Dirichlet belief networks for topic structure learning

Authors: He Zhao, Lan Du, Wray Buntine, Mingyuan Zhou

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on text corpora demonstrate the advantages of the proposed model.
Researcher Affiliation Academia 1Faculty of Information Technology, Monash University, Australia 2Mc Combs School of Business, The University of Texas at Austin, USA
Pseudocode Yes Omitted details of inference as well as the overall algorithm are given in the supplementary materials.
Open Source Code Yes Code available at https://github.com/ethanhezhao/Dir BN
Open Datasets Yes The experiments were conducted on three real-world datasets, detailed as follows: 1) Web Snippets (WS), containing 12,237 web search snippets labelled with 8 categories... 2) Tag My News (TMN), consisting of 32,597 RSS news labelled with 7 categories... 3) Twitter, extracted in 2011 and 2012 microblog tracks at Text REtrieval Conference (TREC)5. It has 11,109 tweets in total... Footnote 5: http://trec.nist.gov/data/microblog.html
Dataset Splits Yes To compute perplexity, we randomly selected 80% of the documents in each dataset to train the models and 20% for testing.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No The paper mentions “Mallet” but does not provide specific version numbers for any software, libraries, or programming languages used for implementation or experimentation.
Experiment Setup Yes For all the models, we ran 3,000 MCMC iterations with 1,500 burnin. For Dir BN, we set a0 = b0 = g0 = h0 = 1.0 and e0 = f0 = 0.01... For all the models, the number of topics in each layer of Dir BN was set to 100, i.e., KT = = K1 = 100.