LERG: Leakage-Safe Event-Relevance Gating for Multimodal Demand Forecasting with Supply-Chain News

Authors

  • Lucas Farrell Thomas Lord Department of Computer Science, USC Viterbi School of Engineering, University of Southern California, Los Angeles, CA, USA
  • Vivian Zhao Thomas Lord Department of Computer Science, USC Viterbi School of Engineering, University of Southern California, Los Angeles, CA, USA
  • Isaac Romero Computer Science and Engineering, College of Engineering, University of Michigan, Ann Arbor, MI, USA

DOI:

https://doi.org/10.66421/int.j.datasci.intell.syst59

Keywords:

demand forecasting, multimodal forecasting, supply-chain news, event gating, time series, language models, leakage-safe evaluation

Abstract

Demand forecasting systems routinely combine endogenous sales histories with exogenous signals, yet news text is unusually difficult to use safely because relevance is sparse, publication time is asynchronous, and naive fusion can inject noise into otherwise stable forecasts. This paper proposes Leakage-Safe Event-Relevance Gating (LERG), an encoder-agnostic multimodal forecasting framework that treats text as a conditional residual signal rather than an always-on input. A time-series model first estimates baseline demand. Supply-chain news observed strictly before the forecast origin is mapped to event scores, a relevance gate suppresses corrections on quiet days, and a scale-aware residual model converts relevant events into demand adjustments. The same interface can accept lexical event scores, sentence embeddings, or LLM-derived event logits without changing the forecasting core. To obtain auditable event timing and avoid unsupported causal claims, evaluation uses a controlled semi-synthetic benchmark anchored to a real retail sales trajectory from the Store Item Demand Forecasting Challenge and injects six explicitly parameterized supply-chain event types into 36 heterogeneous series. On the held-out test period, LERG reduces RMSE from 6.2498 to 5.9169 versus a tuned time-series Ridge model (5.33%) and from 6.2333 to 5.9169 versus SVD early fusion (5.08%). On news-active dates, RMSE falls by 12.96% relative to the time-series baseline while quiet-date RMSE is unchanged. A paired bootstrap across 36 series yields a mean RMSE gain of 5.24% with a 95% interval of 4.83%–5.67%. The results support event-gated, forecast-origin-safe text fusion as a compact and auditable design for news-enhanced demand forecasting.

References

[1] R. Carbonneau, K. Laframboise, and R. Vahidov, “Application of machine learning techniques for supply chain demand forecasting,” Eur. J. Oper. Res., vol. 184, no. 3, pp. 1140–1154, 2008, doi: 10.1016/j.ejor.2006.12.004.

[2] Shen, A. (2023). Compute-efficient scaling of large sequential recommendation models under dynamic user behavior. World Journal of Information Technology.

[3] S. Makridakis, E. Spiliotis, and V. Assimakopoulos, “The M5 competition: Background, organization, and implementation,” Int. J. Forecasting, vol. 38, no. 4, pp. 1325–1336, 2022, doi: 10.1016/j.ijforecast.2021.07.007.

[4] S. Makridakis, E. Spiliotis, and V. Assimakopoulos, “The M4 Competition: 100,000 time series and 61 forecasting methods,” Int. J. Forecasting, vol. 36, no. 1, pp. 54–74, 2020, doi: 10.1016/j.ijforecast.2019.04.014.

[5] A. Vaswani et al., “Attention is All You Need,” in Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 5998–6008.

[6] B. Lim, S. Ö. Arık, N. Loeff, and T. Pfister, “Temporal Fusion Transformers for interpretable multi-horizon time series forecasting,” Int. J. Forecasting, vol. 37, no. 4, pp. 1748–1764, 2021, doi: 10.1016/j.ijforecast.2021.03.012.

[7] D. Salinas, V. Flunkert, J. Gasthaus, and T. Januschowski, “DeepAR: Probabilistic forecasting with autoregressive recurrent networks,” Int. J. Forecasting, vol. 36, no. 3, pp. 1181–1191, 2020, doi: 10.1016/j.ijforecast.2019.07.001.

[8] H. Zhou et al., “Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting,” in Proc. AAAI Conf. Artif. Intell., vol. 35, no. 12, pp. 11106–11115, 2021, doi: 10.1609/aaai.v35i12.17325.

[9] T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting,” in Proc. 39th Int. Conf. Mach. Learn., PMLR 162, pp. 27268–27286, 2022.

[10] Y. Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A Time Series Is Worth 64 Words: Long-Term Forecasting with Transformers,” in Proc. Int. Conf. Learn. Representations, 2023.

[11] A. Zeng, M. Chen, L. Zhang, and Q. Xu, “Are Transformers Effective for Time Series Forecasting?” in Proc. AAAI Conf. Artif. Intell., vol. 37, no. 9, pp. 11121–11128, 2023, doi: 10.1609/aaai.v37i9.26317.

[12] Y. Wang, G. Ding, Z. Zeng, and S. Yang, “Causal-Aware Multimodal Transformer for Supply Chain Demand Forecasting: Integrating Text, Time Series, and Satellite Imagery,” IEEE Access, vol. 13, pp. 176813–176829, 2025, doi: 10.1109/ACCESS.2025.3619552.

[13] B. N. Oreshkin, D. Carpov, N. Chapados, and Y. Bengio, “N-BEATS: Neural basis expansion analysis for interpretable time series forecasting,” in Proc. Int. Conf. Learn. Representations, 2020.

[14] M. Jin et al., “Time-LLM: Time Series Forecasting by Reprogramming Large Language Models,” in Proc. Int. Conf. Learn. Representations, 2024.

[15] N. Gruver, M. Finzi, S. Qiu, and A. G. Wilson, “Large Language Models Are Zero-Shot Time Series Forecasters,” in Advances in Neural Information Processing Systems, vol. 36, 2023, doi: 10.52202/075280-0861.

[16] M. Goswami, K. Szafer, A. Choudhry, Y. Cai, S. Li, and A. Dubrawski, “MOMENT: A Family of Open Time-series Foundation Models,” in Proc. 41st Int. Conf. Mach. Learn., PMLR 235, pp. 16115–16152, 2024.

[17] T. Baltrušaitis, C. Ahuja, and L.-P. Morency, “Multimodal Machine Learning: A Survey and Taxonomy,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 41, no. 2, pp. 423–443, 2019, doi: 10.1109/TPAMI.2018.2798607.

[18] P. Xu, X. Zhu, and D. A. Clifton, “Multimodal Learning With Transformers: A Survey,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 10, pp. 12113–12132, 2023, doi: 10.1109/TPAMI.2023.3275156.

[19] Chen, T., Shang, W., & Guo, Z. (2024). Xsla: An SLA-aware heterogeneous XPU scheduling framework for edge AI inference. World Journal of Information Technology, 2(2), 48–59.

[20] Zou, J., Qin, Y., & Rhee, M. (2024). Beyond model scale: A task-aware reliability benchmark for large language models in clinical decision support. Innovation and Technology Studies, 1(1), 39–47.

[21] Fang, Q. (2024). Bias-aware large language models for equitable educational feedback and assessment. Trends in Social Sciences and Humanities Research, 2(1), 41–50.

[22] Qin, Y., Zou, J., & Rhee, M. (2022). An AI-driven intelligent transaction execution framework for internet finance platforms: Cost-sensitive real-time integrity interdiction. Eurasia Journal of Science and Technology, 4(2), 36–44.

[23] Wang, Y., & Zhang, C. (2023). AERIS: Adaptive executor scoped resource management for privacy preserving data processing in Apache Spark. Eurasia Journal of Science and Technology, 5(2), 25-33.

[24] Zhao, W., & Wang, B. (2022). CAPA: A prediction‑driven autoscaling framework for SLO‑aware cloud‑native machine learning inference. Journal of Computer Science and Electrical Engineering, 4(1), 8-16.

[25] Guo, Z., Chen, T., & Shang, W. (2022). Class‑conditional intermediate‑domain adversarial adaptation for cross‑device acoustic scene classification. Journal of Computer Science and Electrical Engineering, 4(1), 26-35

[26] Fang, Q. (2024). True-EWS: Trustworthy early warning with equalized conformal guarantees for equitable student support in U.S. education. Educational Research and Human Development, 1(2), 44–54.

[27] Shen, A., & Huang, S. (2022). A systems-aware framework for intelligent large-scale recommendation models. Journal of Computer Science and Electrical Engineering, 4(1), 36–44.

Downloads

Published

2026-09-10