摘要
在传统基于关键词属性、情感属性和位置属性提取关键句的文本情感倾向性研究的基础上,提出一种融合全局特征和自身特征双窗口的加权TextRank关键句提取算法(WTTW算法),使用soft_voting对提取的关键句进行情感倾向性分析的方法。从全局特征出发通过关键词特征、位置特征、句子之间的相似度加权求和构建窗口为2的TextRank图模型,即将整个文本作为一个单元,设置长度为2的滑动窗口,从第一句至最后一句顺序进行滑动窗口建立图模型,迭代得到各句子的得分;再根据句子情感特征和标点特征对句子得分进行调整,得到关键句;使用soft_voting对提取的关键句进行情感倾向性分析。在四个不同领域进行实验,实验结果表明,该方法在各种评价指标下均显著优于baseline,具有高效性。
The traditional text sentiment orientation research is based on the key sentence extraction of keyword attributes, sentiment attributes and location attributes. On the basis of that, this paper puts forward the weighted TextRank key sentences extraction algorithm with two windows(WTTW algorithm) which combine the global feature and its own feature. We used soft_voting to analyze the sentiment orientation of the extracted key sentences. The TextRank graph model with two windows was constructed from the global feature through the weighted sum of keyword features, positional features and similarity between sentences. We took the entire text as a unit, set the sliding window with a length of two, proceeded from the first sentence to the last sentence in the order of the sliding window to build a graph model, and iteratively got the score of each sentence. According to the emotional feature and punctuation feature of the sentence, the scores of the sentence was adjusted to get the key sentences. We adopted soft_voting to analyze the sentiment orientation of the extracted key sentences. We carried out the experiments across four domains. The results show that the method is significantly better than the baseline in all the evaluation indicators, which has high efficiency.
作者
宛艳萍
张芳
谷佳真
Wan Yanping;Zhang Fang;Gu Jiazhen(School of Artificial Intelligence and Data Science,Hebei University of Technology,Tianjin 300401,China)
出处
《计算机应用与软件》
北大核心
2022年第4期242-248,共7页
Computer Applications and Software
