Fusing Task-Oriented and Open-Domain Dialogues in Conversational Agents

Authors: Tom Young, Frank Xing, Vlad Pandelea, Jinjie Ni, Erik Cambria11622-11629

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate two baseline models on this task, including the classification-based two-stage models and the two-in-one fused models. We publicly release Fused Chat and the baselines to propel future work on inter-mode dialogue systems.
Researcher Affiliation Academia 1 School of Computer Science and Engineering, Nanyang Technological University, Singapore 2 School of Computing, National University of Singapore, Singapore
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes We publicly release Fused Chat and the baselines to propel future work on inter-mode dialogue systems. 1https://github.com/tomyoung903/Fused Chat
Open Datasets Yes Based on the popular TOD dataset Multi WOZ, we build a new dataset Fused Chat, by rewriting the existing TOD turns and adding new ODD turns. We publicly release Fused Chat and the baselines to propel future work on inter-mode dialogue systems. 1https://github.com/tomyoung903/Fused Chat
Dataset Splits Yes Table 2: Fused Chat is composed of ODD + TOD (prepending ODDs) instances and TOD + ODD (appending ODDs) instances. Partition Training Validation Testing Total Training 3670 4768 8438 Validation 500 500 1000 Testing 500 500 1000 Total 4670 5768 10438
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud computing instance types used for running the experiments.
Software Dependencies No The paper mentions using pre-trained models like Dialo GPT, GPT2, and BERT, and follows approaches like Neural Pipeline, but it does not specify exact version numbers for these models or any other software dependencies like programming languages or libraries.
Experiment Setup No The paper mentions that models are 'fine-tuned' but does not provide specific experimental setup details such as hyperparameter values (e.g., learning rate, batch size, number of epochs), optimizer settings, or other training configurations.