Emotional by design

A new AI system can convert social media discussion about a product into a new design that takes into account user needs more accurately than earlier approaches, according to research in the International Journal of Information and Communication Technology.

The work addresses the problem of complexity in attempting to extract useful information from social network data for product development. Comments and reviews are typically unstructured, meaning they do not follow a fixed format, and also have many variables, such as sentiment, context, and usage scenarios, which makes it difficult to translate into insights about how people feel about products.

A deep-learning framework is at the heart of the system and combines various AI components. Firstly, it uses a multi-scale attention network to identify emotional needs in user comments. Attention in machine learning refers to a mechanism that prioritises the most relevant information in a dataset. The idea of multi-scale processing means it captures both detailed and broad patterns in language. The second component is a generative adversarial network (GAN). This uses two models working against each other, with one generating images and the other evaluating them. In addition, a spatial cross-reconstruction module refines image features, while a semantic correlation module links textual emotion signals to visual attributes. All of this works to improve the link between what the users say about the original product and the new design.

In tests, the model achieved more than 90 per cent accuracy in identifying the users’ emotional needs. This improves on existing methods and suggests that AI might help with data-driven product design informed by user sentiment and social media behaviour.

Wang, C. (2026) ‘Deep learning-based innovative product design driven by social network data’, Int. J. Information and Communication Technology, Vol. 27, No. 49, pp.59–78.