Click Data Denoising via Graph Neural Network Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for denoising click data are labor-intensive and costly, with limited effectiveness due to the need for event tracking and significant noise in the data.
Innovation Solution
A method using graph models to extract feature vectors from click data, dividing them into sets for training binary classification models, predicting labels, and removing noise based on real and prediction labels, eliminating the need for event tracking and reducing costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If event tracking is used to denoise click data, then denoising can be performed, but labor cost and development cost increase significantly
Solution Approach 1:
The patent replaces the mechanical event tracking system with a graph neural network-based automated model. Instead of manually implementing event tracking code in applications, the system uses a trained graph model that automatically processes click data, extracts features, and identifies noise patterns, thereby eliminating the need for manual event tracking implementation while maintaining denoising effectiveness
Solution Approach 2:
The system enables self-service denoising by training the graph model on historical click data with known noise patterns. Once trained, the model autonomously processes new click data without requiring manual intervention or event tracking infrastructure, allowing the system to serve itself by automatically identifying and filtering noise based on learned patterns
2Reliability
If event tracking is implemented to filter noise, then some denoising can be achieved, but the denoising effect is not significant enough
Solution Approach 1:
The patent applies preliminary action by pre-training the graph neural network model on historical click data that contains various noise patterns. During this offline training phase, the model learns to recognize different types of noise and their characteristics. When processing new click data, the pre-trained model can immediately identify and filter noise without requiring real-time event tracking, significantly improving denoising effectiveness while reducing time consumption
Solution Approach 2:
The system transitions from the traditional single-dimension event tracking approach to a multi-dimensional graph-based representation. By constructing graphs where nodes represent users, items, and contexts, and edges represent interactions, the model captures complex relationships and patterns across multiple dimensions, enabling more effective noise identification that goes beyond simple duration-based filtering
3Measurement precision
If graph models are used to extract feature vectors and train binary classification models, then denoising accuracy is enhanced, but model training and prediction processes are added
Solution Approach 1:
The patent segments the denoising process into distinct modular components: graph construction module, feature extraction module, model training module, and prediction module. Each module performs a specific function and can be independently optimized or replaced. This segmentation makes the complex system more manageable and allows for targeted improvements in each component without redesigning the entire system
Data Source
AI summary
A method for denoising click data includes: acquiring a set of click data including pieces of first click data and a real label corresponding to each piece of first click data; extracting feature vectors of each piece of first click data with a graph model; dividing the feature vectors into sets of feature vectors; obtaining trained binary classification models by training binary classification models with the sets of feature vectors; for each of the feature vectors, obtaining prediction values corresponding to the feature vector by predicting the feature vector with the trained binary classification models, and calculating a prediction label of the feature vector based on the prediction values of the feature vector; and removing noise data in the pieces of first click data, based on the pieces of first click data, the real label and the prediction label of each piece of first click data.