改进的谱聚类算法在文本情感分析中的应用

Application of improved spectral clustering algorithm in text sentiment analysis

针对情感分析中使用传统聚类算法所存在的准确率低,聚类方向不确定等问题,文章提出一种将用户反馈机制引入谱聚类算法的新方法。该方法首先由拉谱拉斯矩阵分解得到特征向量,这些特征向量对应于数据不同方面的特征信息,然后让用户对部分特征信息进行检阅并确定自己所需要的聚类方向,最后由系统自动按用户选择进行再聚类,从而得到用户所需要的情感方面分类。实验结果表明,该方法使得聚类结果的准确率获得了一定程度的提高,并解决了聚类方向不确定这一问题。

While using traditional clustering algorithms on sentiment analysis, the accuracy is low and the direction of clustering is uncertain. To solve these problems, this paper proposes a novel incorporating user feedback into the spectral clustering algorithm. Firstly this method takes the top eigenvectors computed from the eigen-decomposition of the Laplacian matrix, which corresponding to the most informative features in different dimensions of the data. And then it lets the user inspect these features and selects his intended clustering dimension. At last, the system clusters all the data along the dimension selected by the user. The experimental results show that the method proposed above is feasible and obtain better performance.