Jiashun Duan

Dr. Jiashun Duan

orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
21 November 2024 in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
Verified
Journal Article
Open access
YPYan Pan
Yan Pan
JGJunhui Gao
Junhui Gao
JDJiashun Duan
Jiashun Duan
JSJuan Shi
Juan Shi
BGBin Guo
Bin Guo
YHYujiao Hu
Yujiao Hu
Instant delivery, which has a strict time limit, has become a fundamental service in modern cities. Increasing demands and labor shortages make efficient instant delivery more challenging. Recently, the industry has recruited Unmanned Aerial Vehicles (UAVs) and part-time couriers to address this critical issue. However, the inherent delivery ability of crowdsourced couriers and the limited battery capacity of UAVs make them hard to solely meet the rapidly increasing instant delivery demands. Therefore, this paper proposes the first air-ground cooperative UAV and Courier Delivery paradigm, called UCD, to develop an optimal courier recruitment plan and assignment of instant delivery tasks for UAVs and couriers. In the UCD paradigm, courier and UAV delivery models are introduced, and a recruitment plan that considers both delivery demand and courier availability is designed. With this foundation, a data-driven algorithm is proposed to optimize delivery task assignments, maximizing the flexibility of UAVs and the capacity of couriers. Comprehensive evaluations are conducted on multiple long-term real-world datasets to demonstrate the superior performance of UCD.

1 October 2022
Verified
Conference Paper
Closed access
JDJiashun Duan
Jiashun Duan
XZXin Zhang
Xin Zhang
CXChi Xu
Chi Xu
Albeit great progress has been witnessed in event extraction, the accuracies achieved up to now by various automatic models still can not meet the performance requirements of some special applications such as disaster monitoring and rescue. It motivates us to introduce a new human-in-loop extraction mode called interactive event extraction (IEE), which works iteratively. Each iteration consists of three main steps: "model recommending candidate results → manual selecting and correcting → model re-training and updating". For candidate recommendation, we build an MRC (Machine Reading Comprehension)-based model that can output several most likely candidate elements, i.e., candidate triggers and arguments, by confidence evaluation. For model re-training and updating, we proposed an incremental learning method named LwF4IEE (Learning without Forgetting for IEE), which employs manual selected and corrected results as hard label and prediction of original model as soft label to avoid catastrophic forgetting. We conduct extensive experiments on datasets constructed from real-world Chinese texts. The results show that when setting the number of candidates to be 5, recalls of triggers and arguments reach 93.80% and 90.58% respectively, which is 11.51% and 11.33% higher compared with the basic MRC-based automatic extraction model. Moreover, LwF4IEE increases the recall of triggers by 2.71% on specific event types and only decreases by 0.24% on other types, achieving the purpose of learning without forgetting.

14 September 2024 in Preprints.org
Verified
Preprint
Open access
PDF file
JDJiashun Duan
Jiashun Duan
XZXin Zhang
Xin Zhang
Event detection provides a crucial foundation for intelligent text analysis and supports numerous downstream applications. However, existing detection models often fail to meet the high accuracy requirements and typically require a human–machine collaborative detection mode involving “machine detection followed by manual audit and correction.” Analysis reveals that the recall rate of machine detection is a major factor limiting the efficiency of human–machine collaboration. To address this at its root and ensure high recall, an event trigger recommendation task is introduced, and a span-based recommendation model for Chinese, SpETR (Span-based Event Trigger Recommender), is developed. This model integrates confidence information from both trigger identification and event classification stages to recommend candidate triggers and their event types for each event. To enhance recall, a semantic boundary smoothing method is proposed, which assigns soft labels to spans around gold triggers based on semantic completeness and part-of-speech, combined with a modified focal loss function to make the model predictions smoother. Experiments on DuEE and a self-built dataset show that SpETR is most effective when the number of recommendations is set to 3, achieving recall rates of 97.76% and 98.70%, respectively, comprising improvements of 4.30% and 5.22% over the optimal baseline models. In terms of human–machine collaborative detection, SpETR saved 30.6% and 25.8% of time compared to traditional detection models on two datasets, and the average F1 score of the annotation results was also improved by 5.35% and 3.48%, respectively.

29 September 2024 in Applied Sciences
Verified
Journal Article
Open access
PDF file
JDJiashun Duan
Jiashun Duan
XZXin Zhang
Xin Zhang
Event detection is a crucial task in information extraction. Existing research primarily focuses on machine automatic detection tasks, which often perform poorly in certain practical applications. To address this, an interactive event-detection mode of “machine recommendation-human review–machine incremental learning” was proposed. In this mode, we study a few-shot continual class-incremental learning scenario, where the challenge is to learn new-class events with limited samples while preserving memory of old class events. To tackle these challenges, we propose a class-incremental learning method for interactive event detection via Interaction, Contrast and Distillation (ICD). We design a replay strategy based on representative and confusable samples to retain the most valuable samples under limited conditions; we introduce semantic-boundary-smoothness contrastive learning for effective learning of new-class events with few samples; and we employ hierarchical distillation to mitigate catastrophic forgetting. These methods complement each other and show strong performance. Experimental results demonstrate that, in the 5-shot 5-round class incremental-learning settings on two Chinese event-detection datasets ACE and DuEE, our method achieves final recall rates of 71.48% and 90.39%, respectively, improving by 6.86% and 3.90% over the best baseline methods.

16 December 2025 in Information Processing & Management
Verified
Journal Article
Closed access
JDJiashun Duan
Jiashun Duan
YPYan Pan
Yan Pan
WWWei Wu
Wei Wu
Fangfang LIFangfang LI
Fangfang LI
XZXiang Zhao
Xiang Zhao
XZXin Zhang
Xin Zhang