.png){kind=icon}
.png){kind=icon}
Return
A Guided Topic Detection Model Based on Data Augmentation and Feature Representation
.png){kind=icon}
PRE
AI
DOI：10.1109/TNSE.2025.3647512.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In social networks, the accurate detection of guided topics is of great significance for maintaining the healthy order of the network. Aiming at the high-dimensionality of its feature space and the hiddenness of users' emotions, a guided topic detection method based on data enhancement and feature representation is proposed. Firstly, to address the problem of sparse effective data and high-dimensional heterogeneity in the early stage of guided topic, GAN network is introduced to realize homomorphic compensation of data and enhance data diversity. Meanwhile, the SC2vec method is designed to realize the low-rank densification of data. In addition, random wandering is introduced to mine the implicit association network among users and to realize the fusion of multi-dimensional information. Secondly, to address the problem of the hidden nature of users' emotional polarity, the internal attributes and external influences of users are mined. The fine-grained emotional influence factors based on linear multiple regression are constructed. At the same time, the evolutionary game theory is introduced to build an emotional interaction game model between users to reveal the dynamic evolution law of users' emotions. The experimental results show that the method not only successfully realizes the low-rank densification of data and the deep mining of implicit user emotions, but also achieves significant improvement in the accuracy of guided topic detection.
Keywords:
Social networks
representation learning
data compensation
guided topics
evolutionary game
Journal
I
IF:
7.9
Papers:
2.5K
Citations:
10.0K