基于注意力机制的酒店评论分类模型

Attention-Based for Hotel Review Classification Model

摘要: 在线评论能够对用户的决策产生重要的影响，一些无良商家会利用这一点通过雇佣水军等方式褒扬自己的商品或诋毁竞争对手的商品，从而影响用户的判断，将利益最大化。为了维护大众消费者的利益，将商品最客观真实的评价展现给用户，为用户提供最可靠的参考，将商品评论分类就显得尤为重要。本文将评论文本和评论发布者特征结合，分别利用融入注意力机制的卷积神经网络模型(ACNN)提取商品评论文本特征和评论发布者特征，综合挖掘其中的信息，从而提高分类的准确率。通过在真实数据集上的多次实验表明，这种方式在评论有效性分类上的准确率达到87.2%，相比只提取评论者特征和只在评论文本中融入注意力机制的分类效果均有提高。

Abstract: Online comments have an important impact on the user’s decision, some unscrupulous merchants will take advantage of this point through hiring people and other ways to praise their own goods or slander competitors’ goods to affect the user’s judgment and maximize the interests. In order to ensure the interests of the public consumers and present the most objective and true evaluation of products to users and provide users with the most reliable reference, it is particularly important to classify product reviews. In view of the above the problem, the author combines the text of comments and the features of comment publishers, respectively uses the convolutional neural networks model (ACNN) with attention mechanism to extract the features of product reviews, so as to improve the accuracy of classification. Several experiments on real dataset show that the accuracy of this method in the classification of comment validity reaches 87.2%, which is better than only extracting the features of reviews and only incorporating the attention mechanism into the comment text.