BioCiteGPT: Recurrent Retrieval-Augmented Large Language Models for Faithful Biomedical Citation Recommendation
Poster Number: P175
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Knowledge Representation and Information Modeling
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Citation recommendation is the task of providing relevant articles to cite for a given text, and plays a key role in scientific research. Due to the ever-growing literature, finding appropriate citations is becoming more challenging and effective citation recommenders are of greater importance. In this work, we present a LLM-based recurrent retrieval-augmented framework for citation recommendation. In addition, our recommender works on a sentence level, as opposed to other models which are article or topic-level. Generative LLMs such as ChatGPT have shown effectiveness on a variety of tasks, but are prone to hallucinations (creating imaginary information), especially when used for citation recommendation. Our approach combines generative LLMs and retrieval augmentation whereby the two approaches are used to enhance each other. In addition, we train a LLaMA2-based discriminator which further distinguishes whether an article is relevant to a given sentence, further enhancing the reliability of our process. We show that combining these approaches improves the reliability of citations generated by ChatGPT. Our study is supported by a dataset we constructed consisting of sentences from PubMed articles related to Alzheimer's disease, along with information about articles cited by those sentences.
Speaker(s):
Jeffrey Zhang, PhD
Yale University
Author(s):
Qianqian Xie, PhD - Yale University; Jeffrey Zhang, PhD - Yale University; Yan Wang, PhD - Yale University; Fongci Lin, PhD - Yale University; Yi-chung Wang, BS - Yale University; Ziqing Ji, BS - Yale University; Haoting Chen, BS - Yale University; Qingyu Chen, PhD - Yale University; Hua Xu, Ph.D - Yale University;
Poster Number: P175
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Knowledge Representation and Information Modeling
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Citation recommendation is the task of providing relevant articles to cite for a given text, and plays a key role in scientific research. Due to the ever-growing literature, finding appropriate citations is becoming more challenging and effective citation recommenders are of greater importance. In this work, we present a LLM-based recurrent retrieval-augmented framework for citation recommendation. In addition, our recommender works on a sentence level, as opposed to other models which are article or topic-level. Generative LLMs such as ChatGPT have shown effectiveness on a variety of tasks, but are prone to hallucinations (creating imaginary information), especially when used for citation recommendation. Our approach combines generative LLMs and retrieval augmentation whereby the two approaches are used to enhance each other. In addition, we train a LLaMA2-based discriminator which further distinguishes whether an article is relevant to a given sentence, further enhancing the reliability of our process. We show that combining these approaches improves the reliability of citations generated by ChatGPT. Our study is supported by a dataset we constructed consisting of sentences from PubMed articles related to Alzheimer's disease, along with information about articles cited by those sentences.
Speaker(s):
Jeffrey Zhang, PhD
Yale University
Author(s):
Qianqian Xie, PhD - Yale University; Jeffrey Zhang, PhD - Yale University; Yan Wang, PhD - Yale University; Fongci Lin, PhD - Yale University; Yi-chung Wang, BS - Yale University; Ziqing Ji, BS - Yale University; Haoting Chen, BS - Yale University; Qingyu Chen, PhD - Yale University; Hua Xu, Ph.D - Yale University;
BioCiteGPT: Recurrent Retrieval-Augmented Large Language Models for Faithful Biomedical Citation Recommendation
Category
Poster Invite
Description
Date: Tuesday (11/12)
Time: 05:00 PM to 06:30 PM
Room: Grand Ballroom (Posters)
Time: 05:00 PM to 06:30 PM
Room: Grand Ballroom (Posters)