This paper introduces MDSF (Multidimensional Data Storytelling Framework), a large-language-model system that automates the data-analysis-to-narrative pipeline for business analytics. It models data as a multidimensional table with breakdowns, indicators, and filters, enumerates subspaces, and mines typed 'insights' (distributions, trends, anomalies, year-over-year and month-over-month comparisons, correlations, root-cause) using augmented-analysis algorithms - Prophet and SR-CNN for time series, 3-sigma and isolation forest for outliers. A multi-angle scoring mechanism plus a Top-K step ranks the insights, and a fine-tuned LLM turns the selected insights into coherent narrative reports with charts and tables; an agent-based mechanism supports real-time, context-aware storytelling continuation as users edit reports. Experiments and user studies report gains over prior methods in insight-ranking accuracy, descriptive quality, and narrative coherence, plus reduced interpretive bias.
This paper introduces MDSF (Multidimensional Data Storytelling Framework), a large-language-model system that automates the data-analysis-to-narrative pipeline for business analytics. It models data …