Bot Detection Using Transient Tasks to Filter Fraudulent Requests
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Bot abuse in online systems leads to wasteful processing resources, system delays, and diminished user experiences due to fraudulent automated requests that are difficult to differentiate from genuine user requests.
Innovation Solution
A task management and detection system that utilizes transient tasks (phantom blocks) and various detection techniques to identify bot usage, including frequency-based, context-based, and user-specific methods, and takes remedial actions such as denying requests or modifying task properties to deter bot usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated web requests are processed without differentiation, then system throughput is maintained, but processing resources are wasted on bot-generated requests
Solution Approach 1:
The system performs preliminary bot detection analysis on incoming web requests before they are fully processed. Detection signals are generated based on request characteristics, and high-confidence bot requests are identified and filtered out early in the processing pipeline, preventing wasteful processing of malicious requests while maintaining throughput for legitimate traffic
Solution Approach 2:
A bot detection system acts as an intermediary layer between incoming web requests and the target application. The detection system analyzes requests, generates detection signals, and determines whether requests are bot-generated before they reach the application, thereby protecting processing resources without blocking legitimate requests
2Reliability
If bot detection analysis is performed on all requests, then bot usage is identified, but system response time increases
Solution Approach 1:
The system applies different levels of detection analysis to different requests based on their characteristics. High-confidence bot requests are identified using streamlined analysis, while legitimate requests pass through with minimal processing. The detection depth is adjusted locally for each request based on detection signal strength and confidence levels
Solution Approach 2:
The system performs partial bot detection analysis on requests that fall into intermediate confidence categories, rather than applying full analysis to all requests. This selective approach focuses computational resources on suspicious requests while allowing clear legitimate requests to pass through with minimal processing delay
3Measurement precision
If detection signals are generated for all requests, then bot usage can be detected, but processing overhead increases
Solution Approach 1:
The system combines multiple detection signals into a unified bot determination. Individual detection signals from various analysis methods are merged and evaluated together to reach a final bot classification, reducing the need for separate processing pipelines and minimizing overall processing overhead
Solution Approach 2:
The system uses request characteristics and patterns as proxies for deep analysis in many cases. By copying and analyzing surface-level request features rather than performing exhaustive inspection of all requests, the system achieves adequate detection precision with reduced processing overhead
Data Source
AI summary
Techniques are provided herein for detecting bot activity. Tasks associated with one or more delivery sub-tasks can be provided to various user devices. The task requests received from the user devices can be processed according to a variety of factors to determine a likelihood that the task requests were initiated by a bot. In some embodiments, transient tasks (e.g., task that are deleted within a relatively short period of time and for which a user device cannot be assigned) may be utilized at any suitable time. User devices that request assignment of such transient tasks may be identified as using a bot, or at least the probability identified for those user devices can be increased, indicating a heightened likelihood that the user device is using a bot. A number of remedial actions can be executed when the likelihood that the user device is using a bot exceeds a threshold.