Malicious Text Detection in Advertisements via Character Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for identifying malicious content in advertisements are inefficient and can be circumvented by advertisers using characters from different Unicode blocks, making it difficult to accurately detect malicious text.
Innovation Solution
An online system analyzes advertisement content by identifying characters and scoring words based on their types, using conditional probabilities to determine if the content is malicious, and comparing these scores to a threshold value to determine eligibility for presentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review methods are used to identify malicious content in advertisements, then detection accuracy is improved, but processing efficiency deteriorates
Solution Approach 1:
The patent introduces an intermediary automated analysis system that acts as a mediator between manual review and advertisement processing. This system uses character type analysis and scoring mechanisms to pre-filter advertisements, flagging only those with suspicious characteristics for manual review. This intermediary layer maintains high detection accuracy by preserving manual review for questionable cases while dramatically improving processing efficiency by automating the screening of clear cases.
2Productivity
If conventional automated systems are used to analyze advertisement content, then processing efficiency is improved, but detection accuracy deteriorates due to circumvention methods
Solution Approach 1:
The patent fundamentally changes the analysis parameter from semantic content interpretation to character type composition analysis. Instead of trying to understand advertisement meaning (which can be manipulated), the system analyzes the underlying character types (e.g., Latin vs. Cyrillic Unicode blocks). This parameter change makes the detection robust against circumvention methods that substitute visually similar characters, maintaining high detection accuracy while preserving automated processing efficiency.
Solution Approach 2:
The patent replaces the mechanical system of semantic analysis (reading and interpreting text meaning) with a more fundamental character-level analysis system. By substituting the approach from high-level semantic understanding to low-level character type identification and scoring, the system achieves both automated efficiency and resistance to character substitution attacks.
3Adaptability or versatility
If character substitution methods are used by advertisers, then ability to circumvent detection is improved, but detection difficulty increases
Solution Approach 1:
The patent inverts the detection approach by not trying to detect malicious intent directly (which becomes increasingly difficult as circumvention methods evolve), but rather by detecting anomalies in character type composition. Instead of asking 'is this text malicious?' the system asks 'does this text contain unusual character type patterns?' This inversion makes detection easier and more reliable, as character type analysis is objective and not susceptible to semantic manipulation.
Data Source
AI summary
An online system receives advertisement requests from one or more advertisers and determines whether an advertisement request includes malicious content before presenting content from the advertisement request to a user. To determine whether the advertisement request includes malicious content, the online system identifies text in the advertisement request, identifies words in the text, and identifies characters in each word. The online system identifies a most common type of character in each word and generates a score for each word based on its constituent characters. For example, a word's score is based on the combination of characters in the word, such as a conditional probability of a word including a type of character given that the word includes a given number of the most common type of character. The scores are analyzed to determine if text in the advertisement request includes malicious content.