Keyphrases are not only statistically or semantically important textual units but also words or phrases that are more likely to attract readers' attention during comprehension. However, existing keyphrase extraction (KPE) studies have mainly focused on improving models' representation learning from input texts, while largely overlooking the connection between keyphrases and human reading behavior. This study investigates whether lightweight webcam-based eye-tracking features can enhance KPE from Chinese academic abstracts in Library and Information Science (LIS).
Motivated by the limited availability of eye-tracking data for Chinese academic reading, we developed a lightweight webcam-based data collection platform by integrating the open-source SearchGazer library. After preprocessing the collected gaze data, we constructed the Chinese LIS eye-tracking corpus (CLIS-ET). We then incorporated three character-level eye-tracking features: first fixation duration (FFD), fixation number (FN) and total fixation duration (TFD), into KPE models to evaluate their effects on extraction performance.
Eye-tracking features improved KPE performance across multiple evaluated architectures, although the magnitude of the gains varied by model and dataset. Five-fold cross-validation on Abstract-320 further showed statistically significant effects for multiple individual and combined eye-tracking features. In particular, the FN + TFD combination reached statistical significance in seven of the eight evaluated KPE architectures, indicating a broadly consistent contribution of complementary reading-behavior signals across model types.
This paper presents a cost-effective eye-tracking methodology to improve keyphrase extraction from academic papers. We also introduce the Chinese academic eye-tracking corpus, which includes key eye-tracking metrics: FN, FFD and TFD. Our findings demonstrate that incorporating these features enhances KPE models, with first fixation duration yielding the most significant performance improvement. The dataset and source code can be accessed at: https://github.com/yan-xinyi/ET_AKE.
