Big Data-Driven Emotional Semantic Mining of Art Design Works and Construction of Personalized Recommendation System

With the rapid expansion of the digital art market and the diversification of user aesthetic needs, traditional art recommendation systems face two core challenges: one is the fragmented representation of emotional semantic features in massive unstructured art data, and the other is the adaptation gap between static recommendation strategies and dynamic user preferences. In response to this situation, this study constructs an intelligent recommendation system for art design works that integrates multimodal cognitive computing and dynamic collaborative filtering. First, an improved Transformer architecture is used to achieve cross-modal feature alignment. Secondly, based on the emotional dimension theory of cognitive psychology, a hierarchical emotional modeling framework is innovatively designed. The bottom layer uses Bi-LSTM (Bidirectional Long Short-Term Memory) to capture the evolution of aesthetic preferences in time series, the middle layer aggregates the group aesthetic consensus in social networks through graph neural networks, and the top layer establishes an interpretable meta-learning recommendation model. Finally, a user interest prediction model with a memory enhancement mechanism is constructed by combining real-time eye tracking data with historical interaction logs. Experimental results show that the improved Transformer model achieved an F1 value of 0.85, a precision of 0.83, and a recall rate of 0.82 in the task of sentiment feature extraction. In addition, in the non-cold start scenario, the aesthetic matching degree of the improved Transformer model increases from 0.85 to 0.91, while the matching degree of the collaborative filtering method increases from 0.75 to 0.85, showing the model’s powerful ability to capture changes in user interests and dynamically adjust recommendations. In the above data conclusions, the recommendation system of this paper’s research architecture can significantly improve the accuracy of personalized recommendations for art design works.