Unity in Diversity: Learning Distributed Heterogeneous Sentence Representation for Extractive Summarization

Authors: Abhishek Singh, Manish Gupta, Vasudeva Varma

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the DUC benchmark datasets (DUC-2001, DUC-2002 and DUC2004) indicate that our model shows significant performance gain of around 1.5-2 points in terms of ROUGE score compared with the state-of-the-art baselines.
Researcher Affiliation Collaboration Abhishek Kumar Singh, Manish Gupta, Vasudeva Varma IIIT Hyderabad, India abhishek.singh@research.iiit.ac.in, manish.gupta@iiit.ac.in, vv@iiit.ac.in. The author is also a Principal Applied Researcher at Microsoft.
Pseudocode No The paper describes its methods using prose, equations, and diagrams, but does not include any pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement or link for the open-source code of the described methodology.
Open Datasets Yes Initial training of our model is done on the Daily Mail corpus, used for the task of single document summarization by (Cheng and Lapata 2016)... We experiment on DUC 2001-2004 datasets... The full DUC data set can be availed by request at http://duc.nist.gov/data.html. The SICK dataset which contains 9927 sentence pairs with a 5,000/4,927 training/test split (Marelli et al. 2014) was used for training the Siam CSTI net.
Dataset Splits Yes Overall, we have 193986 training documents, 12147 validation documents and 10350 test documents in the corpus... DUC 2003 data is used as development set and we perform a 3-fold cross-validation on DUC 2001, 2002 and 2004 datasets with two years of data as training set and one year of data as the test set.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory, or cloud instance types).
Software Dependencies No The paper mentions tools like 'Adam' and 'LIBLINEAR' but does not specify their version numbers or other software dependencies with version information required for reproducibility.
Experiment Setup Yes The size of the hidden units of BLSTM was set to 150. After tuning on the validation set, we fix the dimension m of the latent features from convolutional encoder as 125 and window size k = 5 for HNet system. We use Adam (Kingma and Ba 2014) as the optimizer with mini batches of size 35. Learning rates are set to {0.009, 0.0009}. For our network, we use regularization dropout of {0.2, 0.5}.