This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
This paper presents CROSS, a novel framework for modeling Temporal Text Attribute Graphs (TTAGs). Conventional Temporal Graph Neural Networks (TGNNs) statically embed textual information and rely on encoding mechanisms that prioritize structural information, thereby overlooking the temporal evolution of textual meaning and the interplay between meaning and structure. CROSS addresses these challenges by decomposing the TTAG modeling process into two steps: temporal semantic extraction and semantic-structural integration. A large-scale language model (LLM) is used to understand the temporally evolving context of a node's textual neighbors, and a temporal semantic extractor extracts dynamic meaning. A semantic-structural joint encoder then integrates semantic and structural information to generate a mutually reinforcing representation. Experimental results demonstrate that CROSS achieves state-of-the-art performance on four public datasets and one industrial dataset.
Takeaways, Limitations
•
Takeaways:
◦
We present CROSS, an effective framework for modeling temporal text attribute graphs (TTAGs).
◦
We present a novel method for dynamically extracting temporal meaning from text using large-scale language models (LLMs).
◦
Enhanced representation learning through mutual reinforcement of semantic and structural information.
◦
Experimentally verified performance improvements over existing methods on various datasets. In particular, significant performance improvements were observed in temporal link prediction and node classification tasks.
•
Limitations:
◦
Potential increase in computational costs due to the use of LLM.
◦
Dependency on a specific LLM. Possible performance degradation when using a different LLM.
◦
Generalization performance verification for various types of TTAGs is needed.
◦
Difficulty in understanding and implementing due to the complexity of the framework.
