Evaluating Parameter Efficient Learning for Generation

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 480 - 490 December 7-11, 2022 ©2022 Association for Computational Linguistics A Good Neighbor, A Found Treasure: Mining Treasured Neighbors for Knowledge Graph Entity Typing Zhuoran Jin1,2, Pengfei Cao1,2, Yubo Chen1,2, Kang Liu1,2,3, Jun Zhao1,2 1 School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China 2 National Laboratory of Pattern Recognition, Institute of Automation, CAS, Beijing, China 3 Beijing Academy of Artificial Intelligence, Beijing, China {zhuoran.jin, pengfei.cao, yubo.chen, kliu, jzhao}@nlpr.ia.ac.cn Abstract The task of knowledge graph entity typing (KGET) aims to infer the missing types for entities in knowledge graphs. Some pioneer- ing work has proved that neighbor informa- tion is essential for the task. However, existing methods only leverage the one-hop neighbor information of the central entity, ignoring the multi-hop neighbor information that can pro- vide valuable clues for inference. Besides, we also observe that there are co-occurrence rela- tions between types, which is very helpful in al- leviating the false-negative problem. In this pa- per, we propose a novel method called Mining Treasured Neighbors (MiNer) to make use of these two characteristics. Firstly, we devise a Neighbor Information Aggregation module to aggregate the neighbor information. Then, we propose an Entity Type Inference module to mit- igate the adverse impact of the irrelevant neigh- bor information. Finally, a Type Co-occurrence Regularization module is designed to prevent the model from overfitting the false-negative examples caused by missing types. Experimen- tal results on two widely used datasets indicate that our approach significantly outperforms pre- vious state-of-the-art methods.1 1 Introduction Knowledge graphs (KGs) store huge amounts of structured data in the form of triples, i.e., (head entity, relation, tail entity). Each entity in KGs is labeled with one or more types. As shown in Figure 1, the entity Einstein not only belongs to Scientist type, but also Physicist type. The entity type information is very important and can benefit many natural language processing (NLP) applica- tions, such as entity linking (Chen et al., 2018), relation extraction (Vashishth et al., 2018), knowl- edge graph embedding (Xie et al., 2016) and text generation (Dong et al., 2021). 1https://github.com/jinzhuoran/MiNer/ Newton Einstein England Leibniz Calculus Englander Mathematician Physicist Scientist Mathematics Invent Invent Entity Type Relation Existing Type Missing Type Figure 1: An example of a KG fragment. Large brown circles denote entities and small red circles denote types. Brown solid lines denote relations between entities, red solid lines denote existing types of entities and red dot- ted lines denote missing types of entities. Unfortunately, KGs usually suffer from entity type incompleteness problems. More specifically, one entity may have multiple types, while the anno- tated entity type information is usually incomplete. Take Figure 1 as an example, the entity Newton should be labeled with Scientist, Mathematician, Physicist and Englander types. However, only the Scientist type is annotated in the KG. According to the statistics on the FB15k (Moon et al., 2017), 10% of entities have the /music/artist type, but missing the …

Content truncated to 500 words