Siyi Liu - University of Pennsylvania

About

I'm final-year PhD student at Department of Computer and Information Science, University of Pennsylvania, advised by Prof. Dan Roth

My research spans natural language processing and broader artificial intelligence. Recently, I’m particularly interested in studying different types of conflicts that emerge in modern AI systems, including:

hallucinations, factuality and alignment - model generations that are non-factual, toxic, or unsubstatiated/contradictory to the contexts [2][3]
retrieval, knowledge conflicts, and reasoning - How can LLMs reason through conflicting knowledge from different sources (contextual/parametric knowledge)? [1][4]
perspectives and biases - conflicts across different perspectives and opinions [7] [8] [9]

Some of my other ongoing projects include:

Teaching LLMs to learn from failures/feedback
Deepfake Video Detection
Representation Learning for Authorship Representation

Some of my other past projects include:

Open Source Project (Core Contributor) - Distilled Generative Text Embedding Model
Framing Bias Detection

Selected Publications

[1] ConflictScore: Measuring How Language Models Handle Conflicting Evidence
Siyi Liu, Patrick Xia, et al.
In submission

[2] DeeptraceReward: Learning Human-Perceived Fakeness in Generated Videos with Multimodal LLMs
Xingyu Fu, Siyi Liu, et al.
Neurips GenProCC Workshop 2025

[3] Towards Long Context Hallucination Detection
Siyi Liu , Kishaloy Halder, et al.
NAACL 2025 Findings

[4] Open Domain Question Answering with Conflicting Contexts
Siyi Liu , Qiang Ning, et al.
NAACL 2025 Findings

[5] Using LLM for improving key event discovery: Temporal-guided news stream clustering with event summaries
Nishanth Nakshatri, Siyi Liu, Sihao Chen, Daniel Hopkins, Dan Roth, Dan Goldwasser
EMNLP 2023

[6] Open-Domain Event Graph Induction for Mitigating Framing Bias
Siyi Liu, Hongming Zhang, Hongwei Wang, Kaiqiang Song, Dan Roth, Dong Yu
arXiv

[7] Design Challenges for a Multi-Perspective Search Engine
Sihao Chen*, Siyi Liu* , Xander Uyttendaele, Yi Zhang, William Bruno, Dan Roth
NAACL 2022 Findings

[8] MultiOpEd: A Corpus of Multi-Perspective News Editorials
Siyi Liu, Sihao Chen, Xander Uyttendaele, Dan Roth
NAACL 2021
[Code] [Slides] [Poster] [Talk]

[9] Detecting frames in news headlines and its application to analyzing news framing trends surrounding US gun violence
Siyi Liu, Lei Guo, Kate Mays, Margrit Betke, Derry Tanti Wijaya
CoNLL 2019

[10] Learning to mirror speaking styles incrementally
Siyi Liu*, Ziang Leng*, Derry Wijaya
arXiv