Algorithm-driven multi-scene information efficient propagation method

Abstract

This application discloses an algorithm-driven method for efficient information propagation across multiple scenarios, comprising the following steps: 1) lightweight feature extraction; 2) edge computing deployment; 3) establishing an edge-cloud collaboration mechanism; and 4) implementing accuracy and latency control. This application effectively eliminates redundant features by combining lightweight feature extraction with PCA dimensionality reduction, balancing feature retention rate and computational efficiency, reducing resource consumption while ensuring matching accuracy. It deeply integrates the lightweight FM algorithm with edge computing, replacing centralized cloud computing, significantly reducing propagation latency and saving network bandwidth. It improves the edge-end collaboration and incremental synchronization mechanism, clearly defining responsibilities and achieving full coverage across multiple scenarios, enhancing adaptability to niche scenarios. It constructs a dual closed-loop control mechanism, dynamically adjusting parameters and strategies to achieve both matching accuracy and latency targets. The overall solution is simple and feasible, requiring no complex hardware support, adaptable to heterogeneous needs across multiple scenarios, and highly practical.

Description

Technical Field

[0001] This application relates to the field of algorithm-driven information dissemination technology, and in particular to an algorithm-driven method for efficient information dissemination in multiple scenarios. Background Technology

[0002] With the rapid popularization of internet technology and smart terminals, information dissemination has formed a distinct multi-scenario pattern, encompassing various heterogeneous application scenarios such as fragmented commuting, immersive reading, and real-time office work. Users have also placed higher and more stringent demands on the timeliness and accuracy of information dissemination. Algorithm-driven approaches are the core technical means to achieve efficient information dissemination. Through feature matching and dynamic adjustment of dissemination strategies, precise adaptation of information to scenarios and user needs can be achieved, and this approach is currently widely used in various information distribution scenarios. At the same time, edge computing technology is gradually emerging, providing effective technical support for overcoming the dissemination latency bottleneck caused by centralized cloud computing models. By deploying computing resources at the network edge, it enables local processing of information data, significantly reducing information transmission and computing latency, and further improving information dissemination efficiency.

[0003] However, currently, mainstream multi-scenario information dissemination technologies still employ a core architecture of centralized cloud computing and feature matching. While this can accomplish basic information distribution tasks, it has several shortcomings in practical engineering applications, making it difficult to meet the actual needs of efficient and accurate information dissemination across multiple scenarios. Specifically, on the one hand, the feature extraction stage of existing technologies lacks an effective lightweight processing mechanism, often retaining a large number of redundant features unrelated to matching accuracy. This not only results in the ineffective consumption of computing resources but also significantly prolongs the time spent on feature matching, directly leading to increased information dissemination latency. Even when some technologies attempt to optimize through dimensionality reduction algorithms, they fail to effectively balance feature retention rate and computational efficiency, easily leading to a decrease in matching accuracy. On the other hand, most existing technologies centrally deploy complex feature matching algorithms in the cloud. Even with the introduction of edge computing technology, deep integration of lightweight matching algorithms and edge nodes has not been achieved. Edge nodes only undertake simple data forwarding and preliminary caching, failing to fully leverage the advantages of local computing, and still exhibiting prominent problems such as excessive cloud transmission latency and excessive network bandwidth consumption.

[0004] In summary, in the current field of multi-scenario information dissemination technology, the integration of lightweight feature extraction, edge computing deployment, and algorithm optimization is still insufficient. The edge-cloud collaboration mechanism and dual-indicator control measures are also not perfect. Ultimately, this has led to prominent problems such as high latency, insufficient accuracy, and excessive resource consumption in the information dissemination process, which cannot meet the needs of efficient information dissemination driven by algorithms in multiple scenarios.

[0005] Therefore, we propose an algorithm-driven method for efficient information dissemination across multiple scenarios. Summary of the Invention

[0006] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.

[0007] To achieve the above objectives, the first aspect of this application proposes an algorithm-driven method for efficient information dissemination across multiple scenarios, comprising the following steps:

[0008] 1) First, lightweight feature extraction is performed. A two-stage processing strategy of key feature selection and PCA dimensionality reduction is adopted to extract core features on the scene side and core features on the information side, and dimensionality compression is performed on the selected features.

[0009] 2) Next, edge computing deployment is carried out, adopting a distributed edge node deployment mode, and lightweight feature matching algorithms are deployed to each edge node, with the edge nodes taking priority in handling scene information feature matching requests within their service range;

[0010] 3) Then, establish an edge-cloud collaboration mechanism, clarify the specific division of labor between edge nodes and the cloud, and use an incremental synchronization mechanism to realize data interaction between the two. The cloud is responsible for dynamic updating of feature weights, matching of niche scenarios as a backup, and full data aggregation and analysis, while the edge nodes are responsible for real-time matching calculation, high-frequency scenario data caching, and local feedback data collection.

[0011] 4) Finally, implement accuracy and delay control, establish a dual-indicator closed-loop control mechanism, and monitor and dynamically adjust the matching accuracy and propagation delay of scene information features in real time to ensure that both meet the preset standards.

[0012] In addition, the algorithm-driven multi-scenario information efficient propagation method proposed in this application may also have the following additional technical features:

[0013] As a further description of the above technical solution:

[0014] The core features on the scene side in step 1) specifically include network environment parameters, scene duration, and user instantaneous behavior characteristics, while the core features on the information side specifically include multimodal simplified features and propagation priority coefficients.

[0015] As a further description of the above technical solution:

[0016] The key feature selection in step 1) is completed using a feature importance ranking algorithm. Feature importance is calculated using mutual information, and the calculation formula is as follows: ,in The mutual information value representing the feature and matching accuracy. The set of features to be selected. To match the precision label set, Features

[0017] With precision labels The joint probability, Features The marginal probability is used to filter redundant features with mutual information values ​​below 0.2 during the screening process.

[0018] PCA dimensionality reduction algorithm compresses feature dimensions through feature vector transformation, and its transformation formula is as follows: ,in The feature matrix after dimensionality reduction. This is the original feature matrix after filtering. The PCA feature projection matrix is ​​used to compress the overall feature dimension to 30%-50% of the original dimension using a formula, while ensuring that the feature retention rate meets the requirements. ,in These are the eigenvalues ​​of the characteristic covariance matrix. The feature dimension after dimensionality reduction. This is the original feature dimension, which ensures that the subsequent matching accuracy is not significantly affected, reducing matching time and resource consumption from the source of computation.

[0019] As a further description of the above technical solution:

[0020] The lightweight feature matching algorithm in step 2) employs the lightweight FM algorithm, which achieves accurate matching between the two by calculating the matching score of scene information features. The formula for calculating the matching score is as follows: ,in The scene information feature matching score is used to determine the scene information feature matching score. For global bias,

[0021] , The first, respectively, on the scene side or the information side One core feature value, The first The latent vector of each feature The inner product of the latent vectors. This represents the total feature dimension after dimensionality reduction. Simultaneously, the lightweight FM algorithm is optimized with changes to its iteration count. The matching time for a single scene information feature is controlled within Within.

[0022] As a further description of the above technical solution:

[0023] In step 2), the edge nodes adopt a layered deployment strategy, and are reasonably laid out according to the regional scene distribution density. Each edge node corresponds to a fixed service range to ensure efficient handling of matching requests.

[0024] As a further description of the above technical solution:

[0025] In step 3), the incremental synchronization mechanism has a synchronization period of 5 minutes. The local feedback data synchronized by the edge node to the cloud includes matching accuracy data and propagation delay data. The update parameters pushed by the cloud to the edge node are feature weight update parameters, thereby avoiding the latency and bandwidth consumption caused by full data transmission.

[0026] As a further description of the above technical solution:

[0027] The niche scenarios in step 3) are scenarios where the frequency of matching requests is lower than a preset threshold. When an edge node encounters such a niche scenario and cannot complete an accurate match, it will automatically push the matching request to the cloud. The cloud will then complete the fallback matching and feed the matching results back to the edge node to ensure full coverage matching across multiple scenarios.

[0028] As a further description of the above technical solution:

[0029] Step 4) involves a dual-indicator closed-loop control mechanism, comprising two parts: matching accuracy control and propagation delay control. Matching accuracy is measured by correlation error, calculated using the formula: ,in For correlation error, This represents the total number of matched samples per unit of time. For the first The predicted matching score of each sample. For the first The actual matching score of each sample is used to set a correlation error threshold during the matching accuracy control process. That is, it needs to meet .

[0030] As a further description of the above technical solution:

[0031] Step 4) sets different latency thresholds for different scenarios in the propagation latency control. When the matching time exceeds the latency threshold of the corresponding scenario, the edge node will automatically trigger a lightweight degradation strategy to prioritize ensuring that the propagation latency meets the standard by temporarily simplifying the feature dimensions.

[0032] As a further description of the above technical solution:

[0033] When the matching accuracy falls below the preset correlation error threshold or the matching time exceeds the latency threshold for the corresponding scenario, the edge node will report the relevant abnormal data to the cloud. The cloud then adjusts the feature selection weights and lightweight FM algorithm parameters of the edge node using gradient descent. The parameter adjustment iterative formula is as follows: ,in The adjusted linear weights and latent vector parameters, These are the parameters before adjustment. The learning rate (satisfying) ), For correlation error The constructed loss function enables closed-loop optimization of matching accuracy and propagation delay.

[0034] Advantages of this invention:

[0035] The algorithm-driven method for efficient information propagation in multiple scenarios proposed in this application effectively eliminates redundant features by combining lightweight feature extraction with PCA dimensionality reduction, balancing feature retention rate and computational efficiency, reducing resource consumption while ensuring matching accuracy.

[0036] By deeply integrating the lightweight FM algorithm with edge computing, it replaces centralized cloud computing, significantly reducing propagation latency and saving network bandwidth.

[0037] Improve the end-edge collaboration and incremental synchronization mechanism, clarify the division of labor and achieve full coverage of multiple scenarios, and enhance the adaptability to niche scenarios;

[0038] A dual closed-loop control mechanism is constructed to dynamically adjust parameters and strategies, achieving both matching accuracy and latency targets.

[0039] The overall solution is simple and easy to implement, requires no complex hardware support, adapts to diverse needs in various scenarios, and is highly practical.

[0040] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0041] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0042] Figure 1 This is a schematic diagram of the structure of an algorithm-driven method for efficient information propagation across multiple scenarios according to an embodiment of this application;

[0043] Figure 2 This is a data comparison chart of an algorithm-driven, multi-scenario information efficient propagation method according to an embodiment 2 of this application. Detailed Implementation

[0044] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0045] The following describes, with reference to the accompanying drawings, an algorithm-driven method for efficient information propagation across multiple scenarios according to an embodiment of this application.

[0046] like Figure 1 As shown, the algorithm-driven multi-scenario information efficient propagation method of Embodiment 1 of this application may include the following steps:

[0047] 1) First, lightweight feature extraction is performed. A two-stage processing strategy of key feature selection and PCA dimensionality reduction is adopted to extract core features on the scene side and core features on the information side, and dimensionality compression is performed on the selected features.

[0048] 2) Next, edge computing deployment is carried out, adopting a distributed edge node deployment mode, and lightweight feature matching algorithms are deployed to each edge node, with the edge nodes taking priority in handling scene information feature matching requests within their service range;

[0049] 3) Then, establish an edge-cloud collaboration mechanism, clarify the specific division of labor between edge nodes and the cloud, and use an incremental synchronization mechanism to realize data interaction between the two. The cloud is responsible for dynamic updating of feature weights, matching of niche scenarios as a backup, and full data aggregation and analysis, while the edge nodes are responsible for real-time matching calculation, high-frequency scenario data caching, and local feedback data collection.

[0050] 4) Finally, implement accuracy and delay control, establish a dual-indicator closed-loop control mechanism, and monitor and dynamically adjust the matching accuracy and propagation delay of scene information features in real time to ensure that both meet the preset standards.

[0051] This solution addresses the issues of precision and latency imbalance, high resource consumption, and insufficient adaptability to multiple scenarios in existing technologies through lightweight feature extraction, edge computing deployment, edge-end collaboration, and dual-indicator control. Secondary feature processing reduces computational load from the source, edge deployment reduces cloud transmission latency, edge-end collaboration achieves efficient division of labor, and dual-indicator control ensures effective dissemination. The four elements work together to achieve efficient and accurate information dissemination in multiple scenarios.

[0052] like Figure 1 As shown:

[0053] The core features on the scene side in step 1) specifically include network environment parameters, scene duration, and user instantaneous behavior characteristics; the core features on the information side specifically include multimodal simplified features and propagation priority coefficient.

[0054] This solution clearly defines the key feature dimensions of the scene side and the information side to avoid feature redundancy or missing features. Scene features such as network environment and scene duration directly affect the adaptation of propagation delay, while information features such as multimodal simplified features and propagation priority determine the matching accuracy. The two correspond precisely, providing a reliable data foundation for subsequent feature matching and ensuring the rationality of the matching logic.

[0055] like Figure 1 As shown:

[0056] The key feature selection in step 1) is completed using a feature importance ranking algorithm. Feature importance is calculated using mutual information, and the calculation formula is as follows: ,in The mutual information value representing the feature and matching accuracy. The set of features to be selected. To match the precision label set, Features

[0057] With precision labels The joint probability, Features The marginal probability is used to filter redundant features with mutual information values ​​below 0.2 during the screening process.

[0058] PCA dimensionality reduction algorithm compresses feature dimensions through feature vector transformation, and its transformation formula is as follows: ,in The feature matrix after dimensionality reduction. This is the original feature matrix after filtering. The PCA feature projection matrix is ​​used to compress the overall feature dimension to 30%-50% of the original dimension using a formula, while ensuring that the feature retention rate meets the requirements. ,in These are the eigenvalues ​​of the characteristic covariance matrix. The feature dimension after dimensionality reduction. The original feature dimension is used to ensure that the subsequent matching accuracy is not significantly affected, thereby reducing matching time and resource consumption from the source of computation;

[0059] In this scheme, mutual information is used to accurately screen features that are strongly correlated with matching accuracy, eliminate redundant features, and avoid invalid calculations. The PCA dimensionality reduction algorithm achieves dimensionality compression through feature vector projection. At the same time, the feature retention rate constraint balances computational efficiency and matching accuracy, solving the problem of decreased accuracy or poor compression effect after dimensionality reduction in existing technologies, and providing support for subsequent low-latency matching.

[0060] like Figure 1 As shown:

[0061] The lightweight feature matching algorithm in step 2) employs the lightweight FM algorithm, which achieves accurate matching between the two by calculating the matching score of scene information features. The formula for calculating the matching score is as follows: ,in The scene information feature matching score is used to determine the scene information feature matching score. For global bias,

[0062] , The first, respectively, on the scene side or the information side One core feature value, The first The latent vector of each feature The inner product of the latent vectors. This represents the total feature dimension after dimensionality reduction. Simultaneously, the lightweight FM algorithm is optimized with changes to its iteration count. The matching time for a single scene information feature is controlled within within;

[0063] In this solution, the lightweight FM algorithm achieves accurate calculation of feature interaction relationships without a large number of iterations, compared to complex deep learning models. By using linear terms and feature interaction terms in the formula, the matching degree between the scene and the information is accurately quantified, the number of iterations and matching time are optimized, the algorithm's computational latency is further reduced, and it is adapted to the local computing capabilities of edge nodes. At the same time, it ensures the accuracy of the matching score, achieving the dual goals of lightweight and high precision.

[0064] like Figure 1 As shown:

[0065] In step 2), the edge nodes adopt a layered deployment strategy, and are reasonably laid out according to the regional scene distribution density. Each edge node corresponds to a fixed service range to ensure efficient handling of matching requests.

[0066] This solution employs a layered layout based on scene distribution density, enabling the rational allocation of edge computing resources. This avoids overloading of nodes in high-density scenes and resource waste in low-density scenes. The fixed service range ensures that matching requests are processed nearby, reducing cross-node data transmission time, further reducing propagation latency, improving the adaptation efficiency of edge nodes to local scenes, and strengthening the localization advantages of edge computing.

[0067] like Figure 1 As shown:

[0068] In step 3), the synchronization period of the incremental synchronization mechanism is set to 5 minutes. The local feedback data synchronized by the edge node to the cloud includes matching accuracy data and propagation delay data. The update parameters pushed by the cloud to the edge node are feature weight update parameters, thereby avoiding the latency and bandwidth consumption caused by full data transmission.

[0069] In this solution, incremental synchronization only transmits changed data and parameters. Compared with full synchronization, it can significantly reduce the amount of data transmitted, reduce bandwidth consumption and synchronization latency. The 5-minute synchronization cycle balances synchronization timeliness and computational overhead, ensuring that the cloud can obtain the running data of edge nodes in real time, while avoiding frequent synchronization that consumes too many resources. It focuses on matching accuracy, latency data and feature weight parameters to achieve targeted data synchronization and ensure the consistency of parameters between the end and the edge.

[0070] like Figure 1 As shown:

[0071] The niche scenarios in step 3) are scenarios where the frequency of matching requests is lower than a preset threshold. When an edge node encounters such a niche scenario and cannot complete an accurate match, it will automatically push the matching request to the cloud. The cloud will then complete the fallback matching and feed the matching results back to the edge node to ensure full coverage matching of multiple scenarios.

[0072] In this solution, niche scenarios have low request frequency, and edge nodes lack sufficient sample data to support accurate matching. Relying solely on edge nodes would lead to insufficient adaptation to niche scenarios. By using cloud-based backup matching, the full amount of data and stronger computing power in the cloud can be used to complete matching tasks that edge nodes cannot handle. At the same time, edge nodes are responsible for high-frequency scenarios, realizing a division of labor between efficient edge processing and cloud backup coverage, ensuring no blind spots in multiple scenarios, and solving the problem of insufficient coverage of multiple scenarios in existing technologies.

[0073] like Figure 1 As shown:

[0074] Step 4) involves a dual-indicator closed-loop control mechanism, comprising two parts: matching accuracy control and propagation delay control. Matching accuracy is measured by correlation error, calculated using the formula: ,in For correlation error, This represents the total number of matched samples per unit of time. For the first The predicted matching score of each sample. For the first The actual matching score of each sample is used to set a correlation error threshold during the matching accuracy control process. That is, it needs to meet ;

[0075] This solution uses a correlation error formula to quantify the deviation between the predicted matching score and the actual score, accurately measure the matching accuracy, set a 5% error threshold, clarify the accuracy standard, avoid information mismatch and ineffective dissemination due to low accuracy, and provide data basis for subsequent parameter adjustments through real-time accuracy monitoring, thus building a closed loop for accuracy control and ensuring the accuracy of information dissemination in multiple scenarios.

[0076] like Figure 1 As shown:

[0077] Step 4) sets different latency thresholds for different scenarios in the propagation latency control. When the matching time exceeds the latency threshold of the corresponding scenario, the edge node will automatically trigger a lightweight degradation strategy to prioritize ensuring that the propagation latency meets the standard by temporarily simplifying the feature dimensions.

[0078] This solution addresses different requirements for propagation latency (e.g., latency tolerance is higher in weak network scenarios than in WiFi scenarios). Differentiated threshold settings can improve the targeting of latency management. When latency exceeds the limit, a simplified feature dimension degradation strategy can be used to temporarily reduce the amount of computation, quickly shorten the matching time, prioritize the timeliness of propagation, and avoid affecting the user experience due to excessive latency.

[0079] like Figure 1 As shown:

[0080] When the matching accuracy falls below the preset correlation error threshold or the matching time exceeds the latency threshold for the corresponding scenario, the edge node will report the relevant abnormal data to the cloud. The cloud then adjusts the feature selection weights and lightweight FM algorithm parameters of the edge node using gradient descent. The parameter adjustment iterative formula is as follows: ,in The adjusted linear weights and latent vector parameters, These are the parameters before adjustment. The learning rate (satisfying) ), For correlation error The constructed loss function achieves closed-loop optimization of matching accuracy and propagation delay;

[0081] In this solution, when anomalies occur in accuracy or latency, abnormal data is fed back through edge nodes. The cloud uses gradient descent to adjust relevant parameters. Gradient descent can gradually reduce the loss function value through iteration, thereby achieving precise optimization of parameters. Adjusting feature selection weights and FM algorithm parameters can specifically solve the problems of insufficient accuracy or excessive latency, ensuring that matching accuracy and latency remain stable and meet standards in the long term.

[0082] Example 2, further illustrated below with a specific implementation case:

[0083] In the specific implementation process, cloud servers (used for feature weight updates, niche scenario matching as a fallback, and full data aggregation), distributed edge nodes (using Huawei Atlas 500 edge computing nodes), and terminal devices (mobile phones, tablets, and smart terminals) are employed to cover three typical heterogeneous scenarios: commuting, office, and home. Basic parameters are then preset.

[0084] The correlation threshold between features and matching accuracy is 0.2. After PCA dimensionality reduction, the feature dimension is 40% of the original dimension, and the feature retention rate is ≥95%. The lightweight FM algorithm has 45 iterations and the time for a single match is ≤100ms.

[0085] The incremental synchronization cycle between the edge cloud and the cloud is 5 minutes.

[0086] correlation error threshold =4.5% (≤5%); commuting scenario (weak network) latency threshold 200ms, office scenario (WiFi) latency threshold 80ms, home scenario (WiFi / 5G) latency threshold 100ms;

[0087] The learning rate for gradient descent is α = 0.05 (0 < α < 1).

[0088] The specific implementation steps are as follows:

[0089] Lightweight Feature Extraction: Terminal devices collect scene-side and information-side features in real time. Core scene-side features include network environment parameters (signal strength, network type 4G / 5G / WiFi), scene duration (in minutes), and instantaneous user behavior features (click frequency, dwell time). Core information-side features include multimodal simplified features (image and text keywords, voice and semantic tags), and propagation priority coefficients (levels 1-5, with level 5 being the highest). Mutual information is used to calculate feature importance, and the formula is applied to achieve this. Calculate the mutual information value between each feature and the matching accuracy, remove redundant features with a mutual information value below 0.2 (such as features from user history that are irrelevant to the current scene), and then use the PCA dimensionality reduction algorithm according to the formula. The selected 100-dimensional features are compressed to 40 dimensions, and the feature retention rate is calculated. It meets the requirement of ≥95%, reducing computation time from the source.

[0090] Edge computing deployment: A distributed, layered deployment model is adopted, with three edge nodes corresponding to commuting, office, and home scenarios respectively. The nodes are rationally distributed based on the density of each scenario (edge ​​nodes in the office scenario have higher computing power than those in other scenarios). Each edge node covers a fixed service area, prioritizing scenario-information feature matching requests within its own region. A lightweight FM algorithm is deployed to each edge node, using the formula... The scene information feature matching score was calculated, and the algorithm iteration count was set to 45 times (≤50 times). After testing, the time consumption for matching a single scene-information feature was stable at 85-95ms, which meets the requirement of ≤100ms and realizes localized low-latency matching.

[0091] Edge-cloud collaboration: Clearly define the division of labor between edge nodes and the cloud. Edge nodes are responsible for real-time matching calculations, data caching for high-frequency scenarios (peak commuting hours, office hours, and evenings at home), and collection of local feedback data (matching accuracy, propagation latency). The cloud is responsible for dynamic updates of feature weights, backup matching for niche scenarios (such as niche interest information late at night at home), and full data aggregation and analysis. An incremental synchronization mechanism is adopted, with a synchronization cycle set at 5 minutes. Every 5 minutes, edge nodes synchronize local feedback data to the cloud. After updating feature weights based on the feedback data, the cloud incrementally pushes updated parameters to the edge nodes to avoid latency and bandwidth consumption caused by full data transmission. When an edge node encounters a niche scenario where the frequency of matching requests is lower than a preset threshold (≤5 times per hour), it automatically pushes the matching request to the cloud. The cloud then uses the full data to complete the backup matching and feeds the results back to the edge nodes, ensuring full coverage of the three scenarios.

[0092] Accuracy and Delay Control: Establish a dual-indicator closed-loop control mechanism, matching accuracy control through formulas. Calculate the correlation error and set an error threshold. =4.5%, real-time acquisition of edge node matching results. When the matching accuracy of an edge node is lower than 4.5%, abnormal data is reported to the cloud. Propagation latency control sets differentiated thresholds for three scenarios. When the matching time exceeds the corresponding scenario threshold, the edge node automatically triggers a lightweight degradation strategy, temporarily simplifying the feature dimension from 40 dimensions to 30 dimensions to prioritize meeting latency targets. The cloud adjusts the feature selection weights and lightweight FM algorithm parameters of the edge nodes using gradient descent. The adjustment formula is as follows: ,in For correlation error The constructed loss function, through iterative adjustment, restores the matching accuracy to above 4.5%, achieving closed-loop optimization of matching accuracy and propagation delay.

[0093] In summary, the algorithm-driven multi-scenario information efficient propagation method according to Embodiment 2 of this application (such as...) Figure 2 As shown), the average information propagation latency across multiple scenarios is 88ms, a 62% reduction compared to existing centralized cloud computing models. The average matching accuracy is 95.7%, the correlation error remains stable within 4.3%, network bandwidth consumption is reduced by 58%, and the matching coverage for niche scenarios is 100%. This effectively solves the problems of high latency, low accuracy, and insufficient coverage across multiple scenarios in existing technologies (such as...). Figure 2 ).

[0094] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0095] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0096] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. An algorithm-driven method for efficient information dissemination across multiple scenarios, characterized in that: Includes the following steps: 1) First, lightweight feature extraction is performed. A two-stage processing strategy of key feature selection and PCA dimensionality reduction is adopted to extract core features on the scene side and core features on the information side, and dimensionality compression is performed on the selected features. 2) Next, edge computing deployment is carried out, adopting a distributed edge node deployment mode, and lightweight feature matching algorithms are deployed to each edge node, with the edge nodes taking priority in handling scene information feature matching requests within their service range; 3) Then, establish an edge-cloud collaboration mechanism, clarify the specific division of labor between edge nodes and the cloud, and use an incremental synchronization mechanism to realize data interaction between the two. The cloud is responsible for dynamic updating of feature weights, matching of niche scenarios as a backup, and full data aggregation and analysis, while the edge nodes are responsible for real-time matching calculation, high-frequency scenario data caching, and local feedback data collection. 4) Finally, implement accuracy and delay control, establish a dual-indicator closed-loop control mechanism, and monitor and dynamically adjust the matching accuracy and propagation delay of scene information features in real time to ensure that both meet the preset standards.

2. The algorithm-driven multi-scenario information efficient propagation method according to claim 1, characterized in that, The core features on the scene side in step 1) specifically include network environment parameters, scene duration, and user instantaneous behavior characteristics, while the core features on the information side specifically include multimodal simplified features and propagation priority coefficients.

3. The algorithm-driven multi-scenario information efficient propagation method according to claim 1, characterized in that, The key feature selection in step 1 is completed using a feature importance ranking algorithm. Feature importance is calculated using mutual information, and the calculation formula is as follows: ,in The mutual information value representing the feature and matching accuracy. The set of features to be selected. To match the precision label set, Features With precision labels The joint probability, Features The marginal probability is used to filter redundant features with mutual information values ​​below 0.2 during the screening process. PCA dimensionality reduction algorithm compresses feature dimensions through feature vector transformation, and its transformation formula is as follows: ,in The feature matrix after dimensionality reduction. This is the original feature matrix after filtering. The PCA feature projection matrix is ​​used to compress the overall feature dimension to 30%-50% of the original dimension using a formula, while ensuring that the feature retention rate meets the requirements. ,in These are the eigenvalues ​​of the characteristic covariance matrix. The feature dimension after dimensionality reduction. This is the original feature dimension, which ensures that the subsequent matching accuracy is not significantly affected, reducing matching time and resource consumption from the source of computation.

4. The algorithm-driven multi-scenario information efficient propagation method according to claim 1, characterized in that, The lightweight feature matching algorithm in step 2) employs the lightweight FM algorithm, which achieves accurate matching between the two by calculating the matching score of scene information features. The formula for calculating the matching score is as follows: ,in The scene information feature matching score is used to determine the scene information feature matching score. For global bias, , The first, respectively, on the scene side or the information side One core feature value, The first The latent vector of each feature The inner product of the latent vectors. This represents the total feature dimension after dimensionality reduction. Simultaneously, the lightweight FM algorithm is optimized with changes to its iteration count. The matching time for a single scene information feature is controlled within Within.

5. The algorithm-driven method for efficient information propagation across multiple scenarios according to claim 1, characterized in that, In step 2), the edge nodes adopt a layered deployment strategy, and are reasonably laid out according to the regional scene distribution density. Each edge node corresponds to a fixed service range to ensure efficient handling of matching requests.

6. The algorithm-driven method for efficient information propagation across multiple scenarios according to claim 1, characterized in that, In step 3), the incremental synchronization mechanism has a synchronization period of 5 minutes. The local feedback data synchronized by the edge node to the cloud includes matching accuracy data and propagation delay data. The update parameters pushed by the cloud to the edge node are feature weight update parameters, thereby avoiding the latency and bandwidth consumption caused by full data transmission.

7. The algorithm-driven method for efficient information propagation across multiple scenarios according to claim 1, characterized in that, The niche scenarios in step 3) are scenarios where the frequency of matching requests is lower than a preset threshold. When an edge node encounters such a niche scenario and cannot complete an accurate match, it will automatically push the matching request to the cloud. The cloud will then complete the fallback matching and feed the matching results back to the edge node to ensure full coverage matching across multiple scenarios.

8. The algorithm-driven method for efficient information propagation across multiple scenarios according to claim 1, characterized in that, Step 4) involves a dual-indicator closed-loop control mechanism, comprising two parts: matching accuracy control and propagation delay control. Matching accuracy is measured by correlation error, calculated using the formula: ,in For correlation error, This represents the total number of matched samples per unit of time. For the first The predicted matching score of each sample. For the first The actual matching score of each sample is used to set a correlation error threshold during the matching accuracy control process. That is, it needs to meet .

9. The algorithm-driven method for efficient information propagation across multiple scenarios according to claim 1, characterized in that, Step 4) sets different latency thresholds for different scenarios in the propagation latency control. When the matching time exceeds the latency threshold of the corresponding scenario, the edge node will automatically trigger a lightweight degradation strategy to prioritize ensuring that the propagation latency meets the standard by temporarily simplifying the feature dimensions.

10. The algorithm-driven multi-scenario information efficient propagation method according to claim 9, characterized in that, When the matching accuracy falls below the preset correlation error threshold or the matching time exceeds the latency threshold for the corresponding scenario, the edge node will report the relevant abnormal data to the cloud. The cloud then adjusts the feature selection weights and lightweight FM algorithm parameters of the edge node using gradient descent. The parameter adjustment iterative formula is as follows: ,in The adjusted linear weights and latent vector parameters, These are the parameters before adjustment. The learning rate (satisfying) ), For correlation error The constructed loss function enables closed-loop optimization of matching accuracy and propagation delay.

Images