Few-shot knowledge graph completion (FKGC), which aims to infer new triples for a relation using only a few reference triples of the relation, has attracted much attention in recent years. Most existing FKGC methods learn a transferable embedding space, where entity pairs belonging to the same relations are close to each other. In real-world knowledge graphs (KGs), however, some relations may involve multiple semantics, and their entity pairs are not always close due to having different meanings. Hence, the existing FKGC methods may yield suboptimal performance when handling multiple semantic relations in the few-shot scenario. To solve this problem, we propose a new method named adaptive prototype interaction network (APINet) for FKGC. Our model consists of two major components: 1) an interaction attention encoder (InterAE) to capture the underlying relational semantics of entity pairs by modeling the interactive information between head and tail entities and 2) an adaptive prototype net (APNet) to generate relation prototypes adaptive to different query triples by extracting query-relevant reference pairs and reducing the data inconsistency between support and query sets. Experimental results on two public datasets demonstrate that APINet outperforms several state-of-the-art FKGC methods. The ablation study demonstrates the rationality and effectiveness of each component of APINet.

Few-shot knowledge graph completion (FKGC), which aims to infer missing facts about a relation from only a few reference triples, has recently attracted great attention. The core of solving the FKGC task is to learn a vector representation for each few-shot relation using the corresponding entity represen-tations. To this end, existing models generally enhance entity representations with their direct neighbors. However, a large number of entities have few direct neighbors. Hence, encoding only direct neighborhood is insufficient to obtain satisfactory en-tity representations. In addition, current models typically utilize static embeddings to represent entities, ignoring their diverse semantics, i.e., an entity may show distinct semantics within different few-shot relations. To address these issues, we propose a new FKGC framework, namely TransD-based Multi-hop Meta Learning (TDML). TDML consists of three main components: a multi-hop neighbor encoder to enhance entity representations by aggregating heterogeneous multi-hop neighbors, a transformer encoder to generate the relation meta representations, and a TransD-based relation representation updater that allows each entity to exhibit relation-specific semantics and tune the relation meta representations. Extensive experiments on two public datasets demonstrate that our model outperforms state-of-the-art FKGC methods.

Few-shot knowledge graph completion (FKGC) aims to infer unknown fact triples of a relation using its few-shot reference entity pairs.Recent FKGC studies focus on learning semantic representations of entity pairs by separately encoding the neighborhoods of head and tail entities.Such practice, however, ignores the inter-entity interaction, resulting in low-discrimination representations for entity pairs, especially when these entity pairs are associated with 1-to-N, N-to-1, and N-to-N relations.To address this issue, this paper proposes a novel FKGC model, named Cross-Interaction Attention Network (CIAN) to investigate the inter-entity interaction between head and tail entities.Specifically, we first explore the interactions within entities by computing the attention between the task relation and each entity neighbor, and then model the interactions between head and tail entities by letting an entity to attend to the neighborhood of its paired entity.In this way, CIAN can figure out the relevant semantics between head and tail entities, thereby generating more discriminative representations for entity pairs.Extensive experiments on two public datasets show that CIAN outperforms several state-of-the-art methods.The source code is available at https://github.com/cjlyl/FKGC-CIAN.

One mainstream method in cross-lingual word embeddings is to learn a linear mapping between two monolingual embedding spaces using a training dictionary. Successful linear mappings require isomorphic embedding spaces. However, monolingual embedding spaces are not perfectly isomorphic, and therefore, a linear mapping cannot align them accurately. In this study, we assume that two embedding spaces are composed of near-isomorphic translation pairs (NearITP) and non-isomorphic translation pairs. Owing to the nature of similar substructures, NearITP can make linear mapping work well. Motivated by this, we design a screening strategy to identify NearITP effectively. Based on this strategy, we find that the proportion of NearITP in the commonly used training dictionary is relatively low, leading to sub-optimal results. To address this problem, we propose a general framework that can be combined with any of the mapping methods, which further boosts subsequent mapping. Experimental results demonstrate that our framework is an improvement over existing mapping-based methods, and outperforms state-of-the-art models on two public data sets. Moreover, we show that our framework can be successfully generalized to contextual word embeddings such as multilingual BERT (mBERT), and further enhances the cross-lingual properties of mBERT.