基于网络评论的产品质量分析方法研究与系统设计

发布时间：2018-04-22 18:18

本文选题：网络评论 + 情感分析　；参考：《中国计量大学》2016年硕士论文

【摘要】：电子商务、论坛等网络平台,有效拉近了制造企业与消费者之间的距离,企业可以通过用户评论直接了解客户的使用体验,发现产品在使用阶段暴露的质量问题,为产品的质量改进提供方向。本文以网络评论为研究对象,利用特征词提取、情感分析、序列模式挖掘等方法,实现了产品质量水平评估和质量问题发现。本文主要研究内容如下:(1)质量特征词的提取与语义树构建。本文将质量特征词提取看成是序列标注问题,利用条件随机场进行模型训练,在特征方面,采用了词形、词性、上下文、依存关系、支配词、情感判断等特征,实现了质量特征词的自动提取。构建了描述质量特征词之间语义关系的语义树,提出了基于词语相似性的同义子节点添加和基于词语相关性的隶属子节点添加两种语义树扩展方法。(2)产品质量水平评估与质量问题发现。首先提出了基于情感词典的质量特征词情感得分计算方法,然后分析了否定词和程度副词对情感得分的影响,接着基于语义树和情感得分,提出了产品质量水平评估方法。研究了质量否定句的序列模式,提出了基于序列模式挖掘的产品质量问题发现方法。(3)产品质量分析系统的设计。系统由数据抓取、数据预处理、质量特征词提取、语义树构建、质量水平评估和质量问题发现等六个模块组成。
[Abstract]:E-commerce, forums and other network platforms have effectively narrowed the distance between manufacturing enterprises and consumers. Enterprises can directly understand the experience of customers through user reviews and discover the quality problems exposed by products during the use phase. Provide direction for product quality improvement. In this paper, we use the methods of feature word extraction, emotion analysis and sequential pattern mining to realize product quality level evaluation and quality problem discovery. The main contents of this paper are as follows: 1) extraction of quality feature words and construction of semantic tree. In this paper, the quality feature word extraction is regarded as a sequence tagging problem, and the conditional random field is used to train the model. In the feature aspect, it adopts the features of word form, part of speech, context, dependency, dominant word, emotional judgment, and so on. The automatic extraction of quality feature words is realized. A semantic tree is constructed to describe the semantic relationship between quality features. Two semantic tree extension methods, synonymous sub-node addition based on word similarity and subordinate sub-node addition based on word correlation, are proposed to evaluate product quality level and find quality problems. In this paper, an affective score calculation method based on emotion dictionary is proposed, and then the influence of negative words and degree adverbs on emotion score is analyzed. Then, based on semantic tree and emotion score, a product quality evaluation method is proposed. In this paper, the sequential pattern of quality negatives is studied, and the product quality analysis system based on sequential pattern mining is proposed. The system consists of six modules: data capture, data preprocessing, quality feature extraction, semantic tree construction, quality level evaluation and quality problem finding.
【学位授予单位】：中国计量大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【相似文献】