TY - JOUR
T1 - Automated Social Text Annotation with Joint Multilabel Attention Networks
AU - Dong, Hang
AU - Wang, Wei
AU - Huang, Kaizhu
AU - Coenen, Frans
N1 - Funding Information:
Manuscript received August 30, 2019; revised March 5, 2020; accepted June 11, 2020. Date of publication June 25, 2020; date of current version May 3, 2021. This work was supported in part by the Research Development Fund at Xi’an Jiaotong-Liverpool University (XJTLU) under Contract RDF-14-01-10, in part by the National Natural Science Foundation of China under Grant 61876155, in part by the Natural Science Foundation of Jiangsu Province under Grant BK20181189, and in part by the Key Program Special Fund in XJTLU under Grant KSF-A-01, Grant KSF-T-06, Grant KSF-E-26, Grant KSF-P-02, and Grant KSF-A-10. The work of Hang Dong was supported by the Human Phenotype Project in Health Data Research UK Scotland. (Corresponding author: Wei Wang.) Hang Dong is with the Department of Computer Science, University of Liverpool, Liverpool L69 7ZX, U.K., also with the Department of Computer Science and Software Engineering, Xi’an Jiaotong-Liverpool University, Suzhou 215123, China, and also with the Centre for Medical Informatics, Usher Institute, The University of Edinburgh, Edinburgh EH16 4UX, U.K. (e-mail: hangdong@liverpool.ac.uk).
Publisher Copyright:
© 2012 IEEE.
PY - 2021/5
Y1 - 2021/5
N2 - Automated social text annotation is the task of suggesting a set of tags for shared documents on social media platforms. The automated annotation process can reduce users' cognitive overhead in tagging and improve tag management for better search, browsing, and recommendation of documents. It can be formulated as a multilabel classification problem. We propose a novel deep learning-based method for this problem and design an attention-based neural network with semantic-based regularization, which can mimic users' reading and annotation behavior to formulate better document representation, leveraging the semantic relations among labels. The network separately models the title and the content of each document and injects an explicit, title-guided attention mechanism into each sentence. To exploit the correlation among labels, we propose two semantic-based loss regularizers, i.e., similarity and subsumption, which enforce the output of the network to conform to label semantics. The model with the semantic-based loss regularizers is referred to as the joint multilabel attention network (JMAN). We conducted a comprehensive evaluation study and compared JMAN to the state-of-the-art baseline models, using four large, real-world social media data sets. In terms of F_{1} , JMAN significantly outperformed bidirectional gated recurrent unit (Bi-GRU) relatively by around 12.8%-78.6% and the hierarchical attention network (HAN) by around 3.9%-23.8%. The JMAN model demonstrates advantages in convergence and training speed. Further improvement of performance was observed against latent Dirichlet allocation (LDA) and support vector machine (SVM). When applying the semantic-based loss regularizers, the performance of HAN and Bi-GRU in terms of F_{1} was also boosted. It is also found that dynamic update of the label semantic matrices (JMANd) has the potential to further improve the performance of JMAN but at the cost of substantial memory and warrants further study.
AB - Automated social text annotation is the task of suggesting a set of tags for shared documents on social media platforms. The automated annotation process can reduce users' cognitive overhead in tagging and improve tag management for better search, browsing, and recommendation of documents. It can be formulated as a multilabel classification problem. We propose a novel deep learning-based method for this problem and design an attention-based neural network with semantic-based regularization, which can mimic users' reading and annotation behavior to formulate better document representation, leveraging the semantic relations among labels. The network separately models the title and the content of each document and injects an explicit, title-guided attention mechanism into each sentence. To exploit the correlation among labels, we propose two semantic-based loss regularizers, i.e., similarity and subsumption, which enforce the output of the network to conform to label semantics. The model with the semantic-based loss regularizers is referred to as the joint multilabel attention network (JMAN). We conducted a comprehensive evaluation study and compared JMAN to the state-of-the-art baseline models, using four large, real-world social media data sets. In terms of F_{1} , JMAN significantly outperformed bidirectional gated recurrent unit (Bi-GRU) relatively by around 12.8%-78.6% and the hierarchical attention network (HAN) by around 3.9%-23.8%. The JMAN model demonstrates advantages in convergence and training speed. Further improvement of performance was observed against latent Dirichlet allocation (LDA) and support vector machine (SVM). When applying the semantic-based loss regularizers, the performance of HAN and Bi-GRU in terms of F_{1} was also boosted. It is also found that dynamic update of the label semantic matrices (JMANd) has the potential to further improve the performance of JMAN but at the cost of substantial memory and warrants further study.
KW - Attention mechanisms
KW - automated social annotation
KW - deep learning
KW - multilabel classification
KW - recurrent neural networks (RNNs)
UR - http://www.scopus.com/inward/record.url?scp=85105553414&partnerID=8YFLogxK
U2 - 10.1109/TNNLS.2020.3002798
DO - 10.1109/TNNLS.2020.3002798
M3 - Article
C2 - 32584774
AN - SCOPUS:85105553414
SN - 2162-237X
VL - 32
SP - 2224
EP - 2238
JO - IEEE Transactions on Neural Networks and Learning Systems
JF - IEEE Transactions on Neural Networks and Learning Systems
IS - 5
M1 - 9126211
ER -