Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 27 of the 27 references it cites were found in a registry or an index; none of the 4 sources its claims cite is checked for support yet.

Temporal Graph Attention Network for Spatio-Temporal Feature Extraction in Research Topic Trend Prediction

Open access
Guo, Zhan, Lu, Mingxin, Han, Jin · 2025 · Mathematics

Abstract

Comprehensively extracting spatio-temporal features is essential to research topic trend prediction. This necessity arises from the fact that research topics exhibit both temporal trend features and spatial correlation features. This study proposes a Temporal Graph Attention Network (T-GAT) to extract the spatio-temporal features of research topics and predict their trends. In this model, a temporal convolutional layer is employed to extract temporal trend features from multivariate topic time series. Additionally, a multi-head graph attention layer is introduced to capture spatial correlation features among research topics. This layer learns attention scores from the data by using scaled dot product operations and updates edge weights between topics accordingly, thereby mitigating the issue of over-smoothing. Furthermore, we introduce WFtopic-econ and WFtopic-polit, two domain-specific datasets for Chinese research topics constructed from the Wanfang Academic Database. Extensive experiments demonstrate that T-GAT outperforms baseline models in prediction accuracy, with RMSE and MAE being reduced by 4.8% to 7.1% and 14.5% to 18.4%, respectively, while R2 improved by 4.8% to 7.9% across varying observation time steps on the WFtopic-econ dataset. Moreover, on the WFtopic-polit dataset, RMSE and MAE were reduced by 4.0% to 5.3% and 10.0% to 10.7%, respectively, and R2 improved by 7.6% to 14.4%. These results validate the effectiveness of integrating graph attention with temporal convolution to model the spatio-temporal evolution of research topics, providing a robust tool for scholarly trend analysis and decision making.

Overview

CoChat Summary

TL;DR

The paper proposes a Temporal Graph Attention Network (T-GAT) that combines dilated temporal convolution with multi-head scaled dot-product graph attention to jointly extract temporal trend and spatial correlation features of research topics, and uses it to predict topic popularity trends. Experiments on two new Chinese-language datasets (WFtopic-econ, WFtopic-polit) show T-GAT outperforming LSTM, GRU, T-GCN, and Graph WaveNet baselines on RMSE, MAE, and R².

Key findings

  • On WFtopic-econ, T-GAT reduced RMSE by 4.8%–7.1% and MAE by 14.5%–18.4%, and improved R² by 4.8%–7.9% relative to baselines across varying observation time steps; on WFtopic-polit, RMSE fell 4.0%–5.3%, MAE 10.0%–10.7%, and R² rose 7.6%–14.4%.
  • Absolute T-GAT results: WFtopic-econ RMSE 105.32–113.63, MAE 62.47–67.04, R² 0.612–0.618; WFtopic-polit RMSE 86.209–97.213, MAE 51.820–58.256, R² 0.438–0.465 (Table 2).
  • Spatio-temporal graph models (T-GAT, T-GCN, Graph WaveNet) had RMSE 10.75–43.16 lower than LSTM/GRU on WFtopic-econ, and Graph WaveNet's MAE was 6.14–14.60 lower than T-GCN's, indicating convolutional temporal extraction outperformed recurrent extraction.
  • T-GAT converged fastest (31 epochs on WFtopic-econ, 45 on WFtopic-polit) but had the longest per-epoch training time (0.054 s and 0.035 s); t-tests against Graph WaveNet yielded all p-values < 0.05.

Methods

  • Design: T-GAT stacks spatio-temporal blocks, each with a dilated causal temporal convolutional layer (gating removed, weight normalization and residual connections added) and a multi-head graph attention layer using scaled dot-product attention with adjacency-based masking; outputs are averaged across 4 heads, followed by two fully connected layers.
  • Data: Two datasets built from Wanfang Chinese Academic Database metadata — WFtopic-econ (40,254 economics papers, 1999–2021; 367 nodes, 6212 edges, time span 1991–2021) and WFtopic-polit (146 nodes, 1932 edges). Topics are keywords appearing ≥20 times; edges are keyword co-occurrence; node features are annual citation counts.
  • Analysis: Sliding-window setup (observation window 4, prediction window 1; also tested at 6 and 8 steps), 60/40 chronological train-test split, MSE loss, Adam optimizer (lr 1×10⁻³, weight decay 1×10⁻⁵), hidden dimension 32, kernel size 2, up to 3000 epochs with early stopping; evaluated with RMSE, MAE, and R².

Limitations

  • Stated by the authors: the model's training time is relatively long, and its ability to extract correlation features still requires improvement; they plan to address both in future work, including investigating logarithmic quantization.
  • Evident from the text: the datasets are limited to two Chinese-language domains (economics and politics) from a single database, so generalization to other disciplines or languages is untested; the paper reports no ablation isolating the contribution of the scaled dot-product attention versus the temporal convolution alone.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/math13050686

Cites

27 of 27 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Temporal convolution with a kernel size of 2. Figure 1. Temporal convolution with a kernel size of 2.
Figure 1 Temporal convolution with a kernel size of 2. Figure 1. Temporal convolution with a kernel size of 2.CC-BY · From the PDF
Figure 2 Multi-head graph a tt ention mechanism based on scaled dot product. Figure 2. Multi-head graph attention mechanism based on scaled dot product.
Figure 2 Multi-head graph a tt ention mechanism based on scaled dot product. Figure 2. Multi-head graph attention mechanism based on scaled dot product.CC-BY · From the PDF
Figure 3 Architecture of Temporal Graph A tt ention Network. Figure 3. Architecture of Temporal Graph Attention Network.
Figure 3 Architecture of Temporal Graph A tt ention Network. Figure 3. Architecture of Temporal Graph Attention Network.CC-BY · From the PDF
Figure 4 Performance variation across different observation time steps: ( a ) Mean Absolute Error on WFtopic-econ; ( b ) R-squared on WFtopic-econ; ( c ) Mean Absolute Error on WFtopic-polit; ( d ) R-squared on WFtopic-polit.
Figure 4 Performance variation across different observation time steps: ( a ) Mean Absolute Error on WFtopic-econ; ( b ) R-squared on WFtopic-econ; ( c ) Mean Absolute Error on WFtopic-polit; ( d ) R-squared on WFtopic-polit.CC-BY · From the PDF
Figure 5 Comparison of prediction curves between T-GAT and Graph WaveNet for four research topics: ( a ) investment in research and development; ( b ) supply chain coordination; ( c ) earnings management; ( d ) organization pattern.
Figure 5 Comparison of prediction curves between T-GAT and Graph WaveNet for four research topics: ( a ) investment in research and development; ( b ) supply chain coordination; ( c ) earnings management; ( d ) organization pattern.CC-BY · From the PDF
Figure 6 The correlation features identified by the graph a tt ention mechanism: ( a ) Heatmap representing the adjacency matrix of the top 50 research topics. ( b ) Heatmap representing the a tt ention matrix of the top 50 research topics. ( c ) Actual trends observed across multiple topics. Figure 6. The correlation features identified by the graph attention mechanism: ( a ) Heatmap representing the adjacency matrix of the top 50 research topics. ( b ) Heatmap representing the attention matrix of the top 50 research topics. ( c ) Actual trends observed across multiple topics.
Figure 6 The correlation features identified by the graph a tt ention mechanism: ( a ) Heatmap representing the adjacency matrix of the top 50 research topics. ( b ) Heatmap representing the a tt ention matrix of the top 50 research topics. ( c ) Actual trends observed across multiple topics. Figure 6. The correlation features identified by the graph attention mechanism: ( a ) Heatmap representing the adjacency matrix of the top 50 research topics. ( b ) Heatmap representing the attention matrix of the top 50 research topics. ( c ) Actual trends observed across multiple topics.CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 5 Not valid · 5 Not checked · 55 Own results and interpretations

40 more claims

Create a free account to see all 65 claims and to verify the 4 references still open.

See all claims — free account
  • It may state two separate claims as one. 11
  • The quote only partly states this claim. 2
  • Probably not a substantial claim. 12

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 26 references

Research works printed but not linked: 2

  1. Box, G.E., Jenkins, G.M., Reinsel, G.C., and Ljung, G.M. (2015). Time Series Analysis: Forecasting and Control, John Wiley & Sons.
  2. Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., and Bengio, Y. (May, January 30). Graph Attention Networks. Proceedings of the International Conference on Learning Representations, Vancouver, BC, Canada.

Cited by

Metadata

Authors
Guo, Z., Lu, M., & Han, J.
Affiliations
  • Nanjing University of Information Science and Technology (CN)
  • Nanjing University (CN)
Venue
Mathematics
Publisher
MDPI AG
Published
2025-02-20
Volume
13
Issue
5
Pages
686
ISSN
2227-7390 (ISSN-L)
OpenAlex
W4407782414
Cited by
11
References
26
Citations by year
2026: 8 · 2025: 3
FWCI
2.10
Type
journal article
Language
en
Open access
Locations
  • Mathematics (publishedVersion, cc-by)
  • Preprints.org (submittedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
“Design, Evaluation, and Research &amp; Development of Lightweight Cryptographic Systems for the Application Scenarios of Blockchain” (SYC2022093)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Advanced Graph Neural Networks
  • Social Sciences › Transportation › Traffic Prediction and Management Techniques
  • Physical Sciences › Computer Science › Information Systems › Recommender Systems and Techniques
Keywords
spatiotemporal feature extraction, temporal graph attention network, graph attention networks, temporal convolutional network, over-smoothing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.