Two-Level Machine Learning for Network Traffic Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence systems face challenges in efficiently reducing network traffic due to the complexity of obtaining high-quality data, the need for specialized knowledge to design and integrate AI solutions, and the difficulty in reviewing AI results, which hinders the practical implementation of AI in real-time user intent analysis and communication suppression.
Innovation Solution
A two-level machine learning approach is implemented, using a front-end machine learning model for preliminary real-time analysis and a back-end model for extensive analysis, with confidence scores determining when to extract and process data streams, thereby improving user intent detection and reducing network resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a single machine learning model is used for real-time user intent detection, then response time is reduced, but accuracy and confidence scores deteriorate
Solution Approach 1:
The patent divides the machine learning processing into two separate models: a front-end model for rapid preliminary classification and a back-end model for thorough analysis. This segmentation allows the system to achieve both speed (through the fast front-end model) and accuracy (through the comprehensive back-end model), resolving the contradiction between response time and confidence score.
Solution Approach 2:
The patent introduces a confidence score threshold dimension to determine when to engage the back-end model. By adding this decision layer, the system can quickly reject low-confidence predictions (maintaining speed) while thoroughly analyzing high-confidence cases (improving accuracy), thus resolving the trade-off between response time and measurement precision.
2Measurement precision
If the entire data stream is processed using the back-end machine learning model, then accuracy and confidence scores are improved, but network resources and processing time increase
Solution Approach 1:
The patent extracts only the necessary portion of the data stream (the portion requiring back-end processing based on confidence score thresholds) rather than processing the entire stream. This extraction approach reduces network resource consumption while maintaining high accuracy for the cases that require it, resolving the contradiction between measurement precision and energy usage.
Solution Approach 2:
The patent applies partial action by using the back-end model only for a subset of data streams that meet specific confidence criteria, rather than applying it to all data. This partial processing reduces overall resource consumption while maintaining high accuracy for critical cases, resolving the trade-off between measurement precision and energy usage.
3Speed
If conventional machine learning models are used for real-time analysis, then processing speed is maintained, but adaptability to improve confidence scores deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the back-end model's results are used to retrain the front-end model. This feedback loop enables the system to adapt and improve its confidence scores over time while maintaining real-time processing speed, resolving the contradiction between processing speed and adaptability.
Solution Approach 2:
The patent creates a dynamic system where the front-end model's performance is continuously improved through feedback from the back-end model. This dynamic adaptation allows the system to maintain speed while progressively improving its accuracy and confidence scores, resolving the contradiction between processing speed and adaptability.
Data Source
AI summary
Systems and methods for reducing network traffic associated with a service. In some aspects, the systems and methods provide for using a first machine learning model to process a data stream for a communication with a user and generate a confidence score regarding whether to assign a communication suppression flag to the user account. Based on the confidence score not exceeding a first threshold, a communication suppression flag is not assigned to the user account. Based on the confidence score being between first and second thresholds, at least a portion of the data stream is extracted based on temporal proximity to a time stamp of an intent of the user to not receive further communications. Using a second machine learning model, the extracted portion of the data stream is processed to generate a prediction regarding whether to assign a communication suppression flag to the user account.