中文在线评论的产品特征与观点识别:跨领域的比较研究

发布时间：2018-03-28 20:32

本文选题：在线评论　切入点：中文语境　出处：《管理工程学报》2017年04期

【摘要】：产品特征及观点的识别是细粒度情感分析的重要任务。但是,现有识别算法对中文语境下不同评论领域的适应性尚无定论,算法的鲁棒性也不理想,难以实现跨领域的算法移植。为此,选取词频统计方法、规则匹配、关联规则挖掘、具有句法格式的关联规则、CRF和SVM等6种代表性的识别算法,结合中文在线评论的语言特点,对上述算法引入到中文评论的文本分析中,根据准确率、召回率和F值指标,分析比较统计方法和机器学习方法在产品特征及观点识别上的性能。选择数码相机评论、化妆品评论、书评、酒店评论、影评、手机评论和餐厅评论7类语料3646条评论,分别采用6种算法进行产品特征和观点的抽取。实验表明,不同领域下的特征抽取难度是存在差异的;不同算法适应于不同领域;评论的文本长度对识别准确率和召回率有显著影响;另外,总体上机器学习的算法性能显著高于统计学方法。
[Abstract]:Recognition of product features and viewpoints is an important task in fine-grained emotional analysis. However, there is no conclusion on the adaptability of existing recognition algorithms to different comment areas in Chinese context, nor is the robustness of the algorithms ideal. It is difficult to realize cross-domain algorithm transplantation. Therefore, six representative recognition algorithms, such as word frequency statistic method, rule matching, association rule mining, association rule recognition algorithm with syntactic format and SVM, are selected, and combined with the language characteristics of Chinese online comment. The algorithm is introduced into the text analysis of Chinese comments. According to the accuracy rate, recall rate and F value index, the performance of statistical method and machine learning method in product feature and viewpoint recognition is analyzed and compared. Cosmetic reviews, book reviews, hotel reviews, film reviews, mobile phone reviews and restaurant reviews of seven corpus 3646 comments, using 6 algorithms to extract product features and views. The difficulty of feature extraction in different fields is different; different algorithms are suitable for different fields; the length of text of comments has a significant impact on the recognition accuracy and recall rate; in addition, Overall, the performance of machine learning algorithms is significantly higher than that of statistical methods.
【作者单位】：华侨大学工商管理学院;同济大学经济与管理学院;
【基金】：国家自然科学基金资助项目(70971099、71371144、71402121) 上海市哲学社会科学规划课题一般项目(2013BGL004)
【分类号】：TP391.1

【相似文献】