Research on Subjective Question Automatic Scoring Algorithm based on Multi-pattern Matching and Feature Fusion

Aiming at the current Chinese subjective question automatic scoring technology mostly adopts simple ways such as feature fusion or similarity combination calculation, which leads to large error in subjective question score judgement and low practicality, etc., a subjective question automatic scoring model based on multi-pattern matching and feature fusion is proposed. The model adopts the Aho-Corasick multi-pattern matching algorithm for fast matching of key phrases, designs an edit distance algorithm based on the word segmentation and synonym strategy for word sense matching to calculate the word order similarity, and borrows the cosine algorithm to find the semantic similarity. Further, a feature fusion algorithm is used to consider the three features of key phrase features, word order similarity and semantic similarity to make the final scoring result comprehensively. The experimental results of the comparison between this model and the automatic scoring of subjective questions based on the TF-IDF algorithm show that the scoring results of this model are 8% more accurate than the scoring of the TF-IDF algorithm, the average deviation rate is 0.07 smaller, the maximum deviation rate is reduced by 0.15, which is more close to the manual scoring, and the stability is better.