Call for Customized Conversation: Customized Conversation Grounding Persona and Knowledge

Authors: Yoonna Jang, Jungwoo Lim, Yuna Hur, Dongsuk Oh, Suhyune Son, Yeonsoo Lee, Donghoon Shin, Seungryong Kim, Heuiseok Lim10803-10812

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

Reproducibility Variable Result LLM Response
Research Type Experimental To evaluate the abilities to make informative and customized utterances of pre-trained language models, we utilize BART and GPT-2 as well as transformer-based models. We assess their generation abilities with automatic scores and conduct human evaluations for qualitative results.
Researcher Affiliation Collaboration Yoonna Jang1 , Jungwoo Lim1 , Yuna Hur1 , Dongsuk Oh1, Suhyune Son1, Yeonsoo Lee2, Donghoon Shin2, Seungryong Kim1 , and Heuiseok Lim1 1Department of Computer Science and Engineering, Korea University 2Language AI Lab, NCSOFT
Pseudocode No The paper describes the model architecture and objective functions but does not include any pseudocode or algorithm blocks.
Open Source Code Yes In this work, we introduce a new dataset, call For Customized conversation dataset1 (called Fo Cus), that supports knowledge-grounded answers that reflect user s persona. 1http://github.com/pkchat-focus/Fo Cus
Open Datasets Yes In this work, we introduce a new dataset, call For Customized conversation dataset1 (called Fo Cus), that supports knowledge-grounded answers that reflect user s persona. 1http://github.com/pkchat-focus/Fo Cus
Dataset Splits Yes We split the collected data into train, valid and test sets. The detailed statistics of our dataset are summarized in Table 2. ... Table 2: Statistics of Fo Cus dataset. # Dialogs Train 11,562 Valid 1,445 Test 1,445
Hardware Specification Yes Fine-tuning them on the entire data with 2 epochs takes approximately 10 hours with one RTX-8000 GPU.
Software Dependencies No The paper states, "We implement the models based on the source code of Hugging Face s transformers (Wolf et al. 2020, 2019)," but does not provide specific version numbers for software dependencies like Hugging Face Transformers, PyTorch, TensorFlow, or Python.
Experiment Setup Yes We use a batch size of 4 with a gradient accumulation of 32. Adam optimizer is used, and the learning rate is set as 6.25e-5, where 훽1 = 0.9, 훽2 = 0.999 with linear decay. ... For the utterance generation, we use the nucleus sampling with top-p = 0.9 and sampling temperature with 0.7. The maximum sequence length is set to 20.