Abstract: This paper proposes an automatic scoring algorithm for Tibetan subjective questions based on multi-features combination, which establishes similarity computing model that consists of keyword-form, word-order, sentences length and semantic of sentence. This algorithm combines computing model with maximum similarity matrix to compute the result between standard answer and student’s answer, in the end, scores automatically. The results of experiment prove, comparing with other algorithm, that this algorithm can effectively reduce the average deviation.

Key words: Tibetan subjective question, multi-features combination, semantic similarity, similarity matrix

摘要： 提出了一种藏语句多特征融合的主观题自动评分算法，构建了关键词词形相似度计算模型、词序相似度计算模型、句子长度相似度计算模型和句子语义相似度计算模型。该算法将计算模型与最大相似度矩阵相结合，计算主观题的标准答案与学生答案之间句子、段落的相似度，最终做出自动评分。实验结果表明，与其他方法比较，该算法能有效降低平均误差值。

关键词: 藏语主观题, 多特征融合, 语义相似度, 相似度矩阵