传统评价对象抽取方法较少考虑低频、结构复杂和信息量较大的评价对象.为此,针对中文产品评论提出一种完整评价对象抽取方法.利用词性和语法分析结果提取基础词单元,通过每个词单元的完整值和缺失值对其进行扩展,根据连接稳定性判断扩展后的词单元是否合理,得到候选评价对象集合,采用过滤策略对候选评价对象集合进行过滤,获得最终的评价对象集合.在真实数据集上的实验结果表明,该方法对包含低频和分词结构复杂的评价对象有较好的抽取效果.%Traditional opinion target extraction methods ignore the opinion targets with low frequency,complicated structures and large amount of information.Aiming at this problem,an integrated opinion target extraction method for Chinese product reviews is proposed.A basic word unit set is obtained according to part-of-speech and syntax,and it is extended based on its integrity value and deficiency value.A concatenation stability rule is applied to decide whether the extension is reasonable.After above processing,a candidate opinion target set is obtained.The final opinion target set is the result after some unreasonable candidate opinion targets are filtered from the candidate set by filtering strategies.The experimental results on real data sets show that the proposed method works well for the opinion targets with low frequency and complicated structures.
展开▼
