Multi-view non-negative tensor factorization as relation learning in healthcare data.

Area of Science:

• Data Science
• Machine Learning
• Information Retrieval

Background:

• Discovering patterns in co-occurrence data across multiple domains like healthcare and recommender systems is crucial.
• Integrating information from different data views is challenging due to high dimensionality and missing data.

Purpose of the Study:

• To develop a new method for learning semantic relations from multi-view co-occurrence data.
• To address the challenges of high dimensionality and missing data in complex datasets.

Main Methods:

• Proposed a novel paradigm using tensor factorization to jointly factorize multi-view tensors.
• Developed efficient optimization algorithms specifically designed for high-dimensional and incomplete data.
• Searched for a consistent underlying semantic space across different data views.

Main Results:

• The proposed tensor factorization approach effectively uncovers shared latent structures in multi-view data.
• The algorithms demonstrated effectiveness in handling both high-dimensional and missing data scenarios.
• Experimental results validated the potential and efficacy of the developed methods.

Conclusions:

• The new tensor factorization paradigm offers a powerful approach for semantic relation learning.
• The efficient algorithms provide a robust solution for complex, multi-view data analysis.
• This work advances pattern discovery in domains reliant on integrated, multi-view data analysis.