基于多文档摘要的产品评论挖掘技术研究

发布时间：2018-04-19 19:52

本文选题：情感分类 + 评论挖掘　；参考：《东南大学》2015年硕士论文

【摘要】：对产品评论进行挖掘,从中抽取利益方关切信息的技术和方法成为互联网环境下非结构化数据处理研究的重要方向之一。本文对评论挖掘技术进行研究,提出基于产品特征的抽取式摘要算法获得简洁的摘要,再以摘要为基础进行评论分析,从而从评论中抽取有价值的信息。主要贡献如下：1)根据词向量工具word2vec的运行原理,设计了一套同义词筛选规则,通过词向量对种子词典进行扩充,使用筛选规则筛选词向量的扩充结果,可以有效的提高特征词和情感词的同义词典的质量。2)基于依存句法分析技术实现自动化的情感词极性判断。首先利用开源的搜索引擎技术定位语料库中情感词的相关句子,找到与它具有并列关系(COO)的已知极性情感词,以此判定情感词的极性。相比于传统的基于语料库和连词的判断方法,通过COO关系可以捕捉到广泛意义的并列关系,该方法对于情感倾向性判断的准确率很高。3)针对评价句法结构特点,设计基于产品特征的摘要抽取算法,生成简短的具有代表性和可读性的简洁摘要,支持后续的评论句分类。针对评论的用语特点总结出评价搭配模板抽取评价搭配信息,根据评论句中特征词和情感词搭配关系将句子分到相应的类别中。相比于传统摘要算法,本文提出的摘要算法的结果可以更好的作为分类依据。4)为了实现自动化产品评论挖掘,在上述研究基础之上,从用户需求出发,设计并实现基于产品特征的摘要系统,系统可以辅助用户快速获取产品评论中有价值信息。
[Abstract]:The technology and method of mining product reviews and extracting information of interest from the stakeholders has become one of the important research directions of unstructured data processing in the Internet environment. In this paper, we study the technology of comment mining, propose a product feature-based abstract extraction algorithm to obtain concise summary, and then analyze comments based on summary, so as to extract valuable information from comments. The main contributions are as follows: (1) according to the principle of word vector tool word2vec, a set of synonym selection rules is designed. The seed dictionary is expanded by word vector, and the expansion result of word vector is filtered by screening rule. It can effectively improve the quality of synonyms of feature words and affective words. 2) automatic polarity judgment of affective words can be realized based on dependency syntactic analysis technology. Firstly, the open source search engine is used to locate the relevant sentences of affective words in the corpus, and to find the known polar affective words with its paratactic relationship, and then to judge the polarity of affective words. Compared with the traditional corpus-based and conjunction-based judgment methods, the paratactic relationships of a wide range of meanings can be captured by using COO relations, and the accuracy of this method in judging affective preference is high. 3) to evaluate the syntactic structure characteristics. A product feature-based abstract extraction algorithm is designed to generate a brief, representative and readable summary, which supports the subsequent classification of comment sentences. According to the lexical characteristics of comments, this paper summarizes the evaluation collocation template and extracts the evaluation collocation information, and classifies the sentences into the corresponding categories according to the collocation relationship between the feature words and the affective words in the comment sentences. Compared with the traditional summary algorithm, the result of this paper can be used as classification basis. 4) in order to realize automatic product comment mining, based on the above research, the paper starts from the user's demand. A summary system based on product features is designed and implemented. The system can help users to quickly obtain valuable information in product reviews.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.1

【相似文献】