DCGG: drug combination prediction using GNN and GAE

Ziaee, S. Sina; Rahmani, Hossein; Tabatabaei, Mina; Vlot, Anna H. C.; Bender, Andreas

doi:10.1007/s13748-024-00314-3

Cited by 2 publications

(2 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Despite computational resource constraints, our straightforward two-layer GCN has shown promising results, surpassing our baseline model in performance. For future work, we aim to extend our model to inductive settings, enhancing its applicability to unseen data and cover other domains [49]. Additionally, we plan to explore the integration of attention mechanisms to refine classification accuracy [39].…”

Section: Discussionmentioning

confidence: 99%

“…GCNs have also been applied to various Natural Language Processing (NLP) tasks, including semantic role labeling [32], relation classification [33], and machine translation [34], leveraging GCNs to capture the syntactic structures within sentences. For text classification, GNNs have been previously investigated [7,31,39,[49][50][51]. These approaches typically represented documents or sentences as graphs composed of word nodes [35], or they depended on the less commonly available document citation relationships for graph construction [7].…”

Section: Graph Neural Networkmentioning

confidence: 99%

See 1 more Smart Citation

Exploiting Graph Convolutional Networks for Multi-Label Text Classification: A Novel Link Prediction Framework with Heterogeneous Label-Document Graphs

Allahgholi,

Rahmani,

Soltanzadeh

et al. 2024

Preprint

View full text Add to dashboard Cite

Text classification, a fundamental task in natural language processing, has traditionally relied on convolutional neural networks (CNNs). However, the potential of graph convolutional neural networks (GCNs) remains largely untapped. In this study, we explore the application of GCNs to multi-label text classification. Our approach involves constructing a unified text graph that integrates document similarity, document-label associations, and label similarity. The GCN assigns class labels to documents, learns label interrelations, and considers document similarity using the Term Frequency–Inverse Document Frequency (TF-IDF) metric. Additionally, we propose a novel multi-label text classification method by transforming document classification into a link prediction task. In comparison with classic machine learning algorithms which fail to understand the relation between labels our method learns the association between labels. Our empirical findings indicate that our model surpasses base method by 40% with respect to F1 measure and highlight the effectiveness of our proposed approach in tackling multi-label text classification challenges and showcase the importance of considering label relationships in achieving better classification performance.

show abstract

Section: Discussionmentioning

confidence: 99%