AGTG: Text Sentiment Analysis on Social Network Domain Based on Pre-Training and Regular Optimization Techniques

Zufeng Wu, Jingyou Peng, Ruiting Dai, Songtao Liu

2023

Abstract

The data volume of social networks is very large nowadays, and the information contained in social texts can be used in various application scenarios, especially the information of emotional features revealed in the texts, which is needed to be studied in the fields of opinion control and risk assessment, so text sentiment analysis for social networks is an important research. Many advanced modeling techniques suffer from high arithmetic requirements for large number of parameters, difficulty in extracting keyword information, lack of training data volume leading to overfitting phenomenon and catastrophic forgetting in the fine-tuning phase. Aiming to improve the sentiment classification effect of short texts in the textual task domain of social networks, we propose a new text sentiment analysis algorithm, which contains three important components: 1. Use social networks in the pre-training phase The model is further pre-trained using text data from the task domain in the pre-training phase; 2. A dynamic-static word embedding aggregation method is used to enrich the semantic representation information of the text; 3. The loss function is tuned by adding a trust domain smoothing control adversarial regular optimization method in the fine-tuning phase. Our experiments show that the proposed algorithm achieves new optimal performance in the social network domain.

Paper Citation

in Harvard Style

Wu Z., Peng J., Dai R. and Liu S. (2023). AGTG: Text Sentiment Analysis on Social Network Domain Based on Pre-Training and Regular Optimization Techniques. In Proceedings of the 2nd International Seminar on Artificial Intelligence, Networking and Information Technology - Volume 1: ANIT; ISBN 978-989-758-677-4, SciTePress, pages 338-343. DOI: 10.5220/0012283100003807

in Bibtex Style

@conference{anit23,
author={Zufeng Wu and Jingyou Peng and Ruiting Dai and Songtao Liu},
title={AGTG: Text Sentiment Analysis on Social Network Domain Based on Pre-Training and Regular Optimization Techniques},
booktitle={Proceedings of the 2nd International Seminar on Artificial Intelligence, Networking and Information Technology - Volume 1: ANIT},
year={2023},
pages={338-343},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012283100003807},
isbn={978-989-758-677-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Seminar on Artificial Intelligence, Networking and Information Technology - Volume 1: ANIT
TI - AGTG: Text Sentiment Analysis on Social Network Domain Based on Pre-Training and Regular Optimization Techniques
SN - 978-989-758-677-4
AU - Wu Z.
AU - Peng J.
AU - Dai R.
AU - Liu S.
PY - 2023
SP - 338
EP - 343
DO - 10.5220/0012283100003807
PB - SciTePress