基于语义约束LDA的商品特征和情感词提取

发布时间：2019-07-10 12:20

【摘要】：随着网络购物的发展,Web上产生了大量的商品评论文本数据,其中蕴含着丰富的评价知识.如何从这些海量评论文本中有效地提取商品特征和情感词,进而获取特征级别的情感倾向,是进行商品评论细粒度情感分析的关键.根据中文商品评论文本的特点,从句法分析、词义理解和语境相关等多角度获取词语间的语义关系,然后将其作为约束知识嵌入到主题模型,提出语义关系约束的主题模型SRC-LDA(semantic relation constrained LDA),用来实现语义指导下LDA的细粒度主题词提取.由于SRC-LDA改善了标准LDA对于主题词的语义理解和识别能力,从而提高了相同主题下主题词分配的关联度和不同主题下主题词分配的区分度,可以更多地发现细粒度特征词、情感词及其之间的语义关联性.实验结果表明,SRC-LDA对于细粒度特征和情感词的发现和提取具有较好的效果.
[Abstract]:With the development of online shopping, a large number of commodity review text data have been produced on Web, which contains rich evaluation knowledge. How to effectively extract commodity features and emotional words from these massive comment texts, and then obtain the emotional tendency at the feature level, is the key to fine-grain emotional analysis of commodity reviews. According to the characteristics of Chinese commodity review text, the semantic relationship between words is obtained from syntactic analysis, word meaning understanding and context correlation, and then embedded into the topic model as constraint knowledge. The topic model SRC-LDA (semantic relation constrained LDA), which is constrained by semantic relations, is proposed to realize the fine-granularity subject word extraction of LDA under semantic guidance. Because SRC-LDA improves the semantic understanding and recognition ability of standard LDA for subject words, thus improving the correlation degree of subject word allocation under the same topic and the discrimination degree of subject word allocation under different topics, we can find more fine-granularity feature words, emotional words and their semantic relevance. The experimental results show that SRC-LDA has a good effect on the discovery and extraction of fine-granularity features and emotional words.
【作者单位】：江西财经大学信息管理学院;江西师范大学计算机信息工程学院;数据与知识工程江西省高校重点实验室(江西财经大学);
【基金】：国家自然科学基金(61562032,61662032,61662027,61173146,61363039,61363010,61462037,61562031) 江西省自然科学基金重大项目(20152ACB20003) 江西省高等学校科技落地计划(KJLD12022,KJLD14035)~~
【分类号】：TP391.1

【相似文献】