Stability of topic modeling via matrix factorization

Topic models can provide us with an insight into the underlying latent structure of a large corpus of documents. A range of methodshave been proposed in the literature, including probabilistic topic models and techniques based on matrix factorization. However, inboth cases, standard implementations rely on stochastic elements in their initialization phase, which can potentially lead to differentresults being generated on the same corpus when using the same parameter values. This corresponds to the concept of instabilitywhich has previously been studied in the context of k-means clustering. In many applications of topic modeling, this problem ofinstability is not considered and topic models are treated as being definitive, even though the results may change considerably if theinitialization process is altered. In this paper we demonstrate the inherent instability of popular topic modeling approaches, usinga number of new measures to assess stability. To address this issue in the context of matrix factorization for topic modeling, wepropose the use of ensemble learning strategies. Based on experiments performed on annotated text corpora, we show that a K-Foldensemble strategy, combining both ensembles and structured initialization, can significantly reduce instability, while simultaneouslyyielding more accurate topic models