Call Classification Using Feature Extraction and Probabilistic Hashing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for classifying calls lack accuracy and security, particularly in remote data analysis, where user personal data is vulnerable to unauthorized access and misuse.

Innovation Solution

A system and method that collect call data, extract significant features, generate classification models, and use generative review models to classify calls, incorporating probabilistic hashing for depersonalization and frequency analysis to identify suspicious calls, ensuring secure data handling and improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If remote data analysis is performed on user personal data, then data processing capability is improved, but data security deteriorates

Engineering Contradiction:
Improvedata processing capabilityVSAvoiddata security
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent extracts only the essential features needed for call classification (call duration, time of day, day of week, call type) from the complete call data, transmitting only these extracted features to remote servers for analysis while leaving sensitive personal data on the user's device

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates simplified copies of call data in the form of statistical features and patterns rather than transmitting complete raw data, allowing remote analysis to be performed on representative data samples that do not contain sensitive personal information

Inventive Principle:
Principle #26Copying

2Measurement precision

If comprehensive call data is collected for classification, then classification accuracy is improved, but data transmission volume increases

Engineering Contradiction:
Improveclassification accuracyVSAvoiddata transmission volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system extracts only the most significant features from comprehensive call data (duration, timing, call type indicators) that are sufficient for accurate classification while excluding redundant information, thereby reducing transmission volume without sacrificing classification accuracy

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different processing approaches to different data elements, performing local feature extraction and filtering on the user device before transmission, sending only the processed essential features rather than raw comprehensive data

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4383648A1System and method for classifying calls
Publication Date: 2024.06.12 AO KASPERSKY LAB
  • EP4383648A1 patent drawingFigure 1
  • EP4383648A1 patent drawingFigure 2
  • EP4383648A1 patent drawingFigure 3

AI summary

Disclosed herein are systems and methods for classifying calls on a remote device. In one example, an exemplary method comprises, collecting call data for each call, wherein each call is associated with a unique call identifier, extracting significant features from the collected call data, generating a call classification model based on the extracted significant features, wherein the call classification model comprises a set of rules based on which a predetermined call class is assigned to the call, extracting a text review from the collected call data, generating a generative review model based on the extracted text review, the generative review model used for correlating text reviews with a call class, and classifying the call for which the call data was collected based on the call classification model generated and the generative review model.