Current recognition methods utilizing deep learning usually rely on a large amount of training data in their execution. However, in the Chinese patent field, where training data is relatively scarce, the performance of these methods is often unsatisfactory. In order to solve this problem, this paper proposes a patent technology recognition method based on the fusion of dependency parsing and ALBERT-BiLSTM-CRF model. The method learns the expressions of Chinese patents through dependency grammar analysis, acquires sentence dependencies as new features, and combines the ALBERT-BiLSTM-CRF pre-training model with feature fusion and auxiliary training to help the model better understand the sentence structure, thus improving the accuracy of patent technology recognition. After the empirical study, this patent technology recognition model achieves 90.23% accuracy, 88.67% recall and 89.44% F1-Score, which is a higher performance compared with the benchmark model. It proves the effectiveness of the method in patent technology recognition, which can provide certain reference for the subsequent patent analysis work.