Deep Learning for Abusive User Activity Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to effectively detect abusive activities in online networks due to the need for separate models for each site surface, reliance on hand-engineered features, and difficulty in scaling to adapt against adversarial attacks.
Innovation Solution
A deep learning model that operates directly on raw user activity sequences, using standardized tokens and natural language processing to classify requests as abusive or not, leveraging timing information and frequency of request paths without human curation, and employing an unsupervised outlier-detection model for label generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning models are used for abusive activity detection, then detection capability is provided, but the system requires separate models for each site surface and relies on hand-engineered features, increasing device complexity and reducing adaptability
Solution Approach 1:
The patent merges multiple site surface detection capabilities into a single unified deep learning model. Instead of maintaining separate models for different site surfaces, the system processes user activity sequences from any site surface through one model, reducing complexity while maintaining detection reliability across diverse interfaces
Solution Approach 2:
The deep learning model is designed with universal functionality to detect abusive activities across multiple site surfaces and attack types. The model processes standardized user activity sequences and can identify various abusive patterns (scraping, hacking, spam) without requiring separate specialized models, thereby reducing overall system complexity
2Reliability
If hand-engineered features are used in traditional machine learning models, then detection is performed, but information loss occurs from feature engineering and human curation is required, reducing productivity
Solution Approach 1:
The deep learning model performs self-service by automatically learning relevant features from raw user activity sequences without requiring manual feature engineering. The model autonomously identifies patterns in user behavior data, eliminating the need for human curation and preserving complete information from the original activity sequences
Solution Approach 2:
The patent replaces the mechanical process of hand-engineered feature creation with an automated deep learning system. Instead of manually designing and curating features, the system uses neural networks to automatically extract meaningful patterns from raw data, substituting human effort with computational processing while maintaining or improving detection accuracy
3Reliability
If traditional machine learning models are deployed, then abusive activities are detected, but the models struggle to adapt against evolving adversarial attacks, reducing adaptability
Solution Approach 1:
The deep learning model incorporates dynamic adaptability to respond to evolving adversarial attacks. The system continuously learns from new attack patterns through ongoing training on updated data, allowing it to adapt its detection strategies dynamically rather than relying on static hand-engineered features that become obsolete against new attack vectors
Solution Approach 2:
The model adapts to adversarial attacks by changing its internal parameters through continuous learning and retraining. As new attack patterns emerge, the system updates its weight parameters and feature representations based on new training data, enabling it to maintain detection reliability against evolving threats without requiring manual feature redesign
Data Source
AI summary
In an example embodiment, a deep learning algorithm is introduced that operates directly on a raw sequence of user activity in an online network. This allows the system to scalably leverage more of the available signal hidden in the data and stop adversarial attacks more efficiently than other machine-learned models. More particularly, each specific request path is translated into a standardized token that indicates the type of the request (e.g., profile view, search, login, etc.). This eliminates the need for human curation of features. Then, the standardized request paths are standardized to integers based on the frequency of that request path across all users. This allows information about how common a given type of request is to be provided to the machine-learned model. The integer array is the activity sequence that is fed into the deep learning algorithm.