Cloud mobile phone user information feedback processing method and related equipment

Abstract

The invention discloses a cloud mobile phone user information feedback processing method and related equipment, and relates to the technical field of cloud mobile phones, and the method comprises the steps: collecting multi-mode feedback data of a user, and carrying out the time sequence alignment based on a clock synchronization protocol; performing text sentiment analysis, video anomaly detection and interface layout defect identification on the multi-modal data by using a deep learning model, and fusing spatial and temporal features through a multi-source association algorithm to generate a problem feature set; matching a preset rule according to the problem feature set to generate a dynamic improvement strategy; cloud server resource configuration and interactive interface rendering logic are adjusted in real time based on a strategy, and model parameters are continuously optimized in combination with a closed-loop verification mechanism and federal learning. Through multi-modal data fusion and software and hardware collaborative optimization, accurate positioning of user feedback and adaptive improvement of service are realized, and the resource utilization efficiency and the user experience of the cloud mobile phone are effectively improved.

Description

Technical Field

[0001] This application relates to the technical field of cloud mobile phones, and in particular, to a method for processing user information feedback of cloud mobile phones and related devices. Background Art

[0002] As an emerging mobile computing device, a cloud mobile phone provides high-performance mobile services for users through dynamic scheduling of cloud resources. However, the continuous optimization of its functions and performance highly depends on the effective processing of user feedback. How to efficiently process user feedback to continuously optimize services has become a key challenge. Therefore, there is an urgent need for a method for processing user information feedback of cloud mobile phones to solve the above-mentioned technical problems. Summary of the Invention

[0003] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further elaborated in the Detailed Description section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the protection scope of the claimed technical solution.

[0004] In a first aspect, this application provides a method for processing user information feedback of a cloud mobile phone. The method includes:

[0005] Obtain multimodal feedback data of a user, where the multimodal feedback data includes text data, image data, and video data;

[0006] Extract features from the multimodal feedback data based on a deep learning model to determine a set of problem features of the user feedback;

[0007] Generate a dynamic improvement strategy according to the set of problem features;

[0008] Adjust the resource configuration of the cloud mobile phone server and update the interaction interface based on the dynamic improvement strategy.

[0009] In some embodiments, obtaining the multimodal feedback data of the user includes:

[0010] Real-time collect the screen operation trajectory data of the user through an embedded probe;

[0011] Generate a screen operation heat map based on a preset sampling frequency and the screen operation trajectory data;

[0012] Perform format standardization processing on text, images, and videos respectively to generate preprocessed text, preprocessed images, and preprocessed videos;

[0013] Align the timestamps of the screen operation heat map, preprocessed text, preprocessed images, and preprocessed videos based on a preset clock synchronization protocol to generate the multimodal feedback data of the user.

[0014] In some embodiments, feature extraction is performed on multimodal feedback data based on a deep learning model to determine a set of problem features of user feedback, including:

[0015] Perform semantic understanding analysis on text data based on text sentiment analysis model to determine the sentiment tendency score of user feedback text;

[0016] Perform key frame extraction and timing analysis on video data based on the video analysis model to determine the location of abnormal frames in the video stream and network transmission quality parameters;

[0017] Perform pixel analysis on image data based on image recognition models to identify interface layout defect characteristics;

[0018] Based on the multi-source association algorithm, a spatiotemporal alignment analysis is performed on the sentiment tendency score, abnormal frame position, network transmission quality parameters and interface layout defect characteristics to determine the problem feature set of user feedback.

[0019] In some implementations, generating a dynamic improvement strategy based on a problem feature set includes:

[0020] Determining an interface interaction optimization instruction based on a comparison result of the sentiment tendency score in the question feature set and a first preset threshold;

[0021] Determining a dynamic bandwidth scheduling strategy based on a comparison result of the network transmission quality parameter in the problem feature set and a second preset threshold;

[0022] Determine the interface element reload instruction based on the matching result of the interface layout defect feature in the problem feature set and the preset priority condition;

[0023] Generate dynamic improvement strategies based on interface interaction optimization instructions, bandwidth dynamic scheduling strategies, and interface element reloading instructions.

[0024] In some implementations, adjusting the resource configuration of the cloud phone server and updating the interactive interface based on the dynamic improvement strategy include:

[0025] Determine the network bandwidth allocation priority based on the bandwidth dynamic scheduling strategy in the dynamic improvement strategy;

[0026] Determine the loading priority of the interactive interface layout file based on the interface element reload instructions in the dynamic improvement strategy;

[0027] Based on preset resource mapping rules, network bandwidth allocation priority and loading priority are converted into executable configuration instructions;

[0028] Based on configuration instructions, update resource scheduling strategies and interactive interface rendering logic.

[0029] In some embodiments, it further includes:

[0030] Verify the adjusted cloud phone service effect based on a closed-loop verification mechanism to obtain a verification result;

[0031] Update the parameters of the deep learning model based on the verification result.

[0032] In some embodiments, it further includes:

[0033] Determine the real-time computing power load data of the cloud phone server based on a preset performance monitoring period;

[0034] Generate a dynamic instance allocation policy based on the comparison result between the real-time computing power load data and a preset computing power threshold;

[0035] Determine the video memory capacity ratio in the GPU computing power slicing unit based on the dynamic instance allocation policy;

[0036] Update the resource configuration parameters in the GPU computing power slicing unit based on the video memory capacity ratio.

[0037] In a second aspect, the present application proposes a processing device for cloud phone user information feedback, including:

[0038] A feedback data acquisition unit, configured to acquire multi-modal feedback data of a user, where the multi-modal feedback data includes text data, image data, and video data;

[0039] A feedback feature determination unit, which extracts features from the multi-modal feedback data based on a deep learning model to determine a set of problem features of the user's feedback;

[0040] An improvement strategy generation unit, configured to generate a dynamic improvement strategy according to the set of problem features;

[0041] A resource interaction optimization unit, based on the dynamic improvement strategy, adjusts the resource configuration of the cloud phone server and updates the interaction interface.

[0042] In a third aspect, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program stored in the memory, it implements the steps of the processing method for cloud phone user information feedback according to any one of the first aspects.

[0043] In a fourth aspect, the present application proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the processing method for cloud phone user information feedback according to any one of the first aspects.

[0044] In summary, through multi-modal data fusion analysis and deep learning feature extraction techniques, this application can comprehensively capture multi-dimensional information such as text, images, and videos in user feedback, improving the accuracy of problem location. By adopting a dynamic improvement strategy generation mechanism, it realizes real-time collaborative optimization of server resource configuration and interface interaction, effectively shortening the problem response time. By establishing a spatio-temporal correlation model between multi-modal data, it accurately identifies the internal relationship between user operation behaviors and system anomalies, forming a closed-loop optimization path and enhancing the adaptive improvement ability of the cloud phone service. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of illustrating the preferred embodiments and are not considered to be a limitation of this specification. Also, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0046] Figure 1 is a schematic flowchart of a method for processing cloud phone user information feedback provided by an embodiment of this application;

[0047] Figure 2 is a schematic structural diagram of a device for processing cloud phone user information feedback provided by an embodiment of this application;

[0048] Figure 3 is a schematic structural diagram of an electronic device for processing cloud phone user information feedback provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims, and above-mentioned drawings of this application are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices. The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments.

[0050] Please refer to Figure 1 , which is a schematic flowchart of a method for processing cloud phone user information feedback provided by an embodiment of this application, and specifically may include:

[0051] S110. Obtain the multi-modal feedback data of the user, where the multi-modal feedback data includes text data, image data, and video data;

[0052] Exemplarily, through multi-modal data acquisition technology, comprehensively capture the diverse feedback information generated by the user during the use of the cloud mobile phone. The multi-modal feedback data includes text, image, and video data, respectively reflecting the user's operation behavior and experience problems from different dimensions: the text data directly expresses the demands through the text description actively submitted by the user; the image data can intuitively display the interface layout defects or display anomalies; the video data records the dynamic operation process and the network transmission quality performance. The three cooperate to cover the static description and dynamic behavior of the user feedback, providing a multi-dimensional information basis for subsequent problem analysis.

[0053] The implementation of this step depends on the embedded probe and the standardized preprocessing mechanism. Real-time collect the user operation trajectory through the embedded probe, generate a screen operation heat map in combination with the preset sampling frequency to quantify the user interaction behavior; at the same time, perform format standardization processing on the text, image, and video data to eliminate data heterogeneity, and achieve timestamp alignment of multi-modal data through the clock synchronization protocol. This process ensures the consistency of different modal data in the time series and space dimensions, laying a data quality foundation for the subsequent correlation analysis of the deep learning model, and avoiding analysis deviations caused by data fragmentation.

[0054] S120. Based on the deep learning model, extract features from the multi-modal feedback data to determine the set of problem features of the user feedback;

[0055] Exemplarily, through the multi-modal deep learning framework, realize the structured parsing and semantic association of the user feedback information. First, establish parallel feature extraction channels, and respectively implement domain-adapted feature mining on the three types of feedback data of text, video, and image: the text sentiment analysis model captures the user's subjective evaluation features through semantic understanding, the video time series analysis model detects picture anomalies and network transmission defects, and the image recognition model locates the objective defect indicators of the interface layout. These three feature channels adopt a heterogeneous neural network architecture to ensure the efficient feature representation of different modal data.

[0056] Furthermore, realize cross-modal feature fusion through spatio-temporal correlation modeling to form a unified problem feature space. The system establishes a spatio-temporal mapping relationship between features based on the multi-source correlation algorithm, aligns the operation behavior features and system anomaly indicators by using the timestamp sequence after clock synchronization, and dynamically allocates the weight contribution degree of each modal feature through the attention mechanism. This multi-dimensional feature cross-validation mechanism can effectively distinguish the correlation strength between the user's subjective experience and objective system anomalies, and finally generate a set of problem features including multi-dimensional indicators such as sentiment tendency, hardware performance, and interaction design.

[0057] S130. Generate a dynamic improvement strategy based on the problem feature set;

[0058] Exemplarily, the generation of the dynamic improvement strategy is based on the multi-dimensional analysis results in the problem feature set. The problem features are mapped to specific optimization actions through preset rules and model reasoning. According to key indicators such as sentiment tendency scores, network transmission quality parameters, and interface layout defect features, the corresponding improvement priorities and resource adjustment requirements are respectively matched. For example, when the sentiment score is lower than the preset threshold, an interface interaction optimization instruction is triggered; when the network transmission parameter is abnormal, a dynamic bandwidth scheduling strategy is generated; and the interface defect features drive the interface element reloading instruction. Finally, a multi-dimensional collaborative improvement plan is integrated.

[0059] The formulation of the improvement strategy relies on a multi-source correlation analysis and dynamic weight allocation mechanism to ensure the pertinence and real-time nature of the strategy. Key problem features are screened through a preset correlation threshold, and an optimized instruction set with priority sorting is generated in combination with a reinforcement learning algorithm. For example, for video stuttering problems, GPU computing power and bandwidth resources are preferentially allocated, and for interface interaction problems, an optimized layout file is quickly loaded. The strategy generation process integrates real-time feedback data and historical optimization effects to form a decision-making model that can be dynamically adjusted, ensuring a high degree of adaptation of improvement measures to user needs.

[0060] S140. Based on the dynamic improvement strategy, adjust the resource configuration of the cloud mobile phone server and update the interaction interface.

[0061] Exemplarily, establish a real-time linkage channel between the strategy execution engine and the cloud service infrastructure to achieve an accurate conversion from software optimization instructions to hardware resource configuration. Compile the network bandwidth priority and interface loading strategy into underlying executable instructions through preset resource mapping rules, and rely on the Kubernetes device plugin interface to complete the dynamic allocation of GPU computing power slices and the instant deployment of network QoS strategies. This instruction and resource conversion mechanism ensures that the dynamic improvement strategy can penetrate the virtualization layer and directly reach the physical hardware resources, realizing server resource reconfiguration with millisecond-level response.

[0062] Furthermore, achieve seamless optimization of the user interaction experience through the incremental update technology of the interface rendering engine. Based on the XML layout hot reload mechanism, combined with the spatio-temporal feature data of the screen operation heat map, dynamically adjust the rendering priority of interface elements. Adopt a hierarchical rendering pipeline reconstruction technology to update the layout parameters while maintaining the continuity of the interface. Achieve efficient transmission of interface resource files through the RDMA protocol, ensuring the timing synchronization of the interface optimization strategy and hardware resource adjustment, and forming a closed-loop feedback loop.

[0063] In summary, through the synergistic effect of multi-modal data fusion and deep learning feature extraction, the embodiments of this application significantly improve the intelligent level of user feedback processing. Based on the spatio-temporal correlation modeling of multi-dimensional information such as text, images, and videos, it can accurately identify the internal relationship between user operation behaviors and system service anomalies, forming a two-way mapping mechanism for problem positioning and resource optimization. Through the generation and execution of dynamic improvement strategies, real-time collaborative optimization of cloud server resource configuration and interaction interface update is achieved, effectively balancing the improvement of user experience and the utilization efficiency of hardware resources. The closed-loop verification mechanism ensures the continuous iteration of optimization strategies, enabling the system to have an adaptive service improvement ability. This application not only strengthens the structured analysis ability of user feedback data but also constructs a dynamic optimization path for software and hardware linkage, ensuring the overall stability of the cloud phone while improving the real-time performance of service response.

[0064] In some instances, obtaining the multi-modal feedback data of the user includes:

[0065] Real-time collecting the screen operation trajectory data of the user through an embedded probe;

[0066] Generating a screen operation heat map based on the preset sampling frequency and the screen operation trajectory data;

[0067] Performing format standardization processing on text, images, and videos respectively to generate preprocessed text, preprocessed images, and preprocessed videos;

[0068] Based on the preset clock synchronization protocol, aligning the timestamps of the screen operation heat map, preprocessed text, preprocessed images, and preprocessed videos to generate the multi-modal feedback data of the user.

[0069] Exemplarily, this step realizes the real-time capture of user operation data through an embedded probe, and its technical principle is based on the collaborative work of a lightweight kernel monitoring module and a user-mode probe. The probe hijacks the SurfaceFlinger interface of the display service framework and records touch event coordinates, gesture trajectories, and interface element trigger sequences at a sampling frequency of 1000Hz. The data collection process uses the zero-copy technology to directly map the DMA buffer to ensure the low-latency acquisition of operation trajectory data (<2ms). At the same time, the probe integrates an abnormal operation detection function, filtering out invalid noise data through a preset whitelist of legal operations to improve the signal-to-noise ratio of the original data. This mechanism ensures data integrity while avoiding a significant impact on the runtime performance of the cloud phone (the measured CPU occupancy rate <0.3%).

[0070] Downsample the original operation trajectory data based on a preset sampling frequency (e.g., 100Hz) to generate a quantifiable screen operation heat map. The heat map uses a two-dimensional Gaussian kernel density estimation algorithm to map discrete operation points into a continuous density distribution map, and generates an interaction intensity matrix in the 0-1 interval through normalization processing. This matrix is aligned with the coordinate of the interface element layout XML file to establish the mapping relationship between operation hotspots and UI components. The heat map data is compressed by the Zstandard algorithm (compression rate ≥ 85%) and stored in a circular buffer, supporting a timeout elimination mechanism (default to retain the data of the most recent 15 minutes) to avoid storage resource overload.

[0071] The format standardization process adopts a multi-modal pipeline architecture, including three parallel processing channels: text data is encoded uniformly through Unicode normalization form C (NFC), and non-text characters are removed using regular expressions; image data performs color space conversion (sRGB → Linear RGB), gamma correction, and resolution normalization (target resolution 1920×1080); video data performs key frame extraction (I-frame interval ≤ 30) and frame rate standardization (30fps). Each channel integrates an FFmpeg hardware acceleration module and uses the MediaCodec API to achieve parallel processing, with a processing throughput of 120fps (1080P resolution). The standardized data is encapsulated in the Protobuf v3 format, including standardized parameter metadata, to ensure cross-modal data parsing consistency.

[0072] Adopt the IEEE 1588v2 precision clock synchronization protocol to perform cross-modal timestamp alignment on the screen operation heat map, preprocessed text, image, and video data, and strictly control the timing synchronization error within 1μs. During the alignment process, a global time reference is provided by an NTP server (Stratum 1 clock source), and a sliding window dynamic calibration algorithm (window size 1.3 - 1.5 times the clock error) is combined to compensate for network transmission delays. The finally generated multi-modal feedback data set contains the following metadata fields: timestamp sequence, device hardware fingerprint, operation context label, and data checksum, and is stored in a Ceph distributed storage system (EC 4+2 redundancy strategy) to provide a multi-dimensional data basis with consistent timing and spatial correlation for subsequent feature extraction.

[0073] In some instances, based on a deep learning model, feature extraction is performed on the multi-modal feedback data to determine the set of problem features in the user feedback, including: [[ID=X]] [[ID=Y]]

[0074] Perform semantic understanding and analysis on the text data based on a text sentiment analysis model to determine the sentiment tendency score of the user feedback text; [[ID=X]] [[ID=Y]]

[0075] Extract key frames and perform temporal analysis on video data based on a video analysis model to determine the positions of abnormal frames and network transmission quality parameters in the video stream;

[0076] Perform pixel analysis on image data based on an image recognition model to determine the defect features of the interface layout;

[0077] Perform spatio-temporal alignment analysis on sentiment tendency scores, positions of abnormal frames, network transmission quality parameters, and defect features of the interface layout based on a multi-source association algorithm to determine the set of problem features in the user feedback.

[0078] Exemplarily, text sentiment analysis adopts a deep learning framework that combines the BERT model based on the Transformer architecture and a bidirectional long short-term memory network. The input of the model is the UTF-8 encoded sequence of the user feedback text. After being converted into a 768-dimensional vector by the word embedding layer, the context semantic associations are captured through 12 layers of Transformer encoders. The bidirectional long short-term memory network performs temporal modeling on the output of the encoder to extract the sentiment features in the text. Finally, the sentiment tendency score in the range of 0-1 is output through the Sigmoid activation function to quantify the user satisfaction. The model training adopts a domain adaptation strategy and is fine-tuned in combination with a specific corpus of the cloud phone scenario to ensure the accurate parsing of professional terms such as "lag" and "delay".

[0079] The video analysis module is constructed based on a three-dimensional convolutional neural network, and the input is the key frame sequence of the H.265 encoded video stream. The key frames are sampled and extracted at an interval of 0.5 seconds and converted into a tensor with a size of 256×256×32 after decoding for input into the model. The three-dimensional convolutional kernel slides in the spatio-temporal dimension to detect features such as blurred images, mosaics, and abnormal frame rates. The network transmission quality parameters are obtained by parsing the SEI metadata of the video stream. Combining with the user-mode data plane framework to monitor the DMA queue depth, the real-time packet loss rate and delay metrics are calculated. The positions of abnormal frames are marked with timestamps with an accuracy of ±0.1 seconds, and a spatio-temporal mapping relationship is established with the network parameters.

[0080] Image recognition uses an improved ResNet-50 model, which introduces a dual-dimensional attention mechanism of channels and space. The input is an interface screenshot with a resolution of 1080P. After extracting multi-scale features through the spatial pyramid pooling layer, a 2048-dimensional feature vector is output. The model is trained on the cloud phone interface dataset through transfer learning and can identify layout defects such as misaligned buttons, overlapping text, and distorted icons. The defect features are output in accordance with the COCO annotation standard, including the bounding box coordinates and defect type labels. At the same time, the spatial matching degree score of the interface elements is calculated to quantify the degree of unreasonable layout.

[0081] The multi-source association algorithm realizes cross-modal feature fusion based on cosine similarity calculation and attention mechanism. First, map the sentiment tendency score, abnormal frame timestamp, network quality parameter, and interface defect coordinates to a unified spatio-temporal coordinate system. Use the IEEE 1588v2 protocol to align the timestamp sequences of each modality data and eliminate clock synchronization errors. Dynamically allocate feature weights through the multi-head attention mechanism, establish a causal relationship model between video stuttering and network packet loss rate, and at the same time associate the spatial distribution of interface defects with the user operation heat map. The finally generated problem feature set includes four types of structured indicators: sentiment polarity, abnormal frame index, network quality level, and layout defect priority, providing multi-dimensional decision-making basis for the dynamic improvement strategy.

[0082] In some instances, according to the problem feature set, generate a dynamic improvement strategy, including:

[0083] Determine the interface interaction optimization instruction based on the comparison result between the sentiment tendency score in the problem feature set and the first preset threshold;

[0084] Determine the bandwidth dynamic scheduling strategy based on the comparison result between the network transmission quality parameter in the problem feature set and the second preset threshold;

[0085] Determine the interface element reloading instruction based on the matching result between the interface layout defect feature in the problem feature set and the preset priority condition;

[0086] Generate a dynamic improvement strategy based on the interface interaction optimization instruction, the bandwidth dynamic scheduling strategy, and the interface element reloading instruction.

[0087] Exemplarily, when the sentiment tendency score of the user feedback text is lower than the first preset threshold, trigger the interface interaction optimization instruction generation process. The first preset threshold is obtained by training historical user satisfaction data and is default set to 0.35. The instruction generation module determines the interface element optimization priority based on the quantization result of the sentiment tendency score and combines the interaction intensity matrix of the screen operation heat map. The optimization instruction includes button layout adjustment parameters, floating window transparency adjustment coefficients, and response delay optimization target values, and is implemented through the XML layout file dynamic reloading mechanism to ensure the precise matching of the interface interaction logic and the user operation habit.

[0088] When network transmission quality parameters exceed the second preset threshold, the dynamic bandwidth scheduling policy generation process is initiated. This second preset threshold is dynamically adjusted based on the network service quality benchmark, with defaults including a 2% packet loss rate threshold and a 150ms latency threshold. Based on real-time network quality monitoring data, the policy generation module uses a reinforcement learning algorithm to calculate bandwidth allocation priorities, prioritizing the QoS level of the video streaming channel. The RDMA protocol is used for preemptive allocation of network bandwidth, with a minimum granularity of 1Mbps. Scheduling instructions are issued to the SmartNIC via the gRPC streaming protocol, ensuring millisecond-level response time for policy execution.

[0089] When the interface layout defect characteristics match the preset priority conditions, an interface element reload instruction is generated. The preset priority conditions are determined based on the spatial matching score of the interface elements and the user operation frequency. Interface elements with a score below 0.7 are marked as high-priority optimization targets. The reload instruction contains the XML layout file path, rendering priority parameters and incremental update identifier of the element to be optimized, and seamless replacement of interface elements is achieved through layered rendering pipeline technology. PCIe P2P direct write technology is used to transfer the layout file directly to the GPU video memory to ensure that the interface update delay is less than 200ms.

[0090] This approach integrates interface interaction optimization instructions, dynamic bandwidth scheduling strategies, and interface element reload instructions into a structured dynamic improvement strategy. This integration process utilizes an LSTM-based predictive model to optimize the execution order and resource usage of each strategy to avoid policy conflicts. The resulting dynamic improvement strategy is encapsulated in JSON format and includes the policy's effective time window, resource allocation weights, and rollback mechanism parameters. This strategy is deployed to the cloud phone server through the Kubernetes device plugin interface. The effectiveness of the strategy is verified using an A / B testing platform, forming a closed-loop optimization loop to ensure that improvement measures are highly compatible with user needs and that resources are optimally configured globally.

[0091] In some instances, based on dynamic improvement strategies, the cloud phone server's resource configuration is adjusted and the interactive interface is updated, including:

[0092] Determine the network bandwidth allocation priority based on the bandwidth dynamic scheduling strategy in the dynamic improvement strategy;

[0093] Determine the loading priority of the interactive interface layout file based on the interface element reload instructions in the dynamic improvement strategy;

[0094] Based on preset resource mapping rules, network bandwidth allocation priority and loading priority are converted into executable configuration instructions;

[0095] Based on configuration instructions, update resource scheduling strategies and interactive interface rendering logic.

[0096] Exemplarily, based on the bandwidth dynamic scheduling policy in the dynamic improvement strategy, the network bandwidth allocation priority is converted into an executable configuration instruction through a preset resource mapping rule. Specifically, the bandwidth priority is divided into integer levels from 0 to 255, and the preemptive allocation of network bandwidth is achieved relying on the RoCEv2 protocol. The policy execution module issues the configuration instruction to the intelligent network card supporting the RoCEv2 protocol through the gRPC bidirectional stream, and the minimum allocation granularity is 1 Mbps. At the same time, combined with the custom scheduling logic of the Kubernetes device plugin, the network QoS policy of the cloud mobile phone instance is dynamically adjusted to preferentially guarantee the bandwidth requirements of the video stream transmission channel and ensure that the transmission delay of critical services is less than 50 μs.

[0097] The interface element reload instruction parsing module determines the rendering order of the interactive interface layout file according to the loading priority parameter in the dynamic improvement strategy. The loading priority is dynamically generated based on the spatial matching degree score of the interface elements and the user operation heat map data, and is implemented using the incremental update mechanism of the XML layout file. The optimized layout file is directly transmitted to the GPU video memory through the PCIe P2P direct write technology, and the seamless replacement of the interface elements is achieved in combination with the hierarchical rendering pipeline technology. The RDMA protocol is used for resource file transmission in this process to ensure that the layout update delay is less than 200 ms while maintaining the continuity of the interface interaction.

[0098] The preset resource mapping rule compiles the network bandwidth priority and the interface loading strategy into a low-level executable instruction set. The network configuration instruction generates a CRD resource description file through the Kubernetes Operator to drive the Device Plugin interface to complete the dynamic allocation of GPU computing power slices. The interface rendering instruction is converted into an OpenGL ES rendering command queue and executed asynchronously by the Vulkan graphics pipeline. During the deployment process of the resource configuration instruction, the resource utilization rate is monitored in real time through Prometheus metrics, and the vGPU time slice rotation strategy is dynamically adjusted to ensure the precise matching of hardware resource allocation and policy requirements.

[0099] The update of the resource scheduling policy and the interface rendering logic adopts a timing synchronization mechanism to ensure that the clock error of each module is less than 1 μs through the IEEE 1588v2 protocol. The network bandwidth adjustment and the interface layout reload operation are atomically executed in a unified transaction to avoid policy failure caused by resource competition. After the execution is completed, the service metrics before and after optimization (such as the stuttering rate and response delay) are compared through the A / B test platform, and the verification result is fed back to the policy generation module. The deep learning model parameters are incrementally updated in combination with the federated learning framework to form a closed-loop optimization loop from policy execution to effect verification, continuously improving the collaborative efficiency of resource configuration and interface interaction.

[0100] In some instances, it also includes:

[0101] Verify the adjusted cloud mobile phone service effect based on the closed-loop verification mechanism to obtain the verification result;

[0102] Update the parameters of the deep learning model based on the verification result.

[0103] Exemplarily, the user traffic is allocated to the improvement group and the control group according to a preset ratio through the A / B test platform, and traffic mirroring and canary release are implemented using the Istio service mesh. The improvement group implements dynamic improvement strategies, and the control group maintains the original service configuration. The verification metrics include video stuttering rate, interface response latency, user satisfaction score, and NPS net promoter score. The data collection module monitors the server-side QoS metrics (such as P99 latency, GPU utilization) in real time through Prometheus custom Exporters, and combines client-side buried point technology to capture user interaction behavior data to form a multi-dimensional evaluation data set.

[0104] Use the two-sample T-test to perform a significance analysis of the performance metrics between the improvement group and the control group, and the confidence threshold is set to p < 0.05. For user subjective evaluation data (such as satisfaction score), after quantization using the 5-level Likert scale, the Wilcoxon signed-rank test is used to evaluate the improvement in experience before and after optimization. The system comprehensively calculates the effectiveness score of the improvement strategy based on the reduction ratio of the overall stuttering rate, the percentage increase in satisfaction, and the change rate of resource utilization. When the effectiveness score exceeds the preset threshold, the strategy is determined to be effective and the model update process is triggered.

[0105] The feedback data and verification results corresponding to the effective strategy are encapsulated as training samples and input into the federated learning framework for model parameter iteration. The participating nodes include the local models of each cloud mobile phone instance, and the gradient data is encrypted and transmitted through the homomorphic encryption algorithm (AES-256). The central aggregation server uses the weighted average algorithm to integrate the parameter differences of each node and applies differential privacy technology to add Gaussian noise to ensure the privacy and security of user data. The updated global model parameters are differentially compressed by the bsdiff algorithm to reduce the amount of transmitted data.

[0106] The compressed model parameters are directly transmitted to the GPU video memory through the PCIe P2P direct write technology, and the NCCL communication library is used to achieve parameter synchronization across GPU nodes, with a synchronization bandwidth of up to 125GB / s. During the deployment process, the hot update mechanism is used to dynamically replace the running model instances, and the Kubernetes Operator is used to monitor the model service status to ensure zero downtime during the update process. The updated model immediately participates in the new round of feature extraction and strategy generation, forming a complete closed loop from verification to optimization to improve the adaptive ability.

[0107] In some instances, it also includes:

[0108] Determine the real-time computing power load data of the cloud mobile phone server based on a preset performance monitoring period;

[0109] Generate a dynamic instance allocation strategy based on the comparison result between the real-time computing power load data and the preset computing power threshold;

[0110] Determine the video memory capacity ratio in the GPU computing power slicing unit based on the dynamic instance allocation strategy;

[0111] Update the resource configuration parameters in the GPU computing power slicing unit based on the video memory capacity ratio.

[0112] Exemplarily, use Prometheus custom Exporter to collect the real-time computing power load data of the cloud mobile phone server every 5 seconds, including GPU utilization rate, video memory occupancy rate, and CUDA core activity metrics. The data collection module is integrated into the Kubernetes device plugin framework, directly reads the GPU hardware status information through the NVML interface, and combines with the cAdvisor container monitoring data to construct a multi-dimensional computing power load feature vector. The sliding window mean filtering algorithm is used during the collection process to eliminate instantaneous fluctuation noise and ensure the stability and representativeness of the load data.

[0113] Compare and analyze the real-time computing power load data with the preset computing power threshold. The preset threshold is generated by a dynamic adjustment algorithm based on the historical load distribution. When the GPU utilization rate continuously exceeds the threshold for 3 monitoring periods, trigger the dynamic instance allocation strategy generation process. The strategy generation module calculates the optimal instance allocation plan based on the reinforcement learning algorithm, comprehensively considering the video memory capacity, computing unit isolation requirements, and task priorities. The generated strategy includes the number of vGPU instances, the video memory quota per instance, and the time slice rotation parameters, and is encapsulated as an executable instruction set through the Kubernetes CRD resource description file.

[0114] The dynamic instance allocation strategy drives the GPU computing power slicing unit to reallocate the video memory resources. The video memory capacity ratio is calculated according to the instance quota parameters in the strategy, and a hybrid scheduling mode of space isolation and time slice rotation is adopted. Each vGPU instance is allocated at least 5GB of independent video memory space, and the remaining video memory is dynamically allocated to high-priority tasks as needed. The slicing process is implemented through NVIDIA MIG technology to achieve hardware-level isolation, combined with the high bandwidth characteristics of the PCIe Gen4x16 bus, ensuring that the video memory access latency is less than 10μs. The slicing parameters are sent to the GPU driver layer in real time through the Kubernetes Device Plugin interface.

[0115] After being encoded by the Protobuf v3 protocol, the video memory capacity ratio update instruction is streamed to the cloud mobile phone server through gRPC. After parsing the instruction, the resource configuration engine calls the CUDA API to dynamically adjust the video memory mapping table and reset the vGPU time slice scheduler parameters. The update process uses transactional operations to ensure atomicity and synchronizes the resource configuration status of all nodes through the RDMA protocol. After the update is completed, the validity of video memory allocation is verified through the vGPU performance counter to ensure that the resource adjustment is completely consistent with the policy requirements, forming a complete closed-loop control loop from monitoring to optimization.

[0116] Please refer to Figure 2 , which is a schematic structural diagram of a processing device for cloud mobile phone user information feedback provided by an embodiment of the present application, including:

[0117] A feedback data acquisition unit 21, configured to acquire multimodal feedback data of a user, where the multimodal feedback data includes text data, image data, and video data;

[0118] A feedback feature determination unit 22, configured to extract features from the multimodal feedback data based on a deep learning model and determine a set of problem features of the user's feedback;

[0119] An improvement strategy generation unit 23, configured to generate a dynamic improvement strategy according to the set of problem features;

[0120] A resource interaction optimization unit 24, configured to adjust the resource configuration of the cloud mobile phone server and update the interaction interface based on the dynamic improvement strategy.

[0121] Please refer to Figure 3 , an embodiment of the present application further provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of any method for processing cloud mobile phone user information feedback.

[0122] Since the electronic device introduced in this embodiment is the device adopted for a processing device for cloud mobile phone user information feedback in an embodiment of the present application, based on the method introduced in the embodiment of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiment of the present application will not be described in detail here. As long as the device adopted by those skilled in the art to implement the method in the embodiment of the present application belongs to the scope to be protected by the present application.

[0123] In the specific implementation process, when the computer program 311 is executed by the processor, it can implement any implementation manner in the corresponding embodiment of the first aspect.

[0124] It should be noted that in the above embodiments, each embodiment is described with a particular emphasis. For parts not described in detail in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0125] Those skilled in the art should understand that the embodiments of the present application can provide a method, a system, or a computer program product. Therefore, the present application can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be implemented in the form of a computer program product implemented on one or more computer-readable storage media containing computer-readable program code.

[0126] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0127] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the specified functions in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0128] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the specified functions in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0129] The embodiments of the present application also provide a computer program product, which includes computer software instructions. When the computer software instructions run on a processing device, the processing device is caused to execute Figure 1 the process of a method for processing cloud mobile phone user information feedback corresponding to an embodiment.

[0130] A computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless manner. The computer-readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a server or a data center that includes one or more integrated available media. The available medium may be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0131] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0132] In several embodiments provided in the present application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in electrical, mechanical, or other forms.

[0133] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0134] In addition, the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware and / or software functional units.

[0135] When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device to execute all or part of the steps of the methods of various embodiments of this application.

[0136] The above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of this application.

[0137] Although the preferred embodiments of this specification have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of this specification.

[0138] Obviously, those skilled in the art can make various changes and deformations to this specification without departing from the spirit and scope of this specification. Thus, if these modifications and deformations of this specification fall within the scope of the claims of this specification and their equivalent technologies, this specification is also intended to include these modifications and deformations.

Claims

1. A method for processing user information feedback of cloud phones, characterized in that, including: Obtain the multimodal feedback data of the user, where the multimodal feedback data includes text data, image data, and video data; Based on a deep learning model, perform feature extraction on the multimodal feedback data to determine a set of problem features feedback by the user; Generate a dynamic improvement strategy according to the set of problem features; Based on the dynamic improvement strategy, adjust the resource configuration of the cloud mobile phone server and update the interaction interface.

2. The method according to claim 1, wherein The obtaining of the multimodal feedback data of the user includes: Real-time collect the screen operation trajectory data of the user through an embedded probe; Generate a screen operation heat map based on a preset sampling frequency and the screen operation trajectory data; Perform format standardization processing on text, images, and videos respectively to generate preprocessed text, preprocessed images, and preprocessed videos; Based on a preset clock synchronization protocol, perform timestamp alignment on the screen operation heat map, the preprocessed text, the preprocessed images, and the preprocessed videos to generate the multimodal feedback data of the user.

3. The method according to claim 1, characterized in that The performing of feature extraction on the multimodal feedback data based on a deep learning model to determine a set of problem features feedback by the user includes: Based on a text sentiment analysis model, perform semantic understanding analysis on the text data to determine the sentiment tendency score of the user feedback text; Based on a video analysis model, perform key frame extraction and timing analysis on the video data to determine the abnormal frame position and network transmission quality parameters in the video stream; Based on an image recognition model, perform pixel analysis on the image data to determine the interface layout defect features; Based on a multi-source association algorithm, perform spatio-temporal alignment analysis on the sentiment tendency score, the abnormal frame position, the network transmission quality parameters, and the interface layout defect features to determine a set of problem features feedback by the user.

4. The method according to claim 1, wherein The generating of a dynamic improvement strategy according to the set of problem features includes: Based on the comparison result between the sentiment tendency score in the set of problem features and a first preset threshold, determine an interface interaction optimization instruction; Based on the comparison result between the network transmission quality parameters in the set of problem features and a second preset threshold, determine a bandwidth dynamic scheduling strategy; Based on the matching result between the interface layout defect features in the set of problem features and a preset priority condition, determine an interface element reloading instruction; Generate a dynamic improvement strategy based on the interface interaction optimization instruction, the bandwidth dynamic scheduling strategy, and the interface element reloading instruction.

5. The method according to claim 1, characterized in that, The adjusting of the resource configuration of the cloud mobile phone server and the updating of the interaction interface based on the dynamic improvement strategy includes: Based on the bandwidth dynamic scheduling strategy in the dynamic improvement strategy, determine the network bandwidth allocation priority; Based on the interface element reloading instruction in the dynamic improvement strategy, determine the loading priority of the interaction interface layout file; Based on a preset resource mapping rule, convert the network bandwidth allocation priority and the loading priority into executable configuration instructions; Based on the configuration instructions, update the resource scheduling strategy and the interaction interface rendering logic.

6. The method according to claim 1, wherein It also includes: Verify the adjusted cloud mobile phone service effect based on a closed-loop verification mechanism to obtain a verification result; Update the parameters of the deep learning model based on the verification result.

7. The method according to claim 1, characterized in that It also includes: Determine the real-time computing power load data of the cloud mobile phone server based on a preset performance monitoring period; Generate a dynamic instance allocation strategy based on the comparison result between the real-time computing power load data and a preset computing power threshold; Determine the video memory capacity ratio in the GPU computing power slicing unit based on the dynamic instance allocation strategy; Update the resource configuration parameters in the GPU computing power slicing unit based on the video memory capacity ratio.

8. A processing device for cloud mobile phone user information feedback, characterized in that, It includes: A feedback data acquisition unit for acquiring the multimodal feedback data of the user, wherein the multimodal feedback data includes text data, image data, and video data; A feedback feature determination unit for extracting features from the multimodal feedback data based on a deep learning model to determine a set of problem features of the user feedback; An improvement strategy generation unit for generating a dynamic improvement strategy according to the set of problem features; A resource interaction optimization unit for adjusting the resource configuration of the cloud mobile phone server and updating the interaction interface based on the dynamic improvement strategy.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is used to implement the steps of the processing method for cloud mobile phone user information feedback as described in any one of claims 1 to 7 when executing the computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when executed by the processor, implements the processing method for cloud mobile phone user information feedback as described in any one of claims 1 to 7.

Images