Partial membership latent Dirichlet allocation

Chen, Chao; Zare, Alina; Cobb, J. Tory

doi:10.32469/10355/56468

Cited by 17 publications

(18 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Latent Dirichlet Allocation (LDA) described in [4,5] is a statistical topic model that discovers the hidden theme or topics from a collection of documents. Assuming an imaginary generative process of constructing these documents, LDA then tries to backtrack from these documents to infer a set of topics that are likely to have generated the collection [6]. While being a powerful tool at discovering the thematic structure of text, LDA makes certain assumptions, which make it inappropriate when considering the semantics of a language.…”

Section: Related Workmentioning

confidence: 99%

On discovering most frequent research trends in a scientific discipline using a text mining technique

Lakhanpal

Gupta

Agrawal

2014

Proceedings of the 2014 ACM Southeast Regional Conference

View full text Add to dashboard Cite

Science encompasses many sub-domains such as Computer Science, Physics, M edicine, etc. Each domain such as Computer Science can have many sub-disciplines like Networking, Data M ining, Parallel Processing, etc. M any different techniques have been used in these disciplines to solve open problems and improve existing solutions. Innovations in techniques call for researching prevalent solutions and active work being done in that field. It is thus highly desirable, yet a challenging task to automate the process of identifying current trend of research in any subdiscipline. Automation techniques will allow for faster exploration of methodology and ideas, especially among young researchers or the ones switching to related disciplines, enabling further improvisation and invention. This paper presents a technique for mining the titles and abstracts of research papers to aid in achieving this task. The key idea behind this is that the title and abstract of a research paper encompass within their component words, the core technique, methodology, or aim of that paper. We thus present preliminary ideas of a text mining technique that can efficiently identify trending research topics in a discipline. Our initial experiments exhibit encouraging results.

show abstract

Section: Related Workmentioning

confidence: 99%

On discovering most frequent research trends in a scientific discipline using a text mining technique

Lakhanpal

Gupta

Agrawal

2014

Proceedings of the 2014 ACM Southeast Regional Conference

View full text Add to dashboard Cite

show abstract

“…The appropriate number of topics for our dataset was determined by comparing the intra-topic similarity with inter-topic dissimilarity. Analyses specifying five, ten, fifteen, and twenty topic solutions were run, and the optimum number of topics was determined to be twenty as this maximized the difference between topics (Chen and Wang, 2018). Words in topics were generated by determining the mix between the probability and relevance of words belonging to topics (Sievert and Shirley, 2014).…”

Section: Citation Analysismentioning

confidence: 99%

Information flows and topic modeling in corporate governance

Kushkowski

Shrader

Anderson

et al. 2020

View full text Add to dashboard Cite

PurposeMultiple disciplines such as finance, management and economics have contributed to governance research over time. However, the full intellectual structure of the governance “field” including the exchange of knowledge across disciplines and the large variety of governance topics remains to be uncovered. To appreciate the breadth of corporate governance research, it is necessary to understand the disciplinary sources from which the research stems. This manuscript focuses on the interdisciplinary underpinnings of corporate governance research.Design/methodology/approachThis paper employs bibliometric analysis to trace the evolution of corporate governance using articles included in the ISI Web of Science database between 1990 and 2015. Journals included in these categories encompass a full range of business disciplines and provide evidence of the multi-disciplinary nature of corporate governance. It also uncovers the topics treated by disciplines under the governance umbrella using a machine learning method called latent Dirichtlet allocation (LDA).FindingsCorporate governance research deals with a number of strategy-related topics. Unlike strategy topics that reside in a single discipline, corporate governance crosses disciplinary boundaries and includes contributions from accounting, finance, economics, law and management. Our analysis shows that over 80% of corporate governance articles come from outside the field of management. Our LDA solution indicates that the major topics in governance research include corporate governance theory, control of family firms, executive compensation and audit committees.Originality/valueThe results illustrate that corporate governance is far more interdisciplinary than previously thought. This is an important insight for corporate governance academics and may lead to collaborative research. More importantly, this research illustrates the usefulness of LDA for investigating interdisciplinary fields. This method is easily transferable to other interdisciplinary fields and it provides a powerful alternative to existing bibliometric methods. We suggest a number of topic areas within library and information science where this method may be applied, including collection development, support for interdisciplinary faculty and basic research into emerging interdisciplinary areas.

show abstract

“…Partial Membership Latent Dirichlet Allocation (PM-LDA) [26,27] is an extension of Latent Dirichlet Allocation topic modeling [28] that allows words to have partial membership across multiple topics. The use of partial memberships allows for topic modeling given data sets in which crisp topic assignments (as done by LDA) is insufficient since data points (or words) may straddle multiple topics simultaneously.…”

Section: Partial Membership Latent Dirichlet Allocationmentioning

confidence: 99%

Hyperspectral unmixing with endmember variability using Partial Membership Latent Dirichlet Allocation

Zou

Zare

2017

2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

A semi-supervised Partial Membership Latent Dirichlet Allocation approach is developed for hyperspectral unmixing and endmember estimation while accounting for spectral variability and spatial information. Partial Membership Latent Dirichlet Allocation is an effective approach for spectral unmixing while representing spectral variability and leveraging spatial information. In this work, we extend Partial Membership Latent Dirichlet Allocation to incorporate any available (imprecise) label information to help guide unmixing. Experimental results on two hyperspectral datasets show that the proposed semi-supervised PM-LDA can yield improved hyperspectral unmixing and endmember estimation results.

show abstract

Partial membership latent Dirichlet allocation

Cited by 17 publications

References 19 publications

On discovering most frequent research trends in a scientific discipline using a text mining technique

On discovering most frequent research trends in a scientific discipline using a text mining technique

Information flows and topic modeling in corporate governance

Hyperspectral unmixing with endmember variability using Partial Membership Latent Dirichlet Allocation

Contact Info

Product

Resources

About