HTML Tokenization With Large Language Models for Granular Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing web filtering methods face challenges in blocking specific content types without affecting access to other essential resources, and allowlist filtering requires constant updates due to evolving online content, which is time-consuming and inconvenient.
Innovation Solution
Utilize HTML tokenization and a large language model to analyze webpage content, segmenting HTML into tokens, and determine content descriptions for granular filtering and categorization, enabling real-time and proactive content management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If denylist filtering is used to block specific content, then access to harmful content is restricted, but access to essential resources on the same domain is also blocked
Solution Approach 1:
The patent segments the webpage into multiple HTML tokens representing different content elements (e.g., videos, images, text sections). The web filter then selectively blocks or allows individual tokens rather than entire domains, enabling granular control that prevents harmful content while preserving access to essential resources on the same page.
2Adaptability or versatility
If allowlist filtering is used to control content access, then access to approved content is ensured, but constant updates are required to accommodate new content
Solution Approach 1:
The patent performs preliminary analysis of webpage content by tokenizing HTML and using AI models to categorize content before the user accesses it. This proactive classification enables the system to automatically make filtering decisions without requiring manual allowlist updates, as new content is evaluated and categorized in real-time.
Solution Approach 2:
The web filtering system uses AI-based content analysis to automatically categorize and filter content without human intervention. The system self-manages the filtering process by analyzing webpage tokens, determining their categories, and applying appropriate filters, eliminating the need for manual allowlist maintenance.
3Measurement precision
If granular content filtering is implemented, then filtering accuracy is improved, but system complexity increases
Solution Approach 1:
The patent introduces HTML tokenization as an intermediary step between webpage retrieval and content filtering. By breaking down webpages into standardized tokens and using AI models to categorize them, the system achieves granular filtering accuracy while managing complexity through a structured, modular approach to content analysis.
Data Source
AI summary
A computer program product and method include operations including accessing HTML code from one or more webpages, tokenizing the HTML code to form one or more HTML tokens, submitting each HTML token to a large language model, and obtaining a token content description for each HTML token from the large language model. The operations further include determining, for each of the one or more webpages, whether to rendered the HTML code on the web browser based on the token content descriptions of the HTML tokens formed for the HTML code from the one or more webpages.