基于机器学习的文档级情感分析.pdfVIP

中国科技论文在线 基于机器学习的文档级情感分析# 金鑫，张宪超** （大连理工大学软件学院，辽宁大连 116621 ） 5 摘要：情感分析或者舆情挖掘意在识别评论中表达的观点。例如商人们想要知道消费者是否 对他们的产品满意，推荐系统倾向于推荐那些受到很多好评的商品。本文主要研究文档级别 的情感极性分类问题。为了更好的完成该任务，我们把实验分为三个部分。（1 ）从大量未 标注过的句子中自动标注出主观句子和客观句子，进而用标注过的句子们训练句子级别主客 观检测器。（2 ）通过结合第一步中的主客观检测器和句子之间的邻近度信息提出基于最小 10 割的句子级主客观检测器为给定评论创建只包含主观性信息的摘要。（3 ）最后我们在抽取 的摘要上进行分类任务从而提高分类的准确率。 关键词：情感分析；机器学习；最小割 中图分类号：TP181 15 Document-level Sentiment Analysis based on Machine Learning JIN Xin, ZHANG Xianchao (Software School,Dalian University of Technology, LiaoNing DaLian 116621) Abstract: Sentiment analysis or opinion mining aims at identifying viewpoints of reviews which is 20 quite popular in several fields. For example, businessman wants to find out if customers like their products or not and recommend systems tend to recommend items which receives many positive reviews. This paper focuses on document level polarity classification problem. To do this job better, we (1) use un-labeled sentences to create a huge dataset where sentences are annotated as either subjective or not, and the dataset can further be used to train subjectivity detector, (2) create 25 an extract of a given review which only contain subjective information by using graph-cut-based subjectivity detector. （3 ）Then we perform polarity classification algorithm on this extract which can achieve high precision. Keywords: sentiment analysis; machine learning; minimum cut 30 0 引言 由于情感分析很多潜在的应用，对于文本中观点、情感和主客观性的分析和研究得到了 很多的关注。