The bone sticks excavated from the Han Chang’an City site in Xi’an City, Shanxi Province, have provided valuable information for research into the history of the Western Han Dynasty. Due to the long-term burial and artificial excavation, many of the bone sticks are fractured, resulting in the loss of semantic information, which affects the efficiency of bone stick classification. To improve the efficiency of bone stick classification and imputation, a semantically deficient EWRCA bone stick interpretation classification model is proposed. The model utilizes ERNIE’s 8-layer encoder to capture the deep semantic information of the text, learning from the fractured and incomplete bone stick interpretation data. By integrating the outputs of ERNIE’s multi-layer encoder with the word vectors generated by Word2Vec, the model’s ability to understand the unique vocabulary of bone stick interpretations is enhanced. The combination of the text vector fusion module with the TextRCNN-MHAtt model effectively captures contextual dependencies, strengthening the semantic representation of the text and improving classification accuracy. The fusion attention mechanism is introduced to enhance the accuracy and robustness of the model when processing bone stick interpretations. Experimental results demonstrate that the model achieves classification precision and accuracy of 95.62% and 95.2% for the bone stick interpretations from Han Chang’an City, significantly improving classification accuracy.