Abstract: 【Purpose/significance】In libraries, with diverse information carriers, differences in data structures, metadata formats and storage methods, and similar resource scenarios, traditional library management systems storing structured data have single semantic expressions. Focusing only on content features and extracting semantic features via a single channel, they can′t meet users′ search hab⁃ its, leading to low matching of book information recommendation and user satisfaction. Thus, a new recommendation method is needed. 【Method/process】A library multi-carrier information recommendation method via semantic dual-channel feature extraction is pro⁃ posed. Firstly, adopt the semantic dual-channel feature extraction strategy to extract features of library multi-carrier information con⁃ tent and user demands. Introduce the attention mechanism into the convolutional neural network to create an ATTCNN model. After capturing local features in the convolutional layer, the attention layer selectively focuses on relevant parts of input data by learning weight distribution to extract information content features. Map data to a high-dimensional feature space with a kernel function and conduct clustering to find data′s internal structure and distribution patterns for extracting user demand features. Secondly, use the Apriori algorithm to obtain the matching relationship between them, and screen out information meeting users′ personalized needs based on strong correlations for intelligent recommendation.【Result/conclusion】Experiments demonstrate that when the ATTCNN model and kernel clustering algorithm are used for feature extraction, the F1 score stays above 0.9, the matching degree of recom⁃ mended information exceeds 92%, and user satisfaction is over 89%. This proves the method can precisely recommend information and boost user satisfaction.【Innovation/limitation】The innovation is using a semantic dual-channel feature extraction strategy and in⁃ tegrating multiple algorithms for accurate recommendation.