Sentiment analysis in online shopping reviews and film reviews
With the rapid development and popularization of the Internet, a series of online platforms such as Weibo, Facebook, IMDB and Jingdong online shopping mall have provided a stage for people to express themselves freely since the beginning of the 21st century. The Internet generates a large amount of...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis-Master by Coursework |
Language: | English |
Published: |
Nanyang Technological University
2020
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/143415 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | With the rapid development and popularization of the Internet, a series of online platforms such as Weibo, Facebook, IMDB and Jingdong online shopping mall have provided a stage for people to express themselves freely since the beginning of the 21st century. The Internet generates a large amount of text information every day. How to accurately obtain the emotional tendency from a large amount of text information is a hot topic in the field of sentiment analysis in natural language processing. However, the traditional text sentiment analysis has the following disadvantages: (1) The artificially customized feature extraction strategy is too subjective, some hidden emotional information cannot be effectively expressed. (2) The extracted features are too simple to express the dependency of several keywords effectively in the text.
In this dissertation, Gated Recurrent Units model and Hierarchical Attention Mechanism are combined to make an in-depth study on text sentiment analysis. This dissertation mainly includes the following work:
(1) This dissertation briefly introduces the research background and related basic theoretical knowledge of text sentiment analysis, describes the development history of text sentiment analysis technology, especially points out the advantages of deep learning compared with other traditional learning methods. Also, the basic concepts, structures and principles of CNN, RNN, LSTM, GRU and attention mechanism are introduced in details, whose advantages and disadvantages are compared and analyzed in the following chapters. At the same time, it also notes the problems it faces, such as long-distance text dependence.
(2) On the basis of previous studies, for text review data, this dissertation uses a combined gated recurrent unit network model to extract features. Compared with other network models, the advantages of recurrent neural network is that it can better deal with some tasks related to time series.
(3) By comparing the advantages and disadvantages of several attention mechanisms, this dissertation uses Hierarchical Attention Mechanism in the algorithm model, which adds attention structure hierarchically in the process of dividing text into sentences and sentences into words, so as to calculate the weight of each feature vector better. Aiming at the shortcomings of dictionary based and traditional sentiment analysis methods, this dissertation combines bidirectional gated recurrent unit network and hierarchical attention mechanism model to study the sentiment classification of text data. |
---|