A multi-view contrastive embedding framework for filtering fuzzy requirements of complex products

In complex product development, requirement teams must filter large volumes of user input to identify valid and representative requirements. Compared to professional users, broad user requirements come from diverse sources such as feedback, surveys, and social media, but are often subjective, unstructured, and fuzzy—posing challenges for effective filtering. Existing methods typically overlook this fuzziness. To address this, we propose a multi-view contrastive embedding framework for filtering fuzzy requirements. Requirement triples are modeled as nodes in a knowledge graph and extended into multiple hyper-views for fuzziness-aware representation learning. We integrate knowledge graph embedding with a contrastive learning mechanism. By leveraging multi-view modeling and a fuzziness-aware scoring function, the proposed framework effectively captures and models the degree of fuzziness in user requirements, thereby enabling robust filtering of ambiguous requirements. Experiments on real-world datasets show that our method outperforms existing approaches in filtering and ranking tasks, offering a robust solution for large-scale fuzzy requirement analysis.