Meta-Complementing the Semantics of Short Texts in Neural Topic Models

Authors: Delvin Ce Zhang, Hady Lauw

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate the advantage of our framework.
Researcher Affiliation Academia Delvin Ce Zhang School of Computing and Information Systems Singapore Management University Singapore 178902 cezhang.2018@smu.edu.sg Hady W. Lauw School of Computing and Information Systems Singapore Management University Singapore 178902 hadywlauw@smu.edu.sg
Pseudocode No The paper mentions a learning algorithm in supplementary materials but does not include pseudocode or an algorithm block in the main text.
Open Source Code Yes We include the source code, datasets, and a readme file for reproducibility.
Open Datasets Yes Cora [24] is corpus of academic papers with citations as links. ... In addition, HEP-TH [21] is a corpus of Physics papers with their citations. Web [20] is a Web page hyperlink network where each page is a news article...
Dataset Splits Yes We split 80% documents for training (10% are for validation).
Hardware Specification Yes All the experiments were done on Linux server with a Tesla K80 GPU with 11441Mi B.
Software Dependencies No The paper does not provide specific version numbers for software dependencies beyond mentioning general tools like 'Glove embeddings'.
Experiment Setup Yes We set L = 2 convolutional layers. λgen = 2 and λreg = 0.05. Number of negative samples M = 5 and number of semantic clusters R = 5. L is the median length of the corpus. Local and global learning rates are α1 = 0.001 and α2 = 0.0005. We use 300D Glove embeddings.