Personalized Dialogue Generation with Persona-Adaptive Attention
Authors: Qiushi Huang, Yu Zhang, Tom Ko, Xubo Liu, Bo Wu, Wenwu Wang, H Tang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the superiority of the proposed PAA framework compared to the strong baselines in both automatic and human evaluation. |
| Researcher Affiliation | Collaboration | Qiushi Huang1,2, Yu Zhang2*, Tom Ko3, Xubo Liu1, Bo Wu4, H Tang1* 1 University of Surrey 2 Southern University of Science and Technology 3 Byte Dance AI Lab 4 MIT-IBM Watson AI Lab |
| Pseudocode | No | The paper describes the model architecture and mathematical formulations but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is publicly available for reproducing the results.4 |
| Open Datasets | Yes | We present the experimental results for PAA on the Conv AI2 dataset (Dinan et al. 2019) using both automatic and human evaluations. |
| Dataset Splits | Yes | Conv AI2 is a crowd-sourced dialogue dataset consisting of 8939/1000 multi-turn dialogues conditioned on 1155/100 persona for the train/dev splits. |
| Hardware Specification | Yes | We trained on one Nvidia RTX8000 with a batch size of 32. |
| Software Dependencies | No | The paper mentions 'sacrebleu' and 'GPT2-SMALL model' but does not provide specific version numbers for the software dependencies used in the experiments. |
| Experiment Setup | Yes | We employ the Adam optimizer (Kingma and Ba 2015) with a learning rate of 8 10 6. The weight decay, β1, and β2 for the Adam optimizer are set to 0, 0.9, and 0.999, respectively. The training lasted for 30 epochs with 131,438 samples per epoch. ... We applied the gradient clip with the norm value of 0.1. |