BRFP: An Efficient and Universal Sentence Embedding Learning Model Method Supporting Fused Syntax Combined with Graph Embedding Representation Algorithm

Due to the rapidly growing volume of data on Internet, methods efficiently and accurately processing massive text information have been focus research. In natural language theory, sentence embedding representation is an important method. This paper proposes a new learning model called BRFP (Factorization Process with Bidirectional Restraints) that fuses syntactic information, uses matrix decomposition learn calculates word vectors obtain embedded sentences. experimental chapter, similarity experiments are conducted verify rationality effectiveness analyzed results Chinese English texts current mainstream methods, potential improvement directions summarized. The datasets, including STS, AFQMC, LCQMC, show proposed in this outperforms CNN method terms accuracy F1 value by 7.6% 4.8. comparison experiment vector weighted shows when length longer, or corresponding structure complex, model’s advantages more prominent than TF-IDF SIF methods. Compared method, effect improved 14.4%. it has maximum advantage 7.9%, overall each comparative task between 4 6 percentage points. neural network experiment, compared CNN, RNN, LSTM, ST, QT, InferSent models, significantly 14’OnWN, 14’Tweet-news, 15’Ans.-forum datasets. For example, 14’OnWN dataset, 10.9% over ST 14’Tweet-news dataset 22.9% LSTM 24.07% RNN article also demonstrates generality model, proving universal framework.