Server Video Call Matching Using Face Analysis and ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video call matching services do not effectively improve match satisfaction between users, as they rely solely on profile information such as academic background, region, and interest without considering real-time interaction satisfaction during video calls.
Innovation Solution
A server and operating method that utilize machine learning models trained with face information and satisfaction factors from video call sessions to estimate match satisfaction for standby users, selecting the next video call counterpart based on calculated match satisfaction values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If matching is based on profile information (academic background, region, age, interest), then the matching service can be implemented with simple algorithms, but the match satisfaction during video calls is not effectively improved
Solution Approach 1:
The system collects satisfaction feedback from users during and after video call sessions, then uses this feedback to train machine learning models that improve future matching decisions. The satisfaction data is fed back into the matching algorithm to continuously refine counterpart selection based on actual interaction quality rather than just profile compatibility.
Solution Approach 2:
The matching system transitions from using only static profile parameters (academic background, region, age, interest) to incorporating dynamic satisfaction parameters derived from machine learning model predictions. This parameter expansion allows the system to evaluate potential counterparts based on predicted interaction satisfaction rather than merely demographic or interest-based compatibility.
2Reliability
If machine learning models are trained with face information and satisfaction factors, then match satisfaction can be improved, but the system complexity and data processing requirements increase
Solution Approach 1:
Face information is extracted and stored in advance during user registration or profile setup, rather than processing it in real-time during matching. The machine learning model is trained beforehand with historical satisfaction data, so that during actual matching operations, the system only needs to query pre-computed satisfaction predictions rather than performing complex real-time analysis.
Solution Approach 2:
The machine learning model acts as an intermediary component that bridges the gap between raw face information/satisfaction factors and the matching decision. Instead of directly complexifying the matching algorithm itself, the system introduces a trained ML model that handles the complex pattern recognition, allowing the core matching system to remain relatively simple while still achieving improved match satisfaction.
Data Source
AI summary
A server includes a session management device establishing a video call session between a first electronic apparatus of a first user and a second electronic apparatus of a second user, a communication device receiving match satisfaction calculated from the first electronic apparatus through the video call session, an image analysis device obtaining first feature point distribution information from a face image of the first user and obtaining second feature point distribution information from a face image of the second user, a processor training a machine learning model by using the first feature point distribution information, the second feature point distribution information, and the match satisfaction and estimating the match satisfaction of the first user for each of standby users by using the trained machine learning model when a video call connection between the first electronic apparatus and the second electronic apparatus is terminated, and a user management device selecting a third user, who becomes a next video call counterpart of the first user, from among the standby users by using the estimated match satisfaction.