Research on Chinese Elementary Geometry Text Matching Based on Self-Attention and Bi-LSTM Network

Text matching can assist computers in understanding and processing large amounts of textual data, thereby providing more intelligent and personalized services. Addressing issues such as missing traditional machine learning mathematical pre- training data and incomplete semantic extraction of mathematical formula features, a Chinese mathematical plane geometry text matching model based on self-attention mechanism and bidirectional long short-term memory networks (SABi-LSTM) is proposed. The model constructs a parsing tree for mathematical expressions, divides semantics, and establishes a Bi-LSTM network for feature extraction. The performance and timeliness of the model's network structure are evaluated, optimizing the Bi-LSTM layers and hidden layer nodes of the proposed network to determine the optimal network structure. A self-attention mechanism is introduced for optimizing weight parameters to reduce the impact of noise data. Experimental results indicate that compared to traditional machine learning algorithms, the Bi-LSTM network with the introduced self-attention mechanism improves the accuracy of Chinese mathematical text matching by an average of 0.55%. After further optimizing the construction of word vectors, the recognition accuracy increases by an average of 0.34%, resulting in a final accuracy of 89.49%. The proposed model effectively determines the matching degree of Chinese elementary geometry text.