A multigrained preference analysis method for product iterative design incorporating AI-generated review detection.

Online reviews significantly influence consumer purchasing decisions and serve as a vital reference for product improvement. With the surge of generative artificial intelligence (AI) technologies such as ChatGPT, some merchants might exploit them to fabricate deceptive positive reviews, and competitors may also fabricate negative reviews to influence the opinions of consumers and designers. Attention must be paid to the trustworthiness of online reviews. In addition, the opinions expressed by users are limited, and design details hidden behind reviews also affect the product usage experience. Therefore, on the basis of integrated AI-generated review detection, a multigrained user preference analysis method is proposed in this work. The proposed method utilizes pre-trained language models and designs an authenticity detection model for online reviews. Subsequently, attribute-grained preference analysis is considered a text-filling problem and uses the text-infilling objective for domain-adaptive pretraining, facilitating knowledge transfer. On the basis of the feature selection algorithm, a calculation method for the importance of product design features is proposed by introducing a random idea. The proposed method analyzes user preferences at the granularity of product attributes and design features, enabling targeted cost control and optimization in product development and guiding design decisions. Rigorous comparative and few-shot experiments substantiate the superiority of the proposed method.