With the increasing diversification of cultural product consumption demand, traditional recommendation algorithms that rely on the similarity between products and require detailed description of content and feature extraction can no longer meet the demand. This paper applies a user-based collaborative filtering algorithm, which is driven by user behavior, fits the cultural product recommendation scenario, and optimizes the performance of the recommendation system. This study collects and processes user behavior data, extracts sentiment and topic features in combination with natural language processing technology, and constructs a multi-dimensional user feature vector. The cultural product content is analyzed through TF-IDF, LDA topic modeling, and knowledge graph to improve the recommendation accuracy. To solve the cold start and sparsity problems, the KNN algorithm and SVD matrix decomposition technology are used to make recommendations based on content features. Real-time dynamic recommendation captures user interest changes through the LSTM model and combines DQN reinforcement learning to optimize the recommendation strategy to improve user experience and system performance. Experimental results show that collaborative filtering recommendation based on user behavior performs best, with a precision of 0.45, a recall of 0.40, and an F1 value of 0.42, which can better meet personalized needs. Cold start optimization significantly improves the recommendation effect of new users and new products and improves user engagement. The sparsity problem is alleviated by applying SVD and LDA models; the interaction density of new users increases to 0.35, and that of long-term users increases to 0.80; the recommendation accuracy and relevance are greatly improved. The recommendation algorithm based on user collaborative filtering significantly improved the recommendation accuracy and personalization effect. By solving the cold start, sparsity, and real-time recommendation problems, the recommendation of new users and new products is optimized; user participation is enhanced; the platform competitiveness is improved; the accurate dissemination and promotion of cultural products are promoted.