A personalized design knowledge recommendation service architecture was proposed aiming at the problems of low efficiency of knowledge resource retrieval and utilization and imbalance of effectiveness and accuracy of knowledge recommendation results under the task of personalized product design. The framework combined the product design scene characteristics to determine the design knowledge categories based on the product design literature data and manual data. A new deep learning model ALBERT-BiLSTM-IDCNN-CRF was adopted to effectively extract 9 types of knowledge elements contained in the data, including design objects, design tasks, design performance, design methods, design tools, design principles, design parameters, calculation formulas and design drawings. A five-dimensional design knowledge association model was established by combining the knowledge elements of designers composed of internal employees and literature authors. A complete product design knowledge base was established by using ontology and inference rules based on the model. A design knowledge recommendation service mode based on inference-context-aware activation model was proposed. The hydraulic press design was taken as an example to verify. Results show that the proposed deep learning model achieves better results than other benchmark models in design knowledge extraction. The proposed knowledge recommendation service mode greatly improves the effectiveness of outputting recommendation results while maintaining a high accuracy of the knowledge recommendation results. Required solution knowledge resources can be quickly and accurately recommended for personalized design tasks under specific object attributes, task attributes and performance requirement attributes.