Improving Knowledge-Aware Dialogue Generation via Knowledge Base Question Answering

Authors: Jian Wang, Junhao Liu, Wei Bi, Xiaojiang Liu, Kejing He, Ruifeng Xu, Min Yang9169-9176

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on two benchmark datasets demonstrate that our model has robust superiority over compared methods in generating informative and fluent dialogues.
Researcher Affiliation Collaboration Jian Wang,1 Junhao Liu,2 Wei Bi,3 Xiaojiang Liu,3 Kejing He,1 Ruifeng Xu,4 Min Yang2 1South China University of Technology, Guangzhou, China 2Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China 3Tencent AI Lab, Shenzhen, China 4Harbin Institute of Technology (Shenzhen), China
Pseudocode No The paper describes its methods using prose and mathematical equations but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/siat-nlp/Trans DG.
Open Datasets Yes We use Simple Questions (Bordes et al. 2015) dataset2 to train the KBQA model which consists of 75,910/10,845/21,687 instances for training/validation/testing, respectively. For dialogue generation, we use Reddit (Zhou et al. 2018) single-round dialogue dataset3, which contains 3,384,185 training pairs, 10,000 validation pairs and 20,000 test pairs.
Dataset Splits Yes We use Simple Questions (Bordes et al. 2015) dataset2 to train the KBQA model which consists of 75,910/10,845/21,687 instances for training/validation/testing, respectively. For dialogue generation, we use Reddit (Zhou et al. 2018) single-round dialogue dataset3, which contains 3,384,185 training pairs, 10,000 validation pairs and 20,000 test pairs.
Hardware Specification No The paper does not provide specific details on the hardware used, such as CPU/GPU models or memory specifications.
Software Dependencies No The paper mentions software components like 'Glove' and 'Adam optimizer' and 'Lucene' but does not provide specific version numbers for these or other key software dependencies.
Experiment Setup Yes For KBQA, ... We set the margin γ to be 0.5, and sample 20 negative samples for each gold answer. The model is trained using Adam (Kingma and Ba 2014) optimizer with an initial learning rate 0.001. The batch size is set to 128. For dialogue generation, ... The vocabulary size is set to 30,000. The encoder and decoder have 2-layer GRUs with 512 hidden units for each layer. The dropout rate is set to 0.2. The number of candidate responses k is set to 3, ... We train the model using Adam optimizer with an initial learning rate of 0.0005, and the batch size is set to 100.