Dynamic Lifecycle Prediction of FMCG via Multimodal Reviews Combining BLIP and Prophet

Abstract

The Fast-Moving Consumer Goods (FMCG) industry is characterized by high demand volatility, short product lifecycles, and intense market competition. Accurate prediction of product lifecycles is critical for optimizing inventory management and marketing strategies. Traditional forecasting methods often rely heavily on historical sales data, neglecting the rich semantic information embedded in user-generated content, particularly the interplay between textual reviews and visual imagery. This paper proposes a novel predictive framework that integrates multimodal sentiment analysis with advanced time-series forecasting. We utilize the Bootstrapping Language-Image Pre-training (BLIP) model to extract unified semantic features from consumer reviews, effectively capturing the nuanced sentiment expressed through both text and uploaded product images. These multimodal sentiment indices are subsequently incorporated as external regressors into the Prophet forecasting model, which is adept at handling seasonality and trend shifts. Our empirical analysis, conducted on a large-scale dataset of e-commerce FMCG transactions, demonstrates that the proposed BLIP-Prophet framework significantly outperforms unimodal baselines and traditional time-series models. The results highlight the predictive value of visual consumer feedback in understanding market dynamics and offer a robust tool for decision-makers in the retail sector.