Key points are not available for this paper at this time.
Prioritizing product features is a crucial task for software startups due to scarce resources and rapidly changing markets. Existing prioritization approaches (for example, MoSCoW, Kano) are subjective and struggle to handle large sets of user feedback, resulting in inefficient product strategies and longer time to market. This paper bridges this gap by developing an automated feature prioritization approach based on a combined Natural Language Processing (NLP) and Multi-Criteria Machine Learning (MCML) system. The framework uses transformer-based sentence embeddings (specifically, a fine-tuned BERT model) to capture natural language semantics from user reviews and applies k-means clustering—with K selected via silhouette analysis—to discover latent product features. A sentiment-informed scoring system captures user feedback polarity; multiple criteria (sentiment, frequency, linguistic intensity, and user impact) are evaluated to generate a feature vector. An ensemble model combining Gradient Boosting and Random Forest (with α tuned via grid search on the validation set) is then used to predict a feature priority class and produce a ranked list. The model is applied to a real Google Play Store review dataset; the results show accuracy 0.94, precision 0.94, recall 0.94, and F1-Score 0.93, with a macro-averaged F1 of 0.90 and balanced accuracy of 0.90. The results demonstrate the potential of NLP and multi-criteria learning for automated feature prioritization at scale. Our research offers an automated approach that addresses the challenges faced by software startups and may help enhance product–market fit and accelerate time to market.
Pattyn et al. (Mon,) studied this question.