Multi-task convolutional neural network for behavior sequence embedding modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fraud detection methods for online transaction platforms struggle to balance modeling capabilities with efficiency, leading to impractical latency for real-time fraud detection and ineffective integration of domain knowledge.
Innovation Solution
A multi-task convolutional neural network (MTCNN) model with a one-layer CNN architecture and multi-range kernels processes user behavior sequence data, including page identifiers, item identifiers, and view time, to identify short- and long-term fraudulent patterns, using random label weights and positional encoding for improved fraud detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional fraud detection methods use complex neural network architectures to improve modeling capabilities, then fraud detection accuracy is improved, but system latency increases making real-time detection impractical
Solution Approach 1:
The patent segments the user behavior sequence into multiple fixed-length windows (e.g., 5-time step windows with 3-time step overlap) that are processed in parallel through the CNN model. This segmentation allows the system to analyze behavior patterns in manageable chunks simultaneously, improving both detection accuracy through comprehensive pattern recognition and reducing latency through parallel processing of divided sequences.
2Measurement precision
If conventional methods use multi-layer neural network architectures to improve modeling capabilities, then fraud detection performance is improved, but device complexity and computational overhead increase
Solution Approach 1:
The patent extracts and removes the intermediate pooling layers from traditional multi-layer CNN architectures, retaining only the essential convolutional layers with multi-range kernels. This extraction eliminates unnecessary computational complexity while preserving the core functionality of detecting both short-term and long-term behavioral patterns, achieving simplified architecture with maintained or improved detection performance.
Solution Approach 2:
The patent employs multi-range kernels within a single convolutional layer that can simultaneously capture both short-term and long-term user behavior patterns. This multi-functional kernel design replaces the need for multiple specialized layers, allowing one layer to perform multiple detection functions (short-term fraud, long-term fraud, and their combinations) thereby reducing overall model complexity while maintaining comprehensive detection capabilities.
3Measurement precision
If conventional fraud detection systems process entire user behavior sequences to improve detection accuracy, then fraud identification capability is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary segmentation of the user behavior sequence into fixed-length windows before model processing. By pre-dividing the sequence into manageable 5-time step windows with defined overlap, the system prepares the data for efficient parallel processing. This preliminary action enables the model to quickly process multiple segments simultaneously rather than handling the entire long sequence at once, improving processing efficiency while maintaining comprehensive fraud detection capability.
Data Source
AI summary
Some aspects of the present technology relate to technologies for performing fraud detection on online transaction platforms through user behavior sequence data. In accordance with some configurations, a multi-task convolutional neural network (MTCNN) model is used to predict, in real-time, whether user behavior sequence data is indicative of fraudulent activity. To perform fraud detection in such configurations, a one-layer convolutional neural network architecture with multi-range kernels is employed. The MTCNN model receives a sequence of page browsing signals corresponding to a buyer. Each page browsing signal corresponds to a position in the sequence. One or more portions of the page browsing signals are selected. Each of the one or more portions of the page browsing signals and the corresponding position are embedded in one or more sequence embeddings. A fraud risk for each of the one or more sequence embeddings is predicted utilizing the MTCNN model.