Hoan-Suk Choi

Time-sensitive question answering (QA) requires that the cited evidence be temporally valid at query time. However, existing evaluations focus on answer correctness while overlooking evidence timeliness. We presented a program-verifiable evaluation framework that explicitly decouples answer correctness from temporal validity. Our oracle-answer diagnostic fixes answers (Acc=1.0) while measuring evidence validity, revealing that even perfect answers cite temporally invalid documents 11.8% of the time (TMR=0.118), yielding RAS=0.903 under distance-based decay. We introduced time-consistency metrics (RAS, RAR, TMR) and evaluated on TimeFact-CEO (synthetic, N=4,793) and Wikidata-CEO (semi-structured, N=1,493). A lightweight lexico-temporal reranker improves the temporal validity rate (RAR) from 48.2% to 88.2% (+40.0pp) under an oracle-answer evaluation. Error analysis showed that the retrieval coverage (76.3% at K=3)was the primary bottleneck.We demonstrate the framework in a CEO-tenure setting where validity intervals are explicit and uniquely verifiable. Applying it to other domains is feasible when comparable interval metadata are available, and we treat cross-domain validation in future work.

Lung diseases significantly contribute to global morbidity and mortality, underscoring the importance of accurate and efficient diagnostic tools. This study introduces a lightweight two-dimensional convolutional neural network (LW-2DCNN) integrated with ReciproCAM, an explainable artificial intelligence (XAI) technique, specifically designed to classify multiple pulmonary conditions from chest computed tomography (CT) scans. Our model effectively addresses limitations of existing diagnostic methods by enhancing computational efficiency, diagnostic accuracy, and interpretability. Utilizing a curated dataset comprising 10,000 CT images from a Korean population, the LW-2DCNN achieved exceptional accuracy of 99.80% across five diagnostic categories: normal, lung cancer, pneumothorax, pneumonia, and interstitial lung disease (ILD). Comparative analyses demonstrated superior performance relative to established benchmark models, including EfficientNetB0, ResNet50, InceptionV3, DenseNet121, NASNetMobile, MobileNetV2, and Xception. The integrated ReciproCAM technique significantly enhanced clinical interpretability by highlighting the precise anatomical regions influencing the model’s diagnostic decisions. Ablation studies validated the optimal choice of the Adam optimizer and demonstrated the model’s computational feasibility, predicting and explaining 1000 images in approximately 150 seconds on a CPU-only device. This indicates strong potential for deployment in resource-constrained clinical settings.

The journal retracts the article titled “Innovative Use of Self-Attention-Based Ensemble Deep Learning for Suicide Risk Detection in Social Media Posts”

Suicidal ideation constitutes a critical concern in mental health, adversely affecting individuals and society at large. The early detection of such ideation is vital for providing timely support to individuals and mitigating its societal impact. With social media serving as a platform for self-expression, it offers a rich source of data that can reveal early symptoms of mental health issues. This paper introduces an innovative ensemble learning method named LSTM-Attention-BiTCN, which fuses LSTM and BiTCN models with a self-attention mechanism to detect signs of suicidality in social media posts. Our LSTM-Attention-BiTCN model demonstrated superior performance in comparison to baseline models in the realm of classification and suicidal ideation detection, boasting an accuracy of 0.9405, a precision of 0.9385, a recall of 0.9424, and an F1-score of 0.9405. Our proposed model can aid healthcare professionals in recognizing suicidal tendencies among social media users accurately, thereby contributing to efforts to reduce suicide rates.