Two-Level ML Page Categorization for Dynamic Websites

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for managing web pages in large-scale dynamic websites lack the ability to categorize at a granular level in real-time and adapt to changes in content and structure, limiting their effectiveness in managing complex and dynamic websites.

Innovation Solution

A two-level categorization process using machine learning models, where a first ML model categorizes pages into strands based on first parameters and a second ML model classifies them into sub-strands based on second parameters, incorporating pre-processing and reinforcement learning for improved accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional categorization methods are used, then the system is simple to implement, but it cannot categorize at a granular level in real-time and adapt to changes

Engineering Contradiction:
Improvecategorization granularityVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the categorization task into two distinct levels: strand-level categorization (first ML model) and sub-strand-level classification (second ML model). This segmentation enables granular analysis while managing complexity through modular architecture, where each model handles a specific layer of classification hierarchy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a temporal dimension by processing pages in real-time as they are generated, rather than using retrospective batch processing. This real-time dimensionality allows the system to adapt to dynamic changes in web page content and structure immediately, enhancing both granularity and responsiveness.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Productivity

If retrospective categorization is used, then the processing is simpler, but real-time computation and adaptation to changes is restricted

Engineering Contradiction:
Improvereal-time processing capabilityVSAvoidprocessing time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system performs preliminary extraction of features and parameters from web pages as they are generated. By preparing and storing these features in advance (first parameters for strands, second parameters for sub-strands), the system enables rapid real-time classification without requiring complex computation at the moment of need, thus reducing processing latency.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements continuous real-time processing where pages are constantly monitored, categorized, and classified as they are generated. This continuous operation ensures that the system always has up-to-date information about current web pages, enabling immediate adaptation to changes without interruption or delay.

Inventive Principle:
Principle #20Continuity of useful action

3Measurement precision

If generic level categorization is used, then the system is easier to manage, but granular pattern-based level aggregation is not achieved

Engineering Contradiction:
Improvepattern-based classification accuracyVSAvoidclassification system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the classification task into two hierarchical levels with distinct ML models, each specialized for a specific level. The first model handles broad strand-level categorization using first parameters, while the second model handles detailed sub-strand-level classification using second parameters and associated weights, achieving granular pattern-based accuracy through structured division of labor.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different parameter sets and weighting schemes locally at different levels of the classification hierarchy. Each sub-strand has its own associated weights for the second parameters, allowing the system to optimize classification accuracy for specific patterns while maintaining overall system manageability through localized customization.

Inventive Principle:
Principle #3Local quality

4Productivity

If manual categorization is used, then the system requires less computational resources, but it cannot scale to large-scale dynamic websites

Engineering Contradiction:
Improvepage management throughputVSAvoidcomputational resource consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The system automatically extracts features, categorizes pages into strands, and classifies them into sub-strands using ML models without requiring manual intervention. This self-service capability enables the system to process large numbers of pages at high throughput while scaling computational resources dynamically based on the actual workload and complexity of the website being managed.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4625210A1Page management method and system thereof
Publication Date: 2025.10.01 INFOSYS LTD
  • EP4625210A1 patent drawingFigure 1
  • EP4625210A1 patent drawingFigure 2
  • EP4625210A1 patent drawingFigure 3~4

AI summary

This disclosure relates to a page management method and system thereof. The method includes categorizing (302), using a first ML model (108), a set of pages into at least one of a plurality of strands, based on a set of first parameters. Further, the method includes classifying (304) each of the set of pages, using a second ML model (110), into one of a plurality of substrands, based on an associated set of second parameters and the associated strand from the plurality of strands. Further, the method includes determining (306), for each of the set of pages, a score, based on the associated sub-strand, the weight assigned to each of the subset of second parameters of the associated sub-strand, and values of each of the subset of second parameters of the associated sub-strand. Further, the method includes performing (308) an action on at least one of the set of pages based on the determined score.