限定领域口语对话系统中的商品属性抽取

发布时间：2018-06-25 02:09

本文选题：商品属性抽取 + 词向量　；参考：《中文信息学报》2016年06期

【摘要】：按功能或问题域划分,商品属性抽取(product feature mining)在限定领域的对话系统中属于口语语言理解(spoken language understanding,SLU)的范畴。商品属性抽取任务只关注自然文本中描述商品属性的特定部分,它是细粒度观点抽取(fine-grained opinion mining)的一个重要的子任务。现有的商品属性抽取技术主要建立在商品的评论语料上,该文以手机导购对话系统为背景,将商品属性抽取应用到整个对话过程中,增强对话系统应答的针对性。使用基于CBOW(continuous bag of words)语言模型的word2vector(W2V)对词汇的语义层面建模,提出一个针对口语对话的指数型变长静态窗口特征表达框架,捕捉不同距离词语组合的重要特征,使用卷积神经网络(convolutional neural network,CNN)结合词汇的语义和上下文层面对口语对话语料中的商品属性进行抽取。词嵌入模型给出了当前词和所给定的属性类别是否存在相关性的证据,而所提出的特征表达框架则是为了解决一词多义的问题。实验结果表明,该方法取得了优于研究进展中方法的商品属性识别效果。
[Abstract]:According to function or problem domain, commodity attribute extraction (product feature mining) belongs to the category of colloquial language understanding (spoken language) in a restricted domain dialogue system. The task of commodity attribute extraction is only concerned with the specific part of the natural text that describes the property of the commodity. It is an important sub-task of fine-grained point of view extraction (fine-grained opinion mining). The existing commodity attribute extraction technology is mainly based on the commodity comment corpus. In this paper, the commodity attribute extraction is applied to the whole dialogue process in order to enhance the pertinence of the dialogue system response. Word2vector (W2V) based on CBOW (continuous bag of words) language model is used to model the semantic level of vocabulary, and an exponential variable length static window feature representation framework for oral dialogue is proposed to capture the important features of different distance word combinations. A convolutional neural network (convolutional neural Network CNN) is used to extract the commodity attributes from the spoken dialogue corpus in conjunction with the semantic and contextual aspects of the vocabulary. The word embedding model gives evidence of the correlation between the current word and the given attribute category, and the proposed feature representation framework is designed to solve the polysemous problem of the word. The experimental results show that this method is better than the method in the research progress in the recognition of commodity attributes.
【作者单位】：华南农业大学数学与信息学院;
【基金】：国家自然科学基金(71472068) 广东省大学生科技创新培育专项项目(pdjh2016b0087)
【分类号】：TP391.1

【相似文献】