Stochastic Multi-Modal Recommendation System for Uncertainty Quantification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems face limitations in reconciling popularity bias in training data, failing to provide quantitative uncertainty measures, and accurately representing user and content relationships, especially in polysemy and probabilistic terms.
Innovation Solution
A novel probabilistic approach representing entities as multi-dimensional statistical distributions, such as mixtures of Gaussians, to quantify uncertainty and model user and content interactions, incorporating metadata and activity profiles for improved recommendation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If collaborative filtering with gradient-based learning algorithms is used, then recommendation capability is improved, but inability to reconcile popularity bias occurs
Solution Approach 1:
The patent transforms the recommendation problem from deterministic to probabilistic by changing the parameter representation from fixed values to statistical distributions. Entities are represented as multi-dimensional statistical distributions (e.g., mixtures of Gaussians) rather than fixed vectors, enabling the system to model uncertainty and reconcile popularity bias through probabilistic matching while maintaining recommendation capability.
2Speed
If traditional recommendation systems are used, then processing speed is maintained, but quantitative uncertainty measures are not provided
Solution Approach 1:
The patent adds a new dimension to entity representation by introducing statistical distribution parameters (mean, variance, covariance) alongside traditional embedding vectors. This multi-dimensional representation includes both the central tendency (mean vector) and uncertainty characteristics (covariance matrix), providing quantitative uncertainty measures while maintaining compatibility with existing processing frameworks.
3Device complexity
If simple entity representations are used, then computational complexity is reduced, but multiple patterns of consumption are not revealed
Solution Approach 1:
The patent segments entity representation into distinct statistical components (mean vector, covariance matrix, distribution type) that can be independently computed and stored. This segmentation allows the system to capture multiple patterns of consumption through different distribution characteristics while managing computational complexity through modular processing of each component.
Data Source
AI summary
A system includes a computing platform including processing hardware and a memory storing software code including a trained machine learning (ML) model. The processing hardware executes the software code to receive entity specific data over a network from a user device, identify mapping parameters of the entity specific data, and map, using the trained ML model and the mapping parameters, the entity specific data to a statistical distribution in a multi-dimensional representation space. The software code further compares, using the trained ML model, the mapped statistical distribution to each of one or more predetermined statistical distributions in the multi-dimensional representation space, predicts, to using the trained ML model and the comparison, a matching probability for each of the one or more predetermined statistical distributions relative to the mapped statistical distribution. generates a similarity set based on the prediction, and outputs the similarity set to the user device over the network.