ECR-Chain: Advancing Generative Language Models to Better Emotion-Cause Reasoners through Reasoning Chains

Authors: Zhaopei Huang, Jinming Zhao, Qin Jin

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on the RECCON-DD dataset [Poria et al., 2021]. This dataset supplements causal utterance annotations for each non-neutral utterance in the conversations of the Daily Dialog dataset [Li et al., 2017]. Extensive experimental results over various settings demonstrate the effectiveness of our method for predicting emotion-cause utterances and performing explainable emotion-cause reasoning.
Researcher Affiliation Academia Zhaopei Huang1 , Jinming Zhao2 , Qin Jin1 1Renmin University of China 2Independent Researcher {huangzhaopei, qjin}@ruc.edu.cn, zhaojinming1@gmail.com
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks. It uses diagrams and textual descriptions to explain procedures.
Open Source Code Yes Our code, data and more details are at https://github.com/hzp3517/ECR-Chain.
Open Datasets Yes We conduct experiments on the RECCON-DD dataset [Poria et al., 2021]. This dataset supplements causal utterance annotations for each non-neutral utterance in the conversations of the Daily Dialog dataset [Li et al., 2017].
Dataset Splits Yes Table 1: Dataset statistics. We consider each target utterance as a sample. A conversation may contain several target utterances, forming several samples. Statistics Train Valid Test Samples 4,562 200 1,099
Hardware Specification No The paper mentions using 'Chat GPT (gpt-3.5-turbo-0613)' and 'Vicuna-7B-v1.3' but does not provide specific hardware details such as GPU models, CPU types, or memory used for running the experiments.
Software Dependencies Yes We utilize Chat GPT (gpt-3.5-turbo-0613) as our LLM... For the smaller language model, we opt for Vicuna-7B-v1.3... applied Lo RA fine-tuning [Hu et al., 2021]
Experiment Setup Yes Our total training batch is set to 256 (with gradient accumulation) and the learning rate is set to 1e-3. We train 10 epochs and pick the model that performed best on the validation set to evaluate on the test set.