紀要論文

Reputation Analysis Based on Weakly-Supervised Bi-LSTM-Attention Network

eng

24368083

https://doi.org/10.15002/00025364

法政大学大学院理工学研究科

法政大学大学院紀要. 理工学研究科編

63

1

12

2022-03-24

Version of Record

Due to the emergence of rapid, mass-produced information in the Web2.0 era, a large amount of weakly labeled information (star ratings, etc.) has been widespread. The Weakly-Supervised Deep Embedding (WDE) model is a good choice for utilizing this kind of data. The ratings are treated as weakly-supervised signals for pre-training, fine tuning the whole model with a small amount of manually labeled data. In this research, we proposed to change the original unidirectional transmission into bidirectional in the LSTM layer to capture the semantics in both directions, and an attention mechanism is introduced, which is helpful to capture the important information in the context and improve the accuracy of sentiment classification. Finally, we use TF-IDF and LDA topic models to mine the review topics and extract the consumers’ opinions on different sentiment polarities.

Article