Options
Belford, Mark
Preferred name
Belford, Mark
Official Name
Belford, Mark
Research Output
Now showing 1 - 2 of 2
- PublicationStability of topic modeling via matrix factorizationTopic 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
Scopus© Citations 58 348 - PublicationEnsemble Topic Modeling using Weighted Term Co-associationsTopic modeling is a popular unsupervised technique that is used to discover the latent thematic structure in text corpora. The evaluation of topic models typically involves measuring the semantic coherence of the terms describing each topic, where a single value is used to summarize the quality of an overall model. However, this can create difficulties when one seeks to interpret the strengths and weaknesses of a given topic model. With this in mind, we propose a new ensemble topic modeling approach that incorporates both stability information, in the form of term co-associations, and semantic similarity information, as derived from a word embedding constructed on a background corpus. Our evaluations show that this approach can simultaneously yield higher quality models when considering the produced topic descriptors and document-topic assignments, while also facilitating the comparison and evaluation of solutions through the visualization of the discovered topical structure, the ordering of the topic descriptors, and the ranking of term pairs which appear in topic descriptors.
Scopus© Citations 9 138