Mobile Event Classification via Personalized ML Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mobile device systems fail to effectively classify incoming events such as calls, SMS, and emails based on user preferences, often categorizing significant events as spam, and do not account for personal rules or media types beyond voice calls.
Innovation Solution
A method and system that intercepts incoming events, analyzes their content, compares them to a repository of classified events using machine learning models like naive Bayesian or SVM, and determines a rating based on user preferences to classify events as undesirable if the rating falls below a threshold, employing techniques like gradient descent optimization and cosine distance calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a database of spam numbers is used to filter incoming calls, then the majority of undesirable incoming calls are blocked, but significant events for the user are incorrectly categorized as spam
Solution Approach 1:
The system transitions from a generic spam filtering approach to a personalized classification system that adapts to each user's specific preferences and event patterns. By analyzing individual user behavior patterns and assigning personalized weights to different event types, the system achieves local quality optimization where filtering accuracy is tailored to each user's needs, preventing misclassification of user-specific significant events while maintaining spam blocking effectiveness
Solution Approach 2:
The classification system dynamically adjusts its filtering criteria based on real-time user interactions and learned preferences. The machine learning model continuously updates its understanding of what events are significant to each user, allowing the system to adapt its classification rules dynamically. This dynamic adjustment enables the system to maintain high accuracy in identifying spam while preserving the ability to recognize user-specific important events that may vary over time
2Adaptability or versatility
If a general spam filtering procedure is applied, then voice calls are protected from nuisance communications, but other media types like SMS, emails and HTML blocks are not covered
Solution Approach 1:
The system implements a universal event classification framework that handles multiple media types (voice calls, SMS, emails, HTML blocks) through a single integrated machine learning model. This multi-functional approach allows the same classification mechanism to process diverse event formats, expanding coverage without requiring separate filtering systems for each media type, thereby achieving versatility while managing complexity through unification
3Adaptability or versatility
If user preferences are not considered, then classification rules remain simple, but personal rules of classification cannot be defined
Solution Approach 1:
The system employs self-service learning where the machine learning model automatically captures and incorporates user preferences without requiring explicit configuration or manual rule definition. By analyzing user interactions with events and learning from patterns, the system autonomously builds personalized classification rules, eliminating the need for complex manual setup while achieving high adaptability to individual user needs
Data Source
AI summary
Disclosed herein are methods and systems for classifying incoming events by user's mobile device based on user preferences. In one aspect, an exemplary method comprises: intercepting an incoming event received by a mobile device, analyzing content of the intercepted event to determine one or more attributes of the intercepted event, comparing the intercepted event to a plurality of previously collected and classified events, stored in an event repository, based on the one or more determined attributes to identify one or more similar events, determining a rating value of the one or more similar events based on a matrix of user preferences, wherein the rating value indicates probability that the corresponding event belongs to a particular class of events, and classifying the intercepted event as undesirable on the mobile device if the rating value of the one or more similar events is less than a predetermined threshold value.