Sentiment-Polarized Word Embedding for Multi-label Sentiment Classification

Sentiment analysis of text is an import branch of natural language process. In this paper, we propose a sentiment-polarized word embedding model (SPWE) with emotional dictionary, which is a variant of the C&W. Our model is able to represent and differentiate the emotional semantic of words, whic...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2018 IEEE 4th International Conference on Computer and Communications (ICCC) s. 2312 - 2316
Hlavní autoři: Zhang, Liujie, Zhou, Yanquan, Chen, Ruiqi, Duan, Xiuyu
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.12.2018
Témata:
Dictionaries
Feature extraction
Linear programming
multi-label
Semantics
Sentiment analysis
sentiment classification
sentiment-polarized
Task analysis
Training
word embedding
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Sentiment analysis of text is an import branch of natural language process. In this paper, we propose a sentiment-polarized word embedding model (SPWE) with emotional dictionary, which is a variant of the C&W. Our model is able to represent and differentiate the emotional semantic of words, which is critical in sentiment classification tasks. This weak supervised model can perform large scale corpus training just with an open source emotional dictionary. In addition, we optimize the objective function from our previous hierarchical model and propose a Relax Margin loss function. The result shows it can reduce overfitting and improve generalization ability by restricting approximation degree to probability distribution of the multi-labels. We release our manually labeled comment dataset with multi tags from our previous work, and our code is available in https://github.com/KillersDeath/SPWE.
DOI:10.1109/CompComm.2018.8780754