Logistics Service Quality Sentiment Analysis with Deeper Attention LSTM Model with Aspect Embedding

Abstract

To understand the audience's subjective perception of quality of service (QoS), it is important to analyze the data acquired from the logistics service logs and online evaluation system reasonably and effectively. Based on the analysis, rational improvement measures and decision suggestions can be developed to enhance the QoS. However, modern logistics service departments often face various business needs and service objects at the same time. If the evaluation subjects and their relationships are unclear in the service evaluation data, the sentiment analysis result of the text is a coarse-grained evaluation of the service as a whole. The lack of fine-grained pertinent evaluation results will hinder the improvement of specific management measures. To solve the problem, this paper designs an attention-based long short-term memory network (AT-LSTM) to divide the service reviews into ten topic relations, and then builds a deeper attention LSTM with aspect embedding (AE-DATT-LSTM). The weight-sharing bidirectional LSTM (BiLSTM) trains the topic word vectors and the text word vectors, and fuses the resulting topic features and text features. After the processing of the deep attention mechanism, the sentiment class of each evaluation topic is obtained by the classifier. Finally, several experiments were carried out on different public datasets. The results show that our approach surpasses the previous attention-based sentiment analysis models in accuracy and stability of service quality sentiment analysis. The introduction of topic features and deep attention mechanism is of great significance for the QoS-based sentiment classification b, and provides a feasible method for other fields like public opinion analysis, question answering system, and text reasoning.