemes, and an adaptive fuzzy inference system for real-time correction of the optimization schemes. Combined with network bandwidth fluctuations and terminal device resource status, dynamic parameter adjustments are made, resulting in a comprehensive and accurate assessment of the interactive quality of e-commerce live streaming.

Description

Technical Field

[0001] This invention relates to the field of e-commerce live streaming technology, and more specifically to an e-commerce live streaming real-time interactive quality assessment system based on edge computing. Background Technology

[0002] With the rapid development of internet technology and the changing shopping habits of consumers, e-commerce live streaming has become a new and popular marketing model. In the process of e-commerce live streaming, the quality of real-time interaction plays a crucial role in improving user experience and promoting product sales. More and more merchants and brands are using e-commerce live streaming platforms to promote their products and attract consumers to buy.

[0003] However, existing methods for evaluating the quality of e-commerce live streaming interactions have several shortcomings: Firstly, they often rely on single indicators, such as interaction frequency, number of comments, and number of likes, which fail to comprehensively and accurately reflect the true state of interaction quality and cannot fully measure it, as a single indicator cannot comprehensively consider factors such as the effectiveness of the interaction and the depth of user participation. Secondly, existing evaluation methods often do not adequately consider user experience, neglecting the impact of factors such as the smoothness of the live stream, picture quality, and the host's performance on interaction quality, and cannot meet the personalized needs of users, leading to discrepancies between the evaluation results and actual user experiences. Furthermore, e-commerce live streaming generates massive amounts of data with high real-time requirements, and traditional centralized computing models are prone to delays when processing this data, making them unsuitable for real-time evaluation.

[0004] Therefore, there is an urgent need for an edge computing-based real-time interactive quality assessment system for e-commerce live streaming, which can comprehensively and accurately assess the interactive quality of e-commerce live streaming in real time.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time interactive quality assessment system for e-commerce live streaming based on edge computing. This invention collects multimodal interactive data of the live stream in real time through a distributed edge computing node cluster, and then evaluates and adaptively adjusts the quality of live streaming interaction through a spatiotemporal attention neural network model to solve the problems in the background technology mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a real-time interactive quality assessment system for e-commerce live streaming based on edge computing, including a multimodal interactive data acquisition module: through a distributed edge computing node cluster, multimodal interactive data of the live stream is acquired in real time, and a multidimensional quality feature vector is constructed, wherein the multimodal interactive data includes video encoding parameters, audio quality indicators, user interaction behavior data and network transmission status data;

[0008] Interaction Quality Analysis Module: Utilizes a lightweight spatiotemporal attention neural network model built into the edge nodes to perform real-time fusion analysis of multimodal interaction data and generate real-time interaction quality assessment indicators;

[0009] Evaluation strategy generation module: The generator model of the deep reinforcement learning network is combined with real-time interactive quality evaluation indicators and historical optimization strategies to generate quality optimization schemes. The discriminator model evaluates the effectiveness of the quality optimization schemes based on user terminal feedback information.

[0010] Dynamic adjustment strategy module: The quality optimization scheme is corrected in real time through an adaptive fuzzy inference system. The parameters of the quality optimization scheme generated by the deep reinforcement learning network are dynamically adjusted in combination with the current network bandwidth fluctuations and terminal device resource status.

[0011] Feedback Execution Module: Based on the quality optimization scheme executed after dynamic parameter adjustment, combined with the distributed decision-making mechanism of edge nodes, it collaboratively optimizes video encoding parameters, transmission protocols and resource allocation strategies to dynamically control the quality of live streaming. The edge computing nodes then evaluate the quality of e-commerce live streaming in real time and feed the evaluation results back to the live streaming platform and merchants.

[0012] Optionally, the multimodal interactive data acquisition module includes an edge deployment unit: a lightweight multimodal data acquisition array is deployed at both the edge nodes of the content delivery network (CDN) and the user terminal to form a distributed data acquisition network;

[0013] Multimodal synchronization unit: The distributed data acquisition network synchronously collects multi-dimensional data such as video frame rate, encoding bit rate, audio latency, user likes, comment frequency and network jitter rate through a timestamp alignment mechanism, generating raw data that comprehensively reflects the interaction in the live broadcast room;

[0014] Preprocessing unit: performs filtering, denoising, and normalization on the raw data to construct a structured interactive quality feature matrix;

[0015] Feature extraction unit: Utilizes edge computing resources to perform real-time feature extraction on the interaction quality feature matrix, generating a multi-dimensional quality feature vector containing temporal and spatial features;

[0016] Edge transmission unit: After the multi-dimensional quality feature vector after feature extraction is compressed using hierarchical compression technology, it is transmitted to the analysis node through the CDN edge network.

[0017] Optionally, the interaction quality analysis module includes a data fusion unit: performing spatiotemporal alignment and fusion of video quality indicators, audio synchronization, interaction response latency, and network QoS parameters;

[0018] Model input unit: The fused multidimensional features are input into the pre-trained spatiotemporal attention neural network, which includes a video quality attention branch and an interactive experience attention branch;

[0019] Metric generation unit: The spatiotemporal attention neural network learns and outputs a set of evaluation metrics including screen smoothness score, audio-visual synchronization, interaction response index and comprehensive QoE score;

[0020] Threshold comparison unit: compares and analyzes the real-time output evaluation indicators with the preset dynamic quality thresholds to generate a quality anomaly feature map;

[0021] Result Output Unit: Performs data analysis on interaction quality based on the evaluation index set, generates analysis results, and encodes the analysis results into a quality state feature vector that can be recognized by a deep reinforcement learning network.

[0022] Optionally, the evaluation policy generation module includes a policy generation unit: a generator model in a deep reinforcement learning network receives a quality state feature vector and historical optimization records, and generates a candidate policy set including bitrate adjustment, frame rate optimization, and transmission protocol selection, wherein the historical optimization records include past... The effectiveness of the bitrate adjustment strategy within a given time period;

[0023] Strategy Evaluation Unit: The discriminator model calculates the expected quality improvement benefits of each candidate strategy in the candidate strategy set based on the actual experience data reported by the user terminal. The actual experience data includes lag rate, interaction success rate, and decoding time.

[0024] Optimize the decision-making unit: Employ the reinforcement learning decision ε-greedy algorithm to balance exploring new strategies with utilizing known effective strategies, and select the optimal quality optimization strategy;

[0025] Experience replay unit: Stores the decision-making process in the experience pool of edge nodes for online model updates;

[0026] Model update unit: Periodically synchronizes and updates model parameters between edge nodes to maintain consistency of evaluation strategy.

[0027] Optionally, the dynamic adjustment strategy module includes a strategy parsing unit: decomposing the optimal quality optimization strategy output by the reinforcement learning network into executable parameter adjustment instructions;

[0028] Encoding Adjustment Unit: Dynamically adjusts H.265 / AV1 encoding parameters based on network bandwidth prediction to achieve Pareto optimization of bitrate-image quality;

[0029] Transmission optimization unit: Employs adaptive multipath transmission technology to dynamically select the optimal combination of transmission protocols based on network conditions;

[0030] Resource allocation unit: Through edge computing resource scheduling algorithms, prioritize computing resources for key quality indicators.

[0031] Fault tolerance unit: When a sudden network fluctuation is detected, a degradation strategy is activated to ensure a basic interactive experience.

[0032] Optionally, the feedback execution module includes a quality monitoring unit: continuously monitoring the changes in actual quality indicators after the execution of the optimal quality optimization strategy;

[0033] Parameter adjustment unit: dynamically adjusts the quantization parameters and GOP structure of the video encoder through a fuzzy PID controller;

[0034] Transmission update unit: dynamically updates the forward error correction strategy and retransmission mechanism based on real-time network diagnostic results;

[0035] Resource reallocation unit: Dynamically adjusts the distribution of video processing tasks based on the load status of edge nodes;

[0036] Closed-loop control unit: Feeds the execution results back to the reinforcement learning network to form a continuously optimized control loop.

[0037] Optionally, the spatiotemporal attention neural network includes:

[0038] Spatiotemporal feature extraction and fusion component: In e-commerce live streaming scenarios, the spatiotemporal features of the live streaming process are extracted, and dynamic time warping and affine transformation are used to align and integrate the data in time and space. The spatiotemporal features include heterogeneous data such as video frame rate, audio latency, and user likes / comments / order hotspots.

[0039] Interaction Quality Prediction Component: By learning the relationship between spatiotemporal features and interaction quality in historical data, it predicts the real-time interaction quality of e-commerce live streaming based on the fused spatiotemporal features and outputs a quality assessment.

[0040] Anomaly detection component: Identifies abnormal situations in high-frequency periods or areas of the screen during the live broadcast through an attention mechanism, and takes timely measures to ensure the quality of the live broadcast.

[0041] Optionally, the deep reinforcement learning network includes:

[0042] Quality optimization strategy exploration and generation component: Based on the current network status and the live streaming status of user interaction data, explore new parameter adjustments when bandwidth fluctuates, and generate quality optimization strategies to dynamically adjust bitrate, frame rate, and transmission protocol;

[0043] Model agent optimization strategy component: environment setting reward function, based on the Actor-Critic framework, and adopts a proximal strategy to optimize PPO, and achieves stable updates through importance quality index sampling;

[0044] Resource allocation maximizes benefits component: Based on priority experience replay PER and edge node periodic dual-delay deep deterministic policy gradient TD3, the agent allocates computing resources and network bandwidth according to the real-time requirements of the live broadcast and the resource status of the edge nodes, so as to maximize the quality of live broadcast interaction and resource utilization efficiency, and continuously improve the evaluation accuracy of the spatiotemporal network by using execution feedback.

[0045] A computer device includes: a memory and a processor; the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described edge computing-based e-commerce live streaming real-time interactive quality assessment system.

[0046] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described edge computing-based e-commerce live streaming real-time interactive quality assessment system.

[0047] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0048] This invention utilizes a distributed edge computing node cluster to collect multimodal interactive data from live streams in real time. A lightweight spatiotemporal attention neural network model is used for data fusion processing, and a deep reinforcement learning network generates a quality optimization scheme. An adaptive fuzzy inference system then corrects the optimization scheme in real time. Combined with dynamic parameter adjustments based on network bandwidth fluctuations and terminal device resource status, this achieves real-time, accurate, and comprehensive evaluation of e-commerce live stream interaction quality. It can dynamically generate and adjust quality optimization schemes according to actual conditions, effectively addressing network bandwidth fluctuations and changes in terminal device resources. This not only improves the smoothness, clarity, and interactivity of the live stream but also provides users with a better viewing experience. Furthermore, it offers live streaming platforms and merchants a more reliable basis for quality assessment and optimization, promoting the healthy development of e-commerce live streaming. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 This is a block diagram of the real-time interactive quality assessment system for e-commerce live streaming based on edge computing, as described in this invention.

[0051] Figure 2 This is a flowchart of the evaluation method for the edge computing-based e-commerce live streaming real-time interactive quality evaluation system of the present invention. Detailed Implementation

[0052] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0053] This invention provides, for example Figure 1-2 The e-commerce live streaming real-time interactive quality assessment system based on edge computing shown includes a multimodal interactive data acquisition module: through a distributed edge computing node cluster, it collects multimodal interactive data of the live stream in real time and constructs a multidimensional quality feature vector, wherein the multimodal interactive data includes video encoding parameters, audio quality indicators, user interaction behavior data and network transmission status data;

[0054] Interaction Quality Analysis Module: Utilizes a lightweight spatiotemporal attention neural network model built into the edge nodes to perform real-time fusion analysis of multimodal interaction data and generate real-time interaction quality assessment indicators;

[0055] Evaluation strategy generation module: The generator model of the deep reinforcement learning network is combined with real-time interactive quality evaluation indicators and historical optimization strategies to generate quality optimization schemes. The discriminator model evaluates the effectiveness of the quality optimization schemes based on user terminal feedback information.

[0056] Dynamic adjustment strategy module: The quality optimization scheme is corrected in real time through an adaptive fuzzy inference system. The parameters of the quality optimization scheme generated by the deep reinforcement learning network are dynamically adjusted in combination with the current network bandwidth fluctuations and terminal device resource status.

[0057] Feedback Execution Module: Based on the quality optimization scheme executed after dynamic parameter adjustment, combined with the distributed decision-making mechanism of edge nodes, it collaboratively optimizes video encoding parameters, transmission protocols and resource allocation strategies to dynamically control the quality of live streaming. The edge computing nodes then evaluate the quality of e-commerce live streaming in real time and feed the evaluation results back to the live streaming platform and merchants.

[0058] The working principle of the above technical solution is as follows: First, through a distributed edge computing node cluster, a lightweight data acquisition matrix is ​​deployed at CDN edge nodes and user terminals. This enables low-latency, real-time acquisition of multimodal interactive data from the live stream, including video encoding parameters, audio quality indicators, user interaction behavior data, and network transmission status data. This ensures that the acquired data can effectively describe the interaction quality. Simultaneously, a timestamp alignment mechanism is used to synchronize and align the multi-dimensional data in time. Denoising and normalization preprocessing operations are then performed to construct a structured interaction quality feature matrix. Next, edge computing resources are used to extract temporal-spatial features, generating a multi-dimensional quality feature vector. After processing with layered compression technology, this vector is transmitted to the analysis node. Then, video smoothness, audio-visual synchronization, interaction response latency, and network QoS parameters are spatiotemporally fused and analyzed. This data is input into a pre-trained lightweight spatiotemporal attention neural network model to generate real-time interaction quality assessment indicators that can evaluate the quality of live interaction. These real-time interaction quality assessment indicators are compared with dynamic quality thresholds to output a quality anomaly feature map, which is then encoded into a quality state feature vector that can be recognized through reinforcement learning. Secondly, the deep reinforcement learning network generator receives the quality state feature vector and generates a set of candidate optimization strategies, including bitrate adjustment, frame rate optimization, and transmission protocol selection. The discriminator calculates the expected returns of each strategy based on the actual experience data reported by the user terminal, and selects the optimal strategy using the ε-greedy algorithm, storing it in an edge experience pool for online learning and updating. Finally, through an adaptive fuzzy inference system, the optimization strategies output by reinforcement learning are parsed into executable parameter adjustment instructions. Simultaneously, combined with real-time network bandwidth and terminal resource status, the controller dynamically adjusts the encoding parameters and transmission protocol, achieving Pareto optimization of bitrate-image quality. During network fluctuations, a multi-path transmission fault tolerance mechanism and a dynamic edge resource allocation strategy are activated to prioritize key indicators for evaluating interaction quality. A distributed decision-making mechanism adjusts the distribution of video processing tasks, including dynamically adjusting the GOP structure of the video encoder, enabling forward error correction at the transport layer, and allocating computing resources according to priority to achieve collaborative optimization by edge nodes. The system continuously monitors changes in actual quality indicators after execution and feeds the evaluation results back to the reinforcement learning model.

[0059] The above technical solution achieves the following effects: Low-latency acquisition and preprocessing of multimodal data from live streams is realized through distributed edge node clusters and lightweight data acquisition agents. Simultaneously, a spatiotemporal attention neural network model is used to perform multi-dimensional feature fusion analysis at the edge, reducing resource consumption while ensuring evaluation accuracy and improving data transmission response efficiency. Deep reinforcement learning networks combined with historical optimization strategies and real-time interactive quality evaluation metrics improve audio-visual synchronization and reduce stuttering, enhancing interactive quality. Adaptive fuzzy inference systems dynamically adjust bandwidth fluctuations to optimize resource utilization. Distributed resource scheduling algorithms based on edge nodes improve load balancing, and real-time feedback adjusts encoding parameters and transmission strategies, narrowing the range of live stream quality fluctuations and improving the stability of live stream interaction. The lightweight edge-side model deployment reduces pressure on the central server, lowering bandwidth costs and supporting real-time quality evaluation of tens of thousands of concurrent live streams. This ensures that live stream interaction can support rapid response, optimized user experience, and reliable, stable, and scalable evaluation accuracy in high-concurrency, low-latency, and highly interactive scenarios.

[0060] In one specific embodiment, the multimodal interactive data acquisition module includes:

[0061] Edge Deployment Unit: Lightweight multimodal data acquisition arrays are deployed at both the edge nodes of the Content Delivery Network (CDN) and user terminals to form a distributed data acquisition network. The lightweight multimodal data acquisition arrays include, but are not limited to, comment monitors, like counters, share recorders, and viewing time trackers.

[0062] Multimodal synchronization unit: The distributed data acquisition network synchronously collects multi-dimensional data such as video frame rate, encoding bit rate, audio latency, user likes, comment frequency and network jitter rate through a timestamp alignment mechanism, generating raw data that comprehensively reflects the interaction in the live broadcast room;

[0063] Preprocessing unit: performs filtering, denoising, and normalization on the raw data to construct a structured interactive quality feature matrix;

[0064] Feature extraction unit: Utilizes edge computing resources to perform real-time feature extraction on the interaction quality feature matrix, generating a multi-dimensional quality feature vector containing temporal and spatial features;

[0065] Edge transmission unit: After the multi-dimensional quality feature vector after feature extraction is compressed using hierarchical compression technology, it is transmitted to the analysis node through the CDN edge network.

[0066] The working principle of the above technical solution is as follows: First, a lightweight multimodal data acquisition array is deployed at CDN edge nodes and user mobile terminals / PC clients. For example, a comment monitor captures user comment content and sending timestamps in real time, a like counter counts the number of likes per second and the coordinates of click hotspots, a sharing recorder tracks sharing behavior types and conversion rates, and a viewing duration tracker records user dwell time and page scrolling behavior. This can build a distributed acquisition network, which can comprehensively and timely collect various interactive data in the live broadcast room. For example, CDN nodes cover the push data of video / audio encoding parameters on the broadcaster's end, and user terminals collect experience data such as client stuttering rate and rendering latency. Moreover, bidirectional data is interconnected with low latency through the QUIC proprietary tunnel protocol. Then, the distributed data acquisition network adopts a hybrid clock synchronization mechanism. At the hardware level, GPS / PTP clocks are used, and at the software level, the NTP protocol of the user terminal is used for time synchronization. By integrating and aligning multi-dimensional data, it can provide a comprehensive consideration of live broadcast quality, user behavior, and network environment for subsequent analysis. Secondly, a Kalman filter is used to eliminate transient noise in the original data, and median filtering is employed to process abnormal signals. Then, the heterogeneous data is mapped to a structured interaction quality feature matrix with unified dimensions. Edge computing is used to reduce data transmission latency, and temporal-spatial features of the interaction quality feature matrix are extracted to construct a user behavior map that more effectively represents the interaction quality of the live stream. Finally, since the extracted multidimensional quality feature vector data volume is still relatively large, layered compression technology is used to compress the data to different degrees while ensuring that key information in the evaluation data is not lost, in order to reduce data transmission pressure and cost.

[0067] The effects of the above technical solution are as follows: The distributed data acquisition network covers CDN edge nodes and user terminals. The multimodal data acquisition array can collect multi-dimensional data such as live streaming technical indicators, user interaction behavior, and network status, thereby breaking through the data acquisition dimensions. It not only comprehensively and accurately reflects the interaction in the live streaming room, but also provides a rich and reliable data foundation for subsequent analysis and evaluation. The application of edge computing reduces the latency of data transmission to the central server. At the same time, the timestamp alignment mechanism ensures the synchronous acquisition of multimodal data, enabling real-time acquisition and preprocessing of interactive data, improving data quality and availability, and helping subsequent feature extraction and analysis, thereby improving the accuracy and reliability of analysis results. Layered compression technology reduces data volume without losing key information, reducing data transmission pressure and cost. The use of CDN edge network further optimizes the data transmission path, improves data transmission efficiency, and ensures that feature vectors can be transmitted to analysis nodes quickly and stably, enabling timely detection and response to problems in the live streaming room.

[0068] In one specific embodiment, the interaction quality analysis module includes:

[0069] Data fusion unit: performs spatiotemporal alignment and fusion of video quality indicators, audio synchronization, interactive response latency, and network QoS parameters;

[0070] Model input unit: The fused multidimensional features are input into the pre-trained spatiotemporal attention neural network, which includes a video quality attention branch and an interactive experience attention branch;

[0071] Metric generation unit: The spatiotemporal attention neural network learns and outputs a set of evaluation metrics including screen smoothness score, audio-visual synchronization, interaction response index and comprehensive QoE score;

[0072] Threshold comparison unit: compares and analyzes the real-time output evaluation indicators with the preset dynamic quality thresholds to generate a quality anomaly feature map;

[0073] Result Output Unit: Performs data analysis on interaction quality based on the evaluation index set, generates analysis results, and encodes the analysis results into a quality state feature vector that can be recognized by a deep reinforcement learning network.

[0074] The working principle of the above technical solution is as follows: First, the Dynamic Time Warping (DTW) method is used to perform spatiotemporal alignment and fusion of video quality indicators, audio synchronization, interaction response latency, and network QoS parameters. Specifically, in terms of temporal mapping, PTS / DTS video frame timestamps and AAC frame audio sampling points are aligned at the microsecond level through interpolation, while user interaction events are synchronized with the audio and video streams via the NTP protocol. In terms of spatial mapping, the interaction coordinates of terminals with different resolutions are normalized to a unified coordinate system. Simultaneously, a four-dimensional tensor is constructed to fuse the spatiotemporally aligned multi-dimensional feature data. Then, utilizing the attention mechanism in the spatiotemporal attention neural network architecture, 3D convolutional attention is applied to the video quality branch, while LSTM and self-attention learning are used to process the interaction experience branch. This outputs evaluation indicators such as screen smoothness score, audio-visual synchronization, interaction response index, and comprehensive QoE score. The expression for the screen smoothness score is: In the formula, This is expressed as a score for screen smoothness. Represented as the first The frame rate of each time segment. Represented as the first The number of time segments and the total number of time segments; the expression for audio-visual synchronization is: In the formula, This is represented as audio-visual synchronization. Represented as the first The total number of samples and the number of samples for audio-visual synchronization detection. Represented as the first The video timestamps of each sample Represented as the first Audio timestamps of each sample Represented by the absolute value symbol; the expression for the interaction response index is: In the formula, Represented as the interaction response index, The average latency of the interaction request is expressed as: The expression for the overall QoE score is: ,and In the formula, Represented as, These are respectively represented as the corresponding screen smoothness scores. Audio-visual synchronization Interaction Response Index The weighting coefficients are determined. Secondly, a dynamic threshold is calculated using the exponentially weighted moving average (EWMA) method, and then compared with the evaluation index set output by the spatiotemporal attention neural network. A heatmap encoding is used to mark anomalous spatiotemporal regions, generating a quality anomaly feature map. Finally, the multidimensional feature data is reduced to a 32-dimensional vector encoding, and a state feature vector is output by adding temporal difference features.

[0075] The effects of the above technical solution are as follows: By aligning and fusing multi-dimensional feature data in time and space through DTW, combined with the training and learning of spatiotemporal attention neural networks, the quality of live interaction can be comprehensively and accurately evaluated, improving the accuracy of audio-visual synchronization during live interaction and providing users and operators with a clear quality profile; by using the EWMA method to calculate dynamic thresholds and conduct comparative analysis to generate a quality anomaly feature map, the specific aspects of interaction quality anomalies can be quickly located, enabling operators to conduct targeted problem investigation and repair, improving the efficiency of problem solving. At the same time, the dynamic threshold setting can be adjusted according to different scenarios and needs, showing strong adaptability and flexibility; the reinforcement learning model can continuously adjust and optimize the parameters and strategies of live interaction based on multi-dimensional feature vectors to improve interaction quality, achieve adaptive optimization of the system, and improve the real-time performance of interaction quality evaluation and the efficiency of information transmission.

[0076] In one specific embodiment, the evaluation strategy generation module includes:

[0077] Policy Generation Unit: The generator model in the deep reinforcement learning network receives the quality state feature vector and historical optimization records, and generates a candidate policy set including bitrate adjustment, frame rate optimization, and transport protocol selection. The historical optimization records include past... The effectiveness of the bitrate adjustment strategy within a given time period;

[0078] Strategy Evaluation Unit: The discriminator model calculates the expected quality improvement benefits of each candidate strategy in the candidate strategy set based on the actual experience data reported by the user terminal. The actual experience data includes lag rate, interaction success rate, and decoding time.

[0079] Optimize the decision-making unit: Employ the reinforcement learning decision ε-greedy algorithm to balance exploring new strategies with utilizing known effective strategies, and select the optimal quality optimization strategy;

[0080] Experience replay unit: Stores the decision-making process in the experience pool of edge nodes for online model updates;

[0081] Model update unit: Periodically synchronizes and updates model parameters between edge nodes to maintain consistency of evaluation strategy.

[0082] The working principle of the above technical solution is as follows: First, the identifiable quality state feature vector and the historical optimization records of the execution effect of the bitrate adjustment strategy within a unit time period are used as input and transmitted to the generator model. Based on the Actor-Critic framework, the Lagrange multiplier method is introduced to ensure that the strategy meets the bandwidth constraint condition. The generator model can dynamically adjust the H.265 encoding bitrate, adaptively switch the frame rate mode, and intelligently switch between TCP / UDP / QUIC protocols through the output of the policy network Actor. Then, the actual experience data reported by the user terminal, such as stuttering rate, interaction success rate, and decoding time, are input into the discriminator model. The dual-delay deep deterministic policy gradient TD3 algorithm is used to calculate the expected quality improvement benefit that each candidate strategy may bring after implementation. That is, after designing the reward function, the expected benefit of the candidate strategy is evaluated by the value function, and then the candidate strategies are arranged in descending order of expected quality improvement benefit. For example, a certain bitrate adjustment strategy may improve the clarity of the picture, but it may also increase the network bandwidth requirement. The discriminator model will evaluate the expected quality improvement benefit of the strategy based on the reward function for these factors. Secondly, a reinforcement learning decision-making ε-greedy algorithm is used to strike a balance between exploring new strategies and utilizing known effective strategies. Specifically, a candidate strategy is randomly selected with a certain probability ε for exploration to discover potentially better optimization strategies. Then, the strategy with the highest expected quality improvement benefit is selected with a probability of 1-ε, allowing for continuous experimentation with new strategies while fully utilizing existing experience. A dynamic decay mechanism is incorporated into the ε-greedy algorithm to ultimately select the optimal quality optimization strategy. When multiple strategies have conflicting parameters, Pareto optimality is used for filtering. Finally, a circular buffer is deployed at the edge nodes to store the entire decision-making process, including the quality state feature vector, candidate strategy set, selected optimal quality optimization strategy, and actual effect information. The model is updated online based on priority sampling and batch learning to minimize Bellman error and learn more effective optimization strategies. The model parameters between edge nodes are periodically updated synchronously, and a Byzantine fault tolerance algorithm is used to filter malicious nodes, ensuring the consistency of evaluation strategies across edge nodes and improving the model's stability and reliability.

[0083] The effects of the above technical solution are as follows: By combining a deep reinforcement learning generator and a discriminator model, multiple candidate strategies can be quickly generated and evaluated based on real-time quality status and historical experience, providing multiple options for optimizing live streaming interaction quality. The use of the ε-greedy algorithm helps to discover better quality optimization strategies, thereby improving the picture quality, smoothness, and user interaction experience of the live stream. The actual experience data reported by the user terminal accurately reflects the user's true feelings. By evaluating the strategies through the discriminator model, a user-centric experience is achieved, ensuring that the selected optimization strategy can truly improve the user's live streaming experience. The experience replay unit and model update unit enable the model to have the ability to learn online and continuously optimize. The historical decision-making process data in the experience pool can be used for online model updates, allowing the model to continuously learn new optimization strategies and adapt to different network environments. Regularly synchronizing and updating the model parameters between edge nodes not only clearly guarantees the consistency and stability of the evaluation strategy on different nodes, but also enables the entire evaluation strategy system to continuously evolve and improve. This allows the strategy to dynamically adjust the bitrate, frame rate, and transmission protocol in different live streaming scenarios and network conditions, providing adaptability and flexibility, and ensuring the overall performance and reliability of live streaming quality and user experience.

[0084] In one specific embodiment, the dynamic adjustment strategy module includes:

[0085] Policy parsing unit: decomposes the optimal quality optimization policy output by the reinforcement learning network into executable parameter tuning instructions;

[0086] Encoding Adjustment Unit: Dynamically adjusts H.265 / AV1 encoding parameters based on network bandwidth prediction to achieve Pareto optimization of bitrate-image quality;

[0087] Transmission optimization unit: Employs adaptive multipath transmission technology to dynamically select the optimal combination of transmission protocols based on network conditions;

[0088] Resource allocation unit: Through edge computing resource scheduling algorithms, prioritize computing resources for key quality indicators.

[0089] Fault tolerance unit: When a sudden network fluctuation is detected, a degradation strategy is activated to ensure a basic interactive experience.

[0090] The working principle of the above technical solution is as follows: First, the optimal quality optimization strategy output by the reinforcement learning network is decomposed into executable parameter adjustment instructions, including video coding parameter adjustment instructions, a transmission protocol priority list, and resource allocation weights. For example, if the optimal quality optimization strategy is to improve the stability of live interactive image quality while balancing the performance of user-end devices, then the strategy parsing unit dynamically degrades the encoding based on network bandwidth, prioritizing the use of UDP combined with FEC transmission protocol and allocating multi-path transmission weights to achieve a dynamic balance between image quality and bitrate while improving transmission reliability and resource utilization efficiency. Then, a mapping table is established between the strategy's parameter adjustment instructions and the underlying API. Next, based on the LSTM network's prediction of bandwidth over a short period, the quantization step size and coding block size parameters of H.265 / AV1 encoding are dynamically adjusted according to the prediction results. This allows for finding the optimal balance between bitrate and image quality, enabling a reduction in bitrate without compromising image quality, or an improvement in image quality while maintaining the same bitrate. This achieves Pareto optimization by finding the optimal solution on the bitrate-image quality curve. Secondly, by employing adaptive multipath transmission technology, the optimal combination of transmission protocols is dynamically selected based on real-time monitoring of network conditions such as packet loss rate, latency, and bandwidth. For example, when network conditions are good, a faster protocol is selected; when the network is unstable, a more fault-tolerant protocol is selected to ensure efficient and stable data transmission. Furthermore, the traffic allocation weights for each path are adjusted in real-time using a Kalman filter to balance the load. Simultaneously, by prioritizing key quality indicators such as smoothness of the image, audio-visual synchronization, and interactive response latency, an improved Best-Fit algorithm is used to allocate priority computing resources to these key quality indicators. A preemptive task mechanism is also implemented; for example, for interactive elements with high real-time requirements, more computing resources are immediately preempted to ensure their response speed and stability. Finally, by monitoring network conditions in real-time, a tiered degradation strategy is immediately activated when sudden network fluctuations are detected. Tiered degradation includes disabling background blur effects, pausing and locking frame rate modes, and implementing a low-resolution simplified mode. Degradation includes reducing image quality and minimizing interactive functions.

[0091] The effects of the above technical solution are as follows: By dynamically adjusting encoding parameters, optimizing transmission protocols, and rationally allocating computing resources, high live streaming quality can be maintained under different network environments. Simultaneously, Pareto optimization of bitrate-image quality is achieved, ensuring image clarity while controlling the bitrate and reducing network bandwidth pressure. Adaptive transmission protocol selection ensures data transmission stability and efficiency, reducing the impact of packet loss and latency. Dynamically adjusting network conditions by predicting network bandwidth adapts to various complex and changing network environments, improving the adaptability and reliability of live streaming interaction quality while ensuring live streaming quality and user experience. Edge computing resource scheduling algorithms prioritize the allocation of computing resources for key quality indicators, ensuring that users can still perform basic interactive operations even in poor network conditions. This avoids situations where live streaming cannot proceed normally due to network problems, enhances the anti-interference capability of the live streaming interaction segment, and ultimately ensures that the live streaming interaction segment meets the overall user experience.

[0092] In one specific embodiment, the feedback execution module includes:

[0093] Quality monitoring unit: continuously monitors changes in actual quality indicators after the implementation of the optimal quality optimization strategy;

[0094] Parameter adjustment unit: dynamically adjusts the quantization parameters and GOP structure of the video encoder through a fuzzy PID controller;

[0095] Transmission update unit: dynamically updates the forward error correction strategy and retransmission mechanism based on real-time network diagnostic results;

[0096] Resource reallocation unit: Dynamically adjusts the distribution of video processing tasks based on the load status of edge nodes;

[0097] Closed-loop control unit: Feeds the execution results back to the reinforcement learning network to form a continuously optimized control loop.

[0098] The working principle of the above technical solution is as follows: First, it collects video stuttering rate, audio-visual synchronization error, and interactive response latency quality indicators in real time after implementing the optimal quality optimization strategy. A sliding window is defined to calculate the moving average of these indicators, which serves as the dynamic quality benchmark for anomaly detection. Then, a pre-trained isolated forest model is used to learn and predict the input quality indicators, and the output is a dynamic quality benchmark that compares the identified deviation from the baseline with the anomaly detection. By comparing and analyzing the rules and identifying abnormal indicator data points, we can intuitively understand the actual effect of the optimization strategy and provide a basis for subsequent dynamic adjustments. Then, the quantization parameters of the video encoder directly affect the image quality and bitrate of the live video, while the GOP structure affects the encoding efficiency and playback smoothness. By analyzing the actual quality indicator changes of the quality monitoring unit, the centroid method is used to calculate precise adjustment amounts. A fuzzy PID controller is then used to automatically and flexibly adjust the quantization parameters and GOP structure of the video encoder to improve image clarity and playback smoothness. Secondly, through real-time network diagnostics, we obtain the current bandwidth, latency, and packet loss rate status information of the network. The forward error correction strategy can add redundant information during data transmission, dynamically calculate the proportion of redundant packets based on the real-time packet loss rate, and use RaptorQ encoding to recover the packet loss rate as much as possible so that a certain number of errors can be corrected at the receiving end. The retransmission mechanism retransmits the lost data when data loss is detected. A three-retransmission strategy is used for critical I-frames, and a single retransmission combined with a discard strategy is used for non-critical B / P frames, thereby improving the reliability and efficiency of data transmission. Finally, by monitoring the load of edge nodes in real time and using K-means clustering to identify overloaded nodes and construct a load heatmap, an improved Hungarian algorithm is used to match tasks with nodes based on the results of the load heatmap. Video transcoding tasks on overloaded nodes are migrated to idle nodes for processing, thereby dynamically adjusting the distribution of video processing tasks and achieving balanced resource utilization. This avoids the impact on overall performance due to excessive load on local nodes. The effects of policy execution are fed back to the reinforcement learning network and encoded into a state vector. Node IDs, timestamps, and protocol version metadata of the execution environment are added. Quantitative feedback data is set to trigger online learning of the model. The PPO algorithm is used to update the policy network parameters, and the learning rate is adaptively adjusted to adjust and optimize subsequent policies.

[0099] The effects of the above technical solution are as follows: Based on real-time monitoring of actual quality index changes, the fuzzy PID controller adjusts video encoding parameters, transmission strategies, and resource allocation in a timely manner according to changes in network conditions and quality indicators, ensuring that the live broadcast always maintains a good quality level; through forward error correction strategies and retransmission mechanisms, it can effectively cope with network fluctuations and data loss problems, improve the reliability of data transmission, and enhance the user's viewing experience; dynamically adjusting the distribution of video processing tasks according to the load status of edge nodes avoids the situation where some nodes are overloaded while others are idle, achieving balanced utilization of load resources and task migration efficiency, thereby not only improving the overall performance of the system but also reducing operating costs; the reinforcement learning network continuously adjusts its strategies based on feedback information to adapt to different network environments and user needs, improving the adaptability and stability of live broadcast interaction and providing users with a high-quality and stable live broadcast experience.

[0100] In one specific embodiment, the spatiotemporal attention neural network includes:

[0101] Spatiotemporal feature extraction and fusion component: In e-commerce live streaming scenarios, the spatiotemporal features of the live streaming process are extracted, and dynamic time warping and affine transformation are used to align and integrate the data in time and space. The spatiotemporal features include heterogeneous data such as video frame rate, audio latency, and user likes / comments / order hotspots.

[0102] Interaction Quality Prediction Component: By learning the relationship between spatiotemporal features and interaction quality in historical data, it predicts the real-time interaction quality of e-commerce live streaming based on the fused spatiotemporal features and outputs a quality assessment.

[0103] Anomaly detection component: Identifies abnormal situations in high-frequency periods or areas of the screen during the live broadcast through an attention mechanism, and takes timely measures to ensure the quality of the live broadcast.

[0104] The working principle of the above technical solution is as follows: First, the heterogeneous data collected from e-commerce live streaming is cleaned and normalized preprocessed. Temporal and spatial feature data are extracted through convolutional and recurrent layers. Simultaneously, DTW (Dynamic Time Warp) is used to align the time series of video frame rate and audio latency, eliminating audio-visual asynchrony. User click hotspot coordinates are mapped to a unified coordinate system to unify user device resolution differences, and a four-dimensional tensor is constructed to fuse spatiotemporal feature data. Then, the model is trained using a large amount of historical data to learn the mapping relationship between spatiotemporal features and interaction quality. The model then makes real-time predictions of quality evaluation indicators such as video smoothness, audio-visual synchronization error, and interaction response index, strengthening fine-grained indicators. This allows the model to predict the real-time interaction quality of the current e-commerce live stream and output quality evaluation results. Finally, in e-commerce live streaming, the attention mechanism calculates the anomaly weights for each time slice, focusing on periods or areas of high frequency of stuttering during the live stream, more sensitively capturing anomalies. After dynamic threshold comparison analysis, corresponding measures are taken in a timely manner to adjust encoding parameters, increase the regional bitrate ratio, and optimize transmission strategies.

[0105] The effects of the above technical solution are as follows: By comprehensively extracting spatiotemporal features such as video frame rate, audio latency, and user interaction hotspots, and performing fine alignment and integration, it can more accurately reflect the actual situation of e-commerce live streaming, thereby improving the accuracy of prediction. This allows live streaming operators to have a clearer understanding of the live streaming effect and adjust their live streaming strategies in a timely manner. By using an attention mechanism to focus on periods or areas of high frequency of stuttering, it enables rapid and accurate identification of abnormal situations during the live streaming process, allowing for swift action to address these issues and effectively reduce the impact of abnormal situations on live streaming quality, ensuring a better viewing experience for viewers. Furthermore, when abnormal situations are detected, the live streaming parameters for specific areas can be optimized in a targeted manner to avoid wasting resources, improve resource utilization efficiency, and simultaneously increase user engagement and satisfaction, thereby enhancing user stickiness.

[0106] One specific embodiment of the deep reinforcement learning network includes:

[0107] Quality optimization strategy exploration and generation component: Based on the current network status and the live streaming status of user interaction data, explore new parameter adjustments when bandwidth fluctuates, and generate quality optimization strategies to dynamically adjust bitrate, frame rate, and transmission protocol;

[0108] Model agent optimization strategy component: environment setting reward function, based on the Actor-Critic framework, and adopts a proximal strategy to optimize PPO, and achieves stable updates through importance quality index sampling;

[0109] Resource allocation maximizes benefits component: Based on priority experience replay PER and edge node periodic dual-delay deep deterministic policy gradient TD3, the agent allocates computing resources and network bandwidth according to the real-time requirements of the live broadcast and the resource status of the edge nodes, so as to maximize the quality of live broadcast interaction and resource utilization efficiency, and continuously improve the evaluation accuracy of the spatiotemporal network by using execution feedback.

[0110] The working principle of the above technical solution is as follows: First, real-time acquisition of current network status and user interaction data is used to construct a dynamic action space to adjust transmission protocol selection, frame rate mode switching, bitrate, and FEC redundancy. When bandwidth fluctuations are detected, a deep exploration mode is triggered, where components attempt to explore new combinations of parameter strategies. Based on the exploration results, corresponding quality optimization strategies are generated, dynamically adjusting bitrate, frame rate, and transmission protocol to adapt to changes in network status and maintain good live streaming quality. Then, the agent is optimized based on the Actor-Critic framework and the Proximal Policy Optimization (PPO) algorithm. By setting a reward function in the environment, the agent's behavior is evaluated and feedback is provided based on quality indicators such as live stream clarity, smoothness, and interactive activity. This enables the Actor to generate actions for quality optimization strategies, and the Critic evaluates the value of the actions generated by the Actor, thereby using the PPO algorithm to stably update quality indicators. Finally, optimal resource allocation is performed based on the real-time requirements of the live stream and the resource status of the edge nodes. First, the PER mechanism is used to replay the agent's experience to allocate priorities. Then, the edge nodes are periodically subjected to a dual-delay deep deterministic policy gradient algorithm, which enables the agent to allocate computing resources and network bandwidth according to the real-time requirements of the live stream and the resource status of the edge nodes. Furthermore, the evaluation accuracy of the spatiotemporal network is continuously improved by utilizing execution feedback information.

[0111] The effects of the above technical solutions are as follows: By adjusting bitrate, frame rate, and transmission protocol to adapt to changes in network bandwidth fluctuations, the live stream maintains good quality under different network conditions, reducing stuttering, screen tearing, and other issues, thus improving the user's viewing experience; The Actor-Critic framework and PPO algorithm, combined with feedback from the reward function, can stably and efficiently optimize the agent's strategy, enabling the agent to quickly learn the optimal actions under different live stream conditions, improving the effectiveness of live stream quality optimization; Furthermore, by utilizing the PER and TD3 algorithms, resource allocation is performed based on the real-time needs of the live stream and the resource status of edge nodes, achieving optimal utilization of computing resources and network bandwidth, thereby avoiding resource waste and overload, and improving the overall performance and resource utilization efficiency of the system.

[0112] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0113] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0114] It should be understood that, in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0115] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0116] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A real-time interactive quality assessment system for e-commerce live streaming based on edge computing, characterized in that, It includes a multimodal interaction data acquisition module: through a distributed edge computing node cluster, it collects multimodal interaction data of the live stream in real time and constructs a multidimensional quality feature vector. The multimodal interaction data includes video encoding parameters, audio quality indicators, user interaction behavior data and network transmission status data. Interaction Quality Analysis Module: Utilizes a lightweight spatiotemporal attention neural network model built into the edge nodes to perform real-time fusion analysis of multimodal interaction data and generate real-time interaction quality assessment indicators; Evaluation strategy generation module: The generator model of the deep reinforcement learning network is combined with real-time interactive quality evaluation indicators and historical optimization strategies to generate quality optimization schemes. The discriminator model evaluates the effectiveness of the quality optimization schemes based on user terminal feedback information. Dynamic adjustment strategy module: The quality optimization scheme is corrected in real time through an adaptive fuzzy inference system. The parameters of the quality optimization scheme generated by the deep reinforcement learning network are dynamically adjusted in combination with the current network bandwidth fluctuations and terminal device resource status. Feedback Execution Module: Based on the quality optimization scheme executed after dynamic parameter adjustment, combined with the distributed decision-making mechanism of edge nodes, it collaboratively optimizes video encoding parameters, transmission protocols and resource allocation strategies to dynamically control the quality of live streaming. The edge computing nodes then evaluate the quality of e-commerce live streaming in real time and feed the evaluation results back to the live streaming platform and merchants.

2. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 1, characterized in that, The multimodal interactive data acquisition module includes an edge deployment unit: a lightweight multimodal data acquisition array is deployed at both the edge nodes of the content delivery network (CDN) and the user terminal to form a distributed data acquisition network; Multimodal synchronization unit: The distributed data acquisition network synchronously collects multi-dimensional data such as video frame rate, encoding bit rate, audio latency, user likes, comment frequency and network jitter rate through a timestamp alignment mechanism, generating raw data that comprehensively reflects the interaction in the live broadcast room; Preprocessing unit: performs filtering, denoising, and normalization on the raw data to construct a structured interactive quality feature matrix; Feature extraction unit: Utilizes edge computing resources to perform real-time feature extraction on the interaction quality feature matrix, generating a multi-dimensional quality feature vector containing temporal and spatial features; Edge transmission unit: After the multi-dimensional quality feature vector after feature extraction is compressed using hierarchical compression technology, it is transmitted to the analysis node through the CDN edge network.

3. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 2, characterized in that, The interactive quality analysis module includes a data fusion unit that performs spatiotemporal alignment and fusion of video quality indicators, audio synchronization, interactive response latency, and network QoS parameters. Model input unit: The fused multidimensional features are input into the pre-trained spatiotemporal attention neural network, which includes a video quality attention branch and an interactive experience attention branch; Metric generation unit: The spatiotemporal attention neural network learns and outputs a set of evaluation metrics including screen smoothness score, audio-visual synchronization, interaction response index and comprehensive QoE score; Threshold comparison unit: compares and analyzes the real-time output evaluation indicators with the preset dynamic quality thresholds to generate a quality anomaly feature map; Output Unit: Performs data analysis on interaction quality based on the evaluation index set, generates analysis results, and encodes the analysis results into a quality state feature vector that can be recognized by a deep reinforcement learning network.

4. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 3, characterized in that, The evaluation policy generation module includes a policy generation unit: a generator model in a deep reinforcement learning network receives quality state feature vectors and historical optimization records, and generates a candidate policy set including bitrate adjustment, frame rate optimization, and transmission protocol selection, wherein the historical optimization records include past... The effectiveness of the bitrate adjustment strategy within a given time period; Strategy Evaluation Unit: The discriminator model calculates the expected quality improvement benefits of each candidate strategy in the candidate strategy set based on the actual experience data reported by the user terminal. The actual experience data includes lag rate, interaction success rate, and decoding time. Optimize the decision-making unit: Employ the reinforcement learning decision ε-greedy algorithm to balance exploring new strategies with utilizing known effective strategies, and select the optimal quality optimization strategy; Experience replay unit: Stores the decision-making process in the experience pool of edge nodes for online model updates; Model update unit: Periodically synchronizes and updates model parameters between edge nodes to maintain consistency of evaluation strategy.

5. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 4, characterized in that, The dynamic adjustment strategy module includes a strategy parsing unit: decomposing the optimal quality optimization strategy output by the reinforcement learning network into executable parameter adjustment instructions; Encoding Adjustment Unit: Dynamically adjusts H.265 / AV1 encoding parameters based on network bandwidth prediction to achieve Pareto optimization of bitrate-image quality; Transmission optimization unit: Employs adaptive multipath transmission technology to dynamically select the optimal combination of transmission protocols based on network conditions; Resource allocation unit: Through edge computing resource scheduling algorithms, prioritize computing resources for key quality indicators. Fault tolerance unit: When a sudden network fluctuation is detected, a degradation strategy is activated to ensure a basic interactive experience.

6. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 5, characterized in that, The feedback execution module includes a quality monitoring unit: continuously monitoring the changes in actual quality indicators after the execution of the optimal quality optimization strategy; Parameter adjustment unit: dynamically adjusts the quantization parameters and GOP structure of the video encoder through a fuzzy PID controller; Transmission update unit: dynamically updates the forward error correction strategy and retransmission mechanism based on real-time network diagnostic results; Resource reallocation unit: Dynamically adjusts the distribution of video processing tasks based on the load status of edge nodes; Closed-loop control unit: Feeds the execution results back to the reinforcement learning network to form a continuously optimized control loop.

7. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 6, characterized in that, The spatiotemporal attention neural network includes: Spatiotemporal feature extraction and fusion component: In e-commerce live streaming scenarios, the spatiotemporal features of the live streaming process are extracted, and dynamic time warping and affine transformation are used to align and integrate the data in time and space. The spatiotemporal features include heterogeneous data such as video frame rate, audio latency, and user likes / comments / order hotspots. Interaction Quality Prediction Component: By learning the relationship between spatiotemporal features and interaction quality in historical data, it predicts the real-time interaction quality of e-commerce live streaming based on the fused spatiotemporal features and outputs a quality assessment. Anomaly detection component: Identifies abnormal situations in high-frequency periods or areas of the screen during the live broadcast through an attention mechanism, and takes timely measures to ensure the quality of the live broadcast.

8. The e-commerce live streaming real-time interactive quality assessment system based on edge computing according to claim 7, characterized in that, The deep reinforcement learning network includes: Quality optimization strategy exploration and generation component: Based on the current network status and the live streaming status of user interaction data, explore new parameter adjustments when bandwidth fluctuates, and generate quality optimization strategies to dynamically adjust bitrate, frame rate, and transmission protocol; Model agent optimization strategy component: environment setting reward function, based on the Actor-Critic framework, and adopts a proximal strategy to optimize PPO, and achieves stable updates through importance quality index sampling; Resource allocation maximizes benefits component: Based on priority experience replay PER and edge node periodic dual-delay deep deterministic policy gradient TD3, the agent allocates computing resources and network bandwidth according to the real-time requirements of the live broadcast and the resource status of the edge nodes, so as to maximize the quality of live broadcast interaction and resource utilization efficiency, and continuously improve the evaluation accuracy of the spatiotemporal network by using execution feedback.

9. A computer device, comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the e-commerce live streaming real-time interactive quality assessment system based on edge computing as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the real-time interactive quality assessment system for e-commerce live streaming based on edge computing as described in any one of claims 1 to 8.

Images