机构地区: 北京信息科技大学计算中心,北京100192
出 处: 《智能系统学报》 2017年第4期498-503,共6页
摘 要: 针对目前观点分析方法局限于传统的文本分析技术,只能将舆论观点句分为肯定和否定两极或确定每一极的程度(粗粒度),不能进一步给出舆论观点句是积极的还是消极的程度的问题。本文从认知学角度研究细粒度语义情感计算框架。提出了一种舆情观点句的定量分析方法,该方法将对于某话题的文本集合作为输入,输出一个实数表示文本中所表达观点的能量。本文在NLPIR共享平台上进行了相关实验,给出了粗粒度情感和细粒度情感对观点句识别的对比实验,实验表明,两种方法对观点句的识别性能相差不大;对非观点句细粒度方法好于粗粒度方法。 The current viewpoint analysis method is limited to the traditional text analysis technology, whereby a public opinion sentence can only be divided into positive and negative poles and the extent of each pole (coarse- grained) determined. It is difficult to determine whether a public opinion sentence is active or passive. In this paper, we discuss a computation framework for fine-grained semantic sentiments from the cognitive science viewpoint and propose a quantitative analysis method for public opinion sentences. This method takes the text collection of some topic as input and uses a real number to represent the energy of a viewpoint in the text. We conducted an experiment using the Natural Language Processing and Information Retrieval (NLPIR) sharing platform and a contrasting experiment with respect to view recognition by comparing coarse-grained and fine-grained affects. The experimental results show that the two methods have the same recognition performance regarding sentence viewpoints. For no-opinion sentences, the fine-grained method performs better than the coarse-grained method.