Knowledge representation learning is representing entities and relations in a knowledge graph as dense low-dimensional vectors in the continuous space, which explores the features and properties of the graph. Such a technique can facilitate the computation and reasoning on the knowledge graphs, which benefits many downstream tasks. In order to alleviate the problem of insufficient entity representation learning caused by sparse knowledge graphs, some researchers propose knowledge graph embedding models based on instances and concepts, which utilize the latent semantic connections between concepts and instances contained in the knowledge graphs to enhance the knowledge graph embedding. However, they model instances and concepts in the same space or ignore the transitivity of isA relations, leading to inaccurate embeddings of concepts and instances. To address the above shortcomings, we propose a knowledge graph embedding model that differentiates concepts and instances based on spatial transformation—CIST. The model alleviates the gathering issue of similar instances or concepts in the semantic space by modeling them in different embedding spaces, and adds a learnable parameter to adjust the neighboring range for concept embedding to distinguish hierarchical information of different concepts, thus modeling the transitivity of isA relations. The above features of instances and concepts serve as auxiliary information so that thoroughly modeling them could alleviate the insufficient entity representation learning issue. For the experiments, we chose two tasks, i.e., link prediction and triple classification, and two real-life datasets: YAGO26K-906 and DB111K-174. Compared with state of the arts, CIST achieves an optimal performance in most cases. Specifically, CIST outperforms the SOTA model JOIE by 51.1% on Hits@1 in link prediction and 15.2% on F1 score in triple classification.

Knowledge graph embedding represents entities and relations as low-dimensional continuous vectors. Recently, researchers have attempted to leverage the potential semantic connections between entities with hierarchical relationships in the knowledge graph. However, existing knowledge embedding methods that model entity hierarchical structures face the following challenges: (1) Setting the same embedding dimension for entities at different hierarchical levels can lead to overfitting issues and the waste of storage and computational resources; (2) Manually searching the embedding dimension in discrete space makes it easy to miss the optimal parameters and increases training costs. Therefore, we propose a knowledge embedding method, HEAES, that automatically learns the embedding dimensions for entities based on their hierarchical structure. Specifically, we first propose a new modeling method to capture the hierarchical relationships of entities, and we then train the model on a triple classification task. Then, we adaptively learn the pruning threshold to trim the embedding vectors, automatically learning different embedding dimensions for entities at different hierarchical levels. Experiments on the YAGO26K and DB111K datasets verify that introducing entity hierarchical information helps boost the model performance.

Computing the similarity between graphs is a longstanding and challenging problem with many real-world applications. Recent years have witnessed a rapid increase in neural-network-based methods, which project graphs into embedding space and devise end-to-end frameworks to learn to estimate graph similarity. Nevertheless, these solutions usually design complicated networks to capture the fine-grained interactions between graphs, and hence have low efficiency. Additionally, they rely on labeled data for training the neural networks and overlook the useful information hidden in the graphs themselves. To address the aforementioned issues, in this work, we put forward a contrastive neural graph similarity learning framework, Conga. Specifically, we utilize vanilla graph convolutional networks to generate the graph representations and capture the cross-graph interactions via a simple multilayer perceptron. We further devise an unsupervised contrastive loss to discriminate the graph embeddings and guide the training process by learning more expressive entity representations. Extensive experiment results on public datasets validate that our proposal has more robust performance and higher efficiency compared with state-of-the-art methods.

The traditional model training approach based on negative sampling randomly samples a portion of negative samples for training, which can easily overlook important negative samples and adversely affect the training of knowledge graph embedding models. Some researchers have explored non-sampling model training frameworks that use all unobserved triples as negative samples to improve model training performance. However, both training methods inevitably introduce false negative samples and easy-to-separate negative samples that are far from the model’s decision boundary, and they do not consider the adverse effects of long-tail entities and relations during training, thus limiting the improvement of model training performance. To address this issue, we propose a universal knowledge graph embedding framework based on high-quality negative sampling and weighting, called HNSW-KGE. First, we conduct pre-training based on the NS-KGE non-sampling training framework to quickly obtain an initial set of relatively high-quality embedding vector representations for all entities and relations. Second, we design a candidate negative sample set construction strategy that samples a certain number of negative samples that are neither false negatives nor easy-to-separate negatives for all positive triples, based on the embedding vectors obtained from pre-training. This ensures the provision of high-quality negative samples for model training. Finally, we apply weighting to the loss function based on the frequency of the entities and relations appearing in the triples to mitigate the adverse effects of long-tail entities and relations on model training. Experiments conducted on benchmark datasets FB15K237 and WN18RR using various knowledge graph embedding models demonstrate that our proposed framework HNSW-KGE, based on high-quality negative sampling and weighting, achieves better training performance and exhibits versatility, making it applicable to various types of knowledge embedding models.