基于动态特征词的中文句子相似度计算

目的针对当前常用的汉语句子相似度计算方法存在的问题，结合语言习得特点，提出了一种基于动态特征词的中文句子相似度计算方法。方法首先以特征词作为语块切分边界，提取左右语块信息，采用语义向量空间模型；然后计算2个句子对应的左右组块的相似度；最终将各组块的相似度量值加权求和作为2个句子的相似度。结果实验表明，提出的方法计算结果较为理想，与人工判断的相似度较为一致。结论基于动态特征词的中文句子相似度计算方法在常用句式中具有更好的效果。

Computation method for Chinese sentence similarity based on dynamic feature words

Objective--To propose a computation method for Chinese sentence similarities based on dynamic feature words in combination with the feature of language acquisition because there is some shortcomings in the current similarity computation methods of Chinese sentence. Methods--First, the left and right chunks are extracted with the feature words as chunks segmentation boundary. Then the similarities between left and right chunks from the two sentences are calculated using a semantic vec- tor space model. Finally the overall sentence similarity is defined with a combination of these chunk similarities by the weighted parameter. Results--The experiments show that the proposed method with great precision is much close to the similarity of manual judgment, Conclusion--The similarity computation method of Chinese sentence based on dynamic feature words has better performance in common sentence pattern.