基于限制玻尔兹曼机的无极性标注情感分类研究

已有的网络评论情感分类算法都需要手工标注词汇情感倾向,然而网络评论具有表达形式自由、模式多变、词汇更新速度快等特点,手工标注的方式适应能力较低。为解决上述问题,结合限制玻尔兹曼机和相似差异向量运算,通过降低向量相似度,强调其差异性的方式,提出基于限制玻尔兹曼机的无词汇标注情感分类算法。实验表明,该算法虽褒义精确度稍低于支持向量机,但是在贬义精确度上优于支持向量机,并且不需要进行词汇情感倾向标注,降低了算法的复杂度,提高了泛化能力。

Research of Non-polarity Label Sentiment Classification Based on Restricted Boltzmann Machines

Existing web review sentiment classification requires manual annotation emotional tendencies, however, network review has the characteristics of freedom expression, model changeable and vocabulary update speed and so on, the way marked by hand has low adaptive capacity. To solve the above problem, we combine Restricted Boltzmann Machines with similar-difference vector operation which reduce the similarity and emphasize the differences. Then, a kind of no vocabulary label emotional classification algorithm which bases on Restricted Boltzmann Machines has been proposed. Experimental results show that the complimentary sense accurate is slightly lower than support vector machine algorithm, but the derogatory sense accurate is superior than support vector machine algorithm. The algorithm does not require emotional tendencies label which reduces complexity, and improves the generalization ability.