The integration of Graph Neural Networks (GNNs) with Large Language Models (LLMs) has emerged as a promising paradigm for Graph Question Answering (GraphQA). However, effective methods for encoding complex structural information into the LLM’s latent space remain an open challenge. Current state-of-the-art architectures, such as G-Retriever, typically rely on standard GNNs and aggressive mean pooling to compress entire graph substructures into a single token, creating a severe information bottleneck. This work mitigates this bottleneck by investigating two orthogonal strategies: (1) increasing the bandwidth of the graph-to-LLM interface via multi-token pooling, and (2) enhancing the semantic quality of the graph encoder via global attention mechanisms. We evaluate a suite of hierarchical pruning and clustering-based pooling operators—including Top-k, SAGPool, DiffPool, MinCutPool, and Virtual Node Pooling (VNPool) to project graph data into multiple learnable tokens. Empirically, we demonstrate that while pooling introduces significant instability during soft prompt tuning, the application of Low-Rank Adaptation (LoRA) effectively stabilizes these projections, allowing compressed representations to rival full-graph baselines (achieving ∼73% Hit@1 on WebQSP). Conceptually, we demonstrate that a Graph Transformer with VNPool implementation functions structurally as a single-layer Perceiver IO encoder. Finally, we adapt the FandE (Features and Edges) Score to the generative GraphQA domain. Our analysis reveals that current the GraphQA benchmark suffer from representational saturation, where the target answers are often highly correlated with isolated node features. The implementation of our experiments is available at https://anonymous.4open.science/r/Pool-A85D/README.md.

In this paper, we report experiments on Few- and Zero-shot Knowledge Graph completion, where the objective is to add missing relational links between entities into an existing Knowledge Graph with few or no previous examples of the relation in question. While previous work has used pre-trained embeddings based on the structure of the graph as input for a neural network, nobody has, to the best of our knowledge, addressed the task by only using textual descriptive data associated with the entities and relations, much since current standard benchmark data sets lack such information. We therefore enrich the benchmark data sets for these tasks by collecting textual description data to provide a new resource for future research to bridge the gap between structural and textual Knowledge Graph completion. Our results show that we can improve the results for Knowledge Graph completion for both Few- and Zero-shot scenarios with up to a two-fold increase of all metrics in the Zero-shot setting. From a more general perspective, our experiments demonstrate the value of using textual resources to enrich more formal representations of human knowledge and in the utility of transfer learning from textual data and text collections to enrich and maintain knowledge resources.

Advanced NLP models require huge amounts of data from various domains to produce high-quality representations. It is useful then for a few large public and private organizations to join their corpora during training. However, factors such as legislation and user emphasis on data privacy may prevent centralized orchestration and data sharing among these organizations. Therefore, for this specific scenario, we investigate how gossip learning, a massively-parallel, data-private, decentralized protocol, compares to a shared-dataset solution. We find that the application of Word2Vec in a gossip learning framework is viable. Without any tuning, the results are comparable to a traditional centralized setting, with a loss of quality as low as 4.3%. Furthermore, the results are up to 54.8% better than independent local training.