Yi Zhao

Collaboration between academic and industrial institutions significantly boosts the novelty of NLP research outputs, challenging the effectiveness of isolated industrial efforts. WHY_IT MATTERS: This insight could reshape strategies for fostering impactful research collaborations between academia and industry, ultimately enhancing the quality and innovation of academic publications.

An optimal number of thought leaders can enhance team impact, but too many may stifle innovative ideas.

Moderately difficult research in NLP achieves greater academic impact, revealing a critical balance for researchers to target.

Mixed-gender research teams achieve significantly higher citation counts, revealing an optimal gender ratio that could reshape collaborative practices in academia.

Structural fidelity in webpage generation drops significantly with length, revealing that visual appeal doesn't guarantee functional effectiveness in VLMs.

Ditching the global [CLS] token and embracing patch-level inference unlocks surprising gains in multi-label image recognition, all without training.

Results-based novelty alone beats combined theoretical, methodological, and results-based novelty when it comes to scientific impact, challenging the assumption that more novelty is always better.

LLMs still can't reliably judge the novelty of research papers, even when fine-tuned on peer review data, suggesting current models lack a deep understanding of scientific contributions.

Women in Library and Information Science are significantly more likely to choose qualitative methods, while men gravitate towards theoretical approaches, revealing a stark gender divide in research practices.