Tracking Time Evolution of Collective Attention Clusters in Twitter: Time Evolving Nonnegative Matrix Factorisation.

Area of Science:

• Computational Social Science
• Data Mining
• Machine Learning

Background:

• Micro-blogging platforms like Twitter offer rich data for analyzing user behavior.
• Distinguishing individual users and tracking collective attention shifts within groups is challenging.
• Existing methods may lose temporal connections in sequential data analysis.

Purpose of the Study:

• To develop a novel method for tracking temporal dynamics of user behavior and collective attention on Twitter.
• To address the limitations of traditional Nonnegative Matrix Factorisation (NMF) in handling time-sequential data.
• To propose Time Evolving Nonnegative Matrix Factorisation (TENMF) for analyzing time-series matrices.

Main Methods:

• Formulating the problem as matrix decomposition of time-sequential data.
• Developing and applying Time Evolving Nonnegative Matrix Factorisation (TENMF).
• Representing users and words within time intervals as sequential matrices.

Main Results:

• TENMF successfully decomposes time-sequential matrices and tracks connections between them.
• The proposed TENMF method demonstrates good performance on artificial datasets.
• Experiments with real Twitter data provide valuable insights into user behavior and attention dynamics.

Conclusions:

• TENMF offers a robust approach for analyzing temporal patterns in micro-blogging data.
• The method preserves crucial time-sequential connections lost in standard NMF.
• This research enhances our ability to understand evolving collective attention on social media platforms.