融合BiLSTM和注意力机制的卷烟消费者评价情感分类方法

为从海量评价数据中提取消费者对卷烟产品的情感信息，利用词频、点间互信息和左右信息熵提取烟草领域的专有词汇，通过建立分词补充词典提高文本分词准确性；融合双向长短时记忆神经网络和注意力机制建立BiLSTM-Att卷烟消费者评价情感分类模型，基于2006—2021年2 066个卷烟品牌规格消费者评价数据，对BiLSTM-Att模型进行验证并与其他6种分类方法进行对比。结果表明:统一产品名称后BiLSTM-Att模型F1值提高1.78百分点；BiLSTM-Att模型在情感倾向二分类和三分类中的F1值分别为92.89%和80.12%，具有较高准确性。该方法可为卷烟产品研发、精准营销和品牌发展提供支持。 

An emotion classification method for cigarette consumers'evaluation based on combination of BiLSTM and Attention Mechanism

In order to extract consumers'emotion information about cigarette products from massive evaluation data, word frequency, pointwise mutual information, left and right information entropy were used to extract specialized vocabulary in the field of tobacco, and the accuracy of text segmentation was promoted through establishing a text segmentation supplementary dictionary. An emotion classification model, BiLSTM-Att, for cigarette consumers'evaluation was developed by combining Bi-directional Long Short-Term Memory(BiLSTM)with Attention Mechanism. Based on the consumer evaluation data for 2 066 cigarette brands from 2006 to 2021, the developed BiLSTM-Att model was verified and compared with six opted classification methods. The results showed that: the F1 value of the BiLSTM-Att model increased by 1.78 percentage points after unifying product names; the BiLSTM-Att model featured higher accuracy in the binary and ternary emotion classification with F1 values of 92.89% and 80.12% respectively. This method provides a support for cigarette product design, precision marketing and brand development.