Evasive Listing Detection Using Spell Correction and Binary Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online transactional fraud, particularly price gouging during emergencies, evades detection due to continuous adaptation by fraudsters using evasive listing techniques such as misspellings and special characters, which existing heuristic filters and machine learning models struggle to effectively capture.
Innovation Solution
A system that employs machine learning techniques, including deep learning algorithms and binary classifiers, to detect and classify textual variations in item listings, using context-specific dictionaries and preprocessing components to identify and flag evasive listings, thereby distinguishing between legitimate and fraudulent listings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If heuristic filters and machine learning models are used to detect price gouging, then detection capability is improved, but fraudsters can evade detection by using evasive techniques such as misspellings and special characters
Solution Approach 1:
The system performs preliminary actions by generating multiple potential spell variations and corrections before the actual detection process. It pre-computes possible misspellings, special character substitutions, and alternative representations of keywords that fraudsters might use, thereby preparing the detection mechanism in advance to catch evasive listings.
Solution Approach 2:
The system applies parameter changes by transforming the detection approach from exact keyword matching to flexible pattern matching. It changes the parameters of text representation by considering multiple spell variations, character substitutions, and linguistic transformations, allowing the system to detect fraud across different textual representations.
2Reliability
If keyword filtering is used to identify high demand items, then price gouging detection is improved, but false positives increase due to legitimate listings containing similar keywords
Solution Approach 1:
The system applies local quality by making different parts of the detection process have different functions. It uses lenient matching for keyword identification to capture potential fraud cases, then applies stricter contextual analysis and multiple signal evaluation to filter out false positives, thereby achieving both high sensitivity and high precision in different stages.
Solution Approach 2:
The system performs partial action by not requiring all keywords to match exactly. It uses flexible matching where partial keyword matches, alternative spellings, and contextual similarities can trigger detection, reducing false negatives while maintaining false positive filtering through additional verification signals.
3Reliability
If the detection system is made more complex to capture evasive variations, then detection coverage is improved, but system complexity and computational cost increase
Solution Approach 1:
The system segments the detection process into distinct modular components: keyword identification, spell variation generation, contextual analysis, and final classification. Each module performs a specific function and can be independently optimized or adjusted, reducing overall system complexity while maintaining comprehensive detection coverage.
Data Source
AI summary
Systems and techniques are generally described for detecting evasive terms in item listings. In some examples, a list of keywords associated with a first category of specified items may be received. In some examples, text data associated with a first item may be received. A first plurality of tokens representing a corrected version of the text data may be generated using spell correction techniques. In some cases, first flag data indicating that the text data constitutes evasive text may be generated based at least in part on the generation of the first plurality of tokens. The first flag data and a numerical representation of the first plurality of tokens may be input into a binary classifier trained to generate label data indicating whether text data associated with a given item is evasive or non-evasive. The binary classifier may generate label data indicating that the text data associated with the first item is evasive.