JYJijun Yu

Jijun Yu

*This is an algorithmically generated author record based on publicly available information.

Jijun Yu has a total of 6 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Long-Term Fairness-Aware Recommendation via Adaptive Fairness Metric Selection

    by Jijun Yu, Minghua Xiong

    Fairness in recommendation systems has drawn growing attention due to rising societal and regulatory concerns over algorithmic bias. Existing fairness-aware approaches typically mitigate bias by either removing sensitive attributes via representation learning or leveraging causal-path interventions (e.g., counterfactual or specific-path debiasing) to distinguish genuine causal effects from confounder-induced correlations between sensitive attributes and user preferences. However, when it comes to evaluation, most prior work adopts both Demographic Parity (DP) and Equal Opportunity (EO) as simultaneous criteria, yet overlooks their inherent tension and the causal nature of the sensitive attribute. Specifically, if a sensitive attribute genuinely drives preference variation, enforcing DP forces equal exposure across groups, contradicting natural interest diversity and severely hurting accuracy; conversely, for spurious correlations, relying solely on EO fails to remove confounder-introduced bias. More importantly, these metrics are typically computed in a static, one-shot manner, ignoring that recommendation is an iterative process where even minor initial disparities can be amplified over time through feedback loops, eventually leading to substantial long-term unfairness. Nevertheless, existing studies rarely address such dynamic, long-term fairness implications, leaving a critical gap in both evaluation and optimization. To resolve this, we propose Long-term Fairness-aware Recommendation via Adaptive Fairness Metric Selection (LFR-via-AFMS). Our framework first learns the causal structure to identify whether the sensitive attribute has a genuine causal effect or merely a spurious association with user preferences. Based on this diagnosis, it adaptively selects the most appropriate fairness criterion: Equal Opportunity for true causality, which allows legitimate group differences in preference, and Demographic Parity for spurious correlations, which eliminates unjustified disparities entirely. The adaptively chosen metric is then integrated into an actor–critic reinforcement learning reward to optimize long-term fairness without sacrificing accuracy. Extensive experiments on Alibaba and MovieLens datasets, with five independent runs and statistical significance testing, demonstrate that the proposed method achieves a superior fairness-accuracy trade-off compared with state-of-the-art baselines, and the adaptive metric selection proves indispensable for maintaining both equity and recommendation quality. We validate the causal diagnosis module through simulation studies with known ground truth and sensitivity analyses confirming robustness across threshold choices.

  • Open Access

    Long-Term Fairness-Aware Recommendation via Adaptive Fairness Metric Selection

    by Jijun Yu, Minghua Xiong

    Fairness in recommendation systems has drawn growing attention due to rising societal and regulatory concerns over algorithmic bias. Existing fairness-aware approaches typically mitigate bias by either removing sensitive attributes via representation learning or leveraging causal-path interventions (e.g., counterfactual or specific-path debiasing) to distinguish genuine causal effects from confounder-induced correlations between sensitive attributes and users perference. However, when it comes to evaluation, most prior work adopts both Demographic Parity (DP) and Equal Opportunity (EO) as simultaneous criteria, yet overlooks their inherent tension and the causal nature of the sensitive attribute. Specifically, if a sensitive attribute genuinely drives preference variation, enforcing DP forces equal exposure across groups, contradicting natural interest diversity and severely hurting accuracy; conversely, for spurious correlations, relying solely on EO fails to remove confounder-introduced bias. More importantly, these metrics are typically computed in a static, one-shot manner, ignoring that recommendation is an iterative process where even minor initial disparities can be amplified over time through feedback loops, eventually leading to substantial long-term unfairness. Nevertheless, existing studies rarely address such dynamic, long‑term fairness implications, leaving a critical gap in both evaluation and optimization. To resolve this, we propose Long-term Fairness-aware Recommendation via Adaptive Fairness Metric Selection. Our framework first learns the causal structure to identify whether the sensitive attribute has a genuine causal effect or merely a spurious association with user preferences. Based on this diagnosis, it adaptively selects the most appropriate fairness criterion: Equal Opportunity for true causality, which allows legitimate group differences in preference, and Demographic Parity for spurious correlations, which eliminates unjustified disparities entirely. The adaptively chosen metric is then integrated into an actor-critic reinforcement learning reward to optimize long-term fairness without sacrificing accuracy. Extensive experiments on Alibaba and MovieLens datasets demonstrate that proposed method achieves a superior fairness-accuracy trade-off compared to state-of-the-art baselines, and the adaptive metric selection proves indispensable for maintaining both equity and recommendation quality.

  • Open Access

    A Hybrid Ensemble Learning Approach for Intelligent Cost Prediction in Public Construction Projects

    by Shuojun Lyu, Lan Tu, Jijun Yu

    Budget overruns continue to be a widespread and enduring issue in government-funded construction endeavors, with actual expenditures often surpassing planned figures by more than 20%. Conventional approaches to cost forecasting, which depend primarily on professional expertise and parameter-driven models, find it difficult to account for the intricate, nonlinear interdependencies among the diverse variables influencing project expenditures. Although recent research has highlighted the advantages of machine learning over traditional techniques, relying on a single predictive model frequently leads to poor generalizability and insufficient transparency, thereby limiting real-world implementation. In this study, we develop a hybrid stacking-based ensemble prediction framework that combines four diverse base learners—XGBoost, LightGBM, CatBoost, and Random Forest—unified through a stacking strategy with Ridge Regression serving as the second-level aggregator. The proposed system also includes a structured data preprocessing workflow designed to address the specific properties of publicly available construction datasets and leverages SHapley Additive exPlanations (SHAP) to enhance model transparency. Evaluation on an authentic public-sector construction cost dataset shows that the proposed method attains an R² of 0.971, an RMSE of 16.3×10³ USD, and a MAPE of 5.7%, substantially surpassing each individual baseline model. Component-wise ablation experiments validate the effectiveness of each module within the framework, and SHAP-based interpretation indicates that project duration, gross floor area, and number of floors serve as the most significant cost-influencing factors. These results demonstrate that hybrid ensemble strategies, when paired with interpretability tools, provide a viable and dependable solution for intelligent cost estimation in public construction project management.


  • Open Access

    Deep Reinforcement Learning for Boosting Individual and Aggregate Diversity in Product Recommendation Systems

    by Jijun Yu, Shuojun Lyu, Peihui Chen

    The significance of individual and aggregate diversities has been recognized as an important aspect of the recommendation algorithm. Traditional diversity-enhancing recommendation systems have predominantly optimized accuracy and diversity through static rating processes, neglecting the dynamic, long-term user-recommender interactions. This study proposes a novel approach employing deep reinforcement learning (DRL) to derive an optimal selection policy that enhances accuracy and individual and aggregate diversities through maximizing expected cumulative rewards. This approach employs Deep Reinforcement Learning (DRL) to create a dynamic recommendation list that integrates diversity using a Markov decision process based on user feedback and ratings. It calculates the rewards by examining the proximity of the recommended list to an ideal distribution across all recommendations, thereby promoting both individual and collective diversity in the items selected. The learning mechanism is powered by the Deep Deterministic Policy Gradient (DDPG) method to refine the selection strategy. Comprehensive tests on datasets from Alibaba and Grouplens confirm that our method achieves accuracy and substantially improves diversity compared to baselines.