On Automatic Scoring Algorithm for Subjective Questions based on LSA and Dependency Parsing

Aiming at the problems of lack of sample data and unknown words often seen in the development of automatic scoring system for subjective questions, we propose an automatic scoring algorithm based on latent semantic analysis and dependency parsing for subjective questions. First, the questions, standard answers and candidates' answers are used as corpus for latent semantic analysis training to generate document vectors and word vectors. Then, the standard answers and candidates' answers are divided into sentences, and the sentences are analysed by dependency parsing. Then, the document vectors, word vectors and dependency parsing results are used as the input of the automatic scoring algorithm, three automatic scoring schemes are proposed according to the number of sentences and their differences between the standard answer and the candidates' answer. The experimental results show that this algorithm has obvious advantages and potential value compared with the control method.