Learning towards Selective Data Augmentation for Dialogue Generation

Authors: Xiuying Chen, Mingzhe Li, Jiayi Zhang, Xiaoqiang Xia, Chen Wei, Jianwei Cui, Xin Gao, Xiangliang Zhang, Rui Yan

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on two publicly available datasets, i.e., Daily Dialog and Open Subtitles, show that our framework can improve the response generation performance with respect to various metrics.
Researcher Affiliation Collaboration Xiuying Chen1*, Mingzhe Li2*, Jiayi Zhang3, Xiaoqiang Xia3, Chen Wei3, Jianwei Cui3, Xin Gao1 , Xiangliang Zhang4, Rui Yan5 1Computational Bioscience Research Center, KAUST 2Ant Group 3 Xiaomi AI Lab 4 University of Notre Dame 5Gaoling School of Artificial Intelligence, Renmin University of China
Pseudocode No The paper describes its methodology using prose and mathematical equations but does not include any explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statement about releasing its source code or a link to a code repository for the described methodology.
Open Datasets Yes Following Cai et al. (2020), we conduct experiments on two English conversation datasets: (1) Daily Dialog (Li et al. 2017), a collection of real-world dialogues widely used in open-domain dialogue generation. ... (2) Open Subtitles (Lison and Tiedemann 2016), a group of human-human conversations converted from movie transcripts.
Dataset Splits Yes We split the Daily Dialog dataset to 54,889/6,005/5,700, and Open Subtitles to 64,000/8,000/8,000.
Hardware Specification Yes We implement our models in Tensor Flow on an NVIDIA GTX 1080 Ti GPU.
Software Dependencies No The paper mentions implementing models in 'Tensor Flow' but does not provide specific version numbers for TensorFlow or any other software dependencies.
Experiment Setup Yes We truncate the input dialog to 20 words, the minimum decoding step is 10, and the maximum step is 30. The default σ in Equation 1 is set to 0.6 except in the augmentation percentage analysis. The batch size is set to 16, and we limit the vocabulary size to 50K.