Yuning Shi

Dr. Yuning Shi

orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
2 February 2024
Verified
Conference Paper
Open access
YSYuning Shi
Yuning Shi
KMKimura Masaomi
Kimura Masaomi
Biomedical named entity recognition (NER) is a crucial subtask in the field of information extraction within natural language processing (NLP). Its primary objective is to identify and classify entities in biomedical text, playing a pivotal role in applications such as medical information retrieval and biomedical knowledge discovery. In this paper, we propose several enhanced versions of BERT-BiLSTM-CRF and BERT-IDCNN-CRF by incorporating an attention mechanism or lambda layer to improve entity recognition accuracy. Specifically, we utilize the attention mechanism to enable the model to learn interrelationships among all words in the input sequence. Additionally, we employ the lambda layer to enhance the model's capacity for capturing semantic relationships between words and considering word order. This integration results in superior accuracy in entity recognition. We evaluate our proposed methods using the i2b2 2010 dataset and six additional biomedical datasets from the Biomedical Language Understanding and Reasoning Benchmark (BLURB), including JNLPBA, BC2GM, BC5CDR, AnatEM, BioNLP-CG, and NCBI-disease. Experimental results demonstrate that our proposed methods achieve higher accuracy than the original methods, indicating superior capabilities in medical knowledge extraction for our models.

14 November 2026 in Multimedia Systems
Verified
Journal Article
Closed access
PDF file
YSYuning Shi
Yuning Shi
QSQingwei Song
Qingwei Song
NKNaoyuki Kubota
Naoyuki Kubota


12 October 2026 in International Journal of Machine Learning and Cybernetics
Verified
Journal Article
Open access
PDF file
YSYuning Shi
Yuning Shi
NKNaoyuki Kubota
Naoyuki Kubota
Achieving robust human action recognition (HAR) under cross-domain and low-resource conditions remains challenging due to substantial distribution shifts and limited annotated data. Existing transfer learning methods primarily rely on static knowledge transfer or feature distribution alignment, limiting their ability to capture higher-order structural relationships and adaptively transfer knowledge across domains. To address these limitations, we propose a topology-guided transfer learning framework that leverages source-domain structural knowledge to facilitate target-domain adaptation. Specifically, we construct an unsupervised topology-preserving graph to capture the intrinsic structure of the source-domain feature space, which serves as a transferable topological memory. A gated attention fusion module dynamically retrieves and integrates the structural information most relevant to each target-domain sample. Finally, topology-aware alignment and knowledge distillation are jointly employed to preserve structural consistency during training and distill topological knowledge into the target model for memory-free inference. We evaluate the proposed method on two cross-domain transfer tasks using UCF101 as the shared source domain, with HMDB51 and a self-collected dataset serving as the respective target domains. Experimental results demonstrate that our approach consistently outperforms representative transfer learning baselines across multiple backbones and cross-domain settings. These findings confirm that integrating transferable topological memory with adaptive knowledge transfer effectively improves cross-domain HAR performance.

14 June 2026 in JMIR Preprints
Verified
Preprint
Open access
YZYuqi Zhang
Yuqi Zhang
YSYuning Shi
Yuning Shi
JYJunichi Yamamoto
Junichi Yamamoto
NKNaoyuki Kubota
Naoyuki Kubota
BACKGROUND Continuous, preventive monitoring of cognitive and physical function outside clinical visits is of growing interest, but faces two gaps. Evaluation is typically tied to fixed centers, limiting access for those who cannot easily travel; and conventional test batteries are isolated, repetitive tasks that elicit low engagement. Serious games improve motivation but are usually single-domain and report only endpoint scores, whereas fixed living-lab installations cannot be relocated. No existing approach combines engagement, access, and a standardized environment in one system. OBJECTIVE This study aimed to develop the Mobile Escape-Game Lab (MEGL), a relocatable serious game platform integrating four cognitive and physical task modules into a single escape-game experience, and to evaluate its usability, workload, and user experience in healthy adults before deployment in broader populations. METHODS In this single-session development and usability study, MEGL chained four fixed-order modules—an ankle-controlled game (Balance Wood), a digital Trail Making Test (d-TMT), a tabletop Block Design Test (BDT), and an object-search task—guided by a scenario-based chatbot, deployed within TLL+ (Trailer Living Lab+), a relocatable trailer providing a standardized, daily-living-style environment with pre-installed sensing. Healthy adults were recruited by convenience sampling at Tokyo Metropolitan University and tested individually in April-May 2026. We assessed usability with the System Usability Scale (SUS; 0-100), per-module workload with the NASA Task Load Index (NASA-TLX; 1-7), and experience with a 6-item scale (1-5) and open-ended questions. Data were summarized descriptively with 95% CIs. Process data were collected only as an exploratory illustration, not to assess function. RESULTS Of 14 enrolled adults, 12 (11 men, 1 woman; mean age 30.2 years, SD 4.2) were analyzed. The mean SUS score was 71.2 (95% CI 64.8-77.7), slightly above the benchmark of 68, indicating acceptable usability. Per-module NASA-TLX workload was modest and broadly similar, ranging from 2.86 (95% CI 2.35-3.37) for d-TMT to 3.43 (95% CI 2.83-4.03) for object search. The overall-experience composite was positive (mean 3.90; 95% CI 3.53-4.28), led by enjoyment (4.33), while willingness to reuse was lowest (3.67). Qualitative feedback highlighted the immersive escape-game framing and chatbot guidance as strengths, and instruction clarity and task flow as the main area for improvement. CONCLUSIONS To our knowledge, MEGL is the first serious game platform to combine an escape-game-based multimodal task protocol with a relocatable, standardized deployment environment. Unlike single-domain serious games and fixed living labs, it jointly addresses engagement, access, and standardization in one system. By demonstrating acceptable usability and a positive experience in healthy adults, it provides a usable, reproducible foundation for integrated cognitive and physical task experiences in portable, standardized settings. With further development in broader and older populations, MEGL could bring standardized functional-task experiences closer to the communities that motivated its design.