Research on Entity Relation Extraction Based on BiLSTM-CRF Classical Probability Word Problems

Mathematical automatic problem solving is a very challenging task in the field of artificial intelligence. The key premise of problem-solving automatically is to understand the problem's meaning. For mathematical word problems with rich semantics, varied forms, and difficult to be understood by machines, this study focuses on solving the difficulty of overlapping entity relations recognition and multiple entity relation extractions across sentences, taking the word problem of the classical probability as the research object, an entity relations extraction method based on sequence annotation is proposed. The BiLSRM-CRF model is used to improve the effect of question comprehension. The experimental study found that, compared with the selected combination of different features based on the CRF model alone, the BiLSTM-CRF model can obtain the effect of the approximate CRF model at a superior cost, and improve the recognition effect of a few relations. Meanwhile, the accuracy of the overall problem understanding also gets improved.