| This paper investigates the clustering of text sequences pro- duced during the writing process, referred to as writing bursts. These units exhibit strong variability in their linguistic, temporal, and behav- ioral properties, which makes their categorization particularly challeng- ing. We propose an unsupervised framework that combines hybrid text embeddings with Multi-View Nonnegative Matrix Factorization (MV- NMF) in order to jointly exploit complementary feature spaces. More specifically, our approach integrates linguistic, temporal, event-based, and process-related views to better capture the multidimensional nature of writing bursts. Experiments conducted on real-time writing-process data show that the proposed framework yields meaningful burst group- ings, supported by expert analysis and Silhouette-based validation. These results highlight the relevance of multi-view unsupervised learning for modeling writing dynamics and open promising perspectives for the anal- ysis of text production processes. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.