Research on Data Security Access Detection of Language Technology Resources Using Deep Learning and Long-Range Correlation Analysis

This paper collects sample data of language technology resource queries and access, preprocesses the data, and extracts feature values to construct feature vectors. It employs an improved deep learning method to detect malicious access behaviors of resources, and then conducts long-range correlation analysis (Detrended Fluctuation Analysis, DFA) on the detection results. Experimental results indicate that with the increase of experiment times, support vector machines (SVM), BP neural networks (BPNN), naive Bayes methods, and the improved deep learning method with attention mechanism proposed in this paper exhibit different trends in recall rate, precision rate, and accuracy rate of language technology resource access detection. The experimental method proposed in this paper demonstrates better practical application effects in language technology resource access detection with average recall rate is 89.2%, precision rate is 90.5% and accuracy rate is 87.6%, respectively, and the experimental results also exhibit long-range correlations. Thus, we can effectively monitor the data security access to language technology resources using the method proposed in this paper.