Topic-Aware Multi-turn Dialogue Modeling
Authors: Yi Xu, Hai Zhao, Zhuosheng Zhang14176-14184
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on three public datasets show TADAM can outperform the state-of-the-art method, especially by 3.3% on E-commerce dataset that has an obvious topic shift. |
| Researcher Affiliation | Academia | 1 Department of Computer Science and Engineering, Shanghai Jiao Tong University 2 Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering, Shanghai Jiao Tong University, Shanghai, China 3 Mo E Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University |
| Pseudocode | Yes | Algorithm 1 Topic-aware Segmentation Algorithm |
| Open Source Code | Yes | Both datasets and code are available at https://github.com/xyease/TADAM |
| Open Datasets | Yes | For Chinese, we annotate a dataset including 505 phone records of customer service on banking consultation. For English, we build dataset including 711 dialogues by joining dialogues from existing multi-turn dialogue datasets: Multi WOZ Corpus2 (Budzianowski et al. 2018) and Stanford Dialog Dataset (Eric et al. 2017). (Footnote 2: https://doi.org/10.17863/CAM.41572) Additionally, Ubuntu Corpus (Lowe et al. 2015), Douban Corpus (Wu et al. 2017), and E-commerce Corpus (Zhang et al. 2018) were used. |
| Dataset Splits | No | The paper mentions training on datasets and evaluating on test sets but does not explicitly provide details about a separate validation dataset split with percentages or counts. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions using "pre-trained BERT (Devlin et al. 2018)" and links to "https://github.com/huggingface/transformers", implying the use of the HuggingFace Transformers library, but does not provide specific version numbers for Python, PyTorch, TensorFlow, or the Transformers library itself. |
| Experiment Setup | Yes | For topic-aware segmentation: In both datasets, we set range R = 8, jump step K = 2 (value of 1 will lead to fragmentation), window size d = 2 and threshold θcost=0.6. For response selection: We apply topic-aware segmentation algorithm to Ubuntu, Douban and E-commerce with range R = 2, 2, 6... the max input sequence length is set to 350... max number of segments is 10. We set the learning rate as 2e-5 using Bert Adam with a warmup proportion of 10%. Our model is trained with batch size of {20,32,20} and epoch of {3,3,4}... the α of word-level weights is set to 0.5. |