單樣本開集關係連接預測任務（One-shot Open-set Relation Link Prediction）
專注於多關係圖中的連接預測。目標是利用大量已知關係類型的
連接進行訓練，希望模型在僅給定一個新類別的單一連接的情況下，能
夠識別圖中真正屬於該新類別的連接，並進而將其補充至關係圖中，以
完善圖的結構。這在利用關係圖結構儲存數據的應用場景中具有廣泛的
潛在價值，而其挑戰在於新類別關係的識別與結合。現有方法多依賴於
元學習（Meta-learning），這是一種專為少樣本學習設計的演算法。在模
型架構上，主要包含一個鄰居編碼器（neighbor encoder）來提取局部圖
資訊，以及一個三元組編碼器（triple encoder），將連接映射至度量空間
（metric space），並透過計算與給定連接的相似度來進行匹配。我們透
過研究第一篇提出此問題的論文： Gmatching，發現該方法在某些測試的
新類別上表現較差，而且這些類別通常存在尾部實體局部圖資訊不足的現
象；此外，標註大量已知類型的連接需耗費高昂的標記成本。因此，基
於上述觀察與考量，我們提出了一種適用於半監督式One-shot Open-set
Relation Link Prediction 任務的模型semi-MORE。semi-MORE 採用偽
標籤（pseudo-labeling） 機制來利用未標記數據，並結合部分2-hop 鄰居
資訊，以增強局部圖資訊不足的樣本，進而提升neighbor encoder 的效
果。此外，我們採用MORE 模型中的CNN 架構作為triple encoder 的基
礎框架，以結合分群方法，以進一步提升連接預測的能力。此外，我們提
出了一項新的評估指標： ranking，以更全面地衡量模型的表現。實驗結
果顯示，雖然semi-MORE在整體ranking 表現上未必全面超越大多數基
於Meta-learning 的方法，但在反映模型缺失連結識別能力的Hits@1 指標
上，我們的方法展現了明顯優勢。具體而言，在達到相同Hits@1 準確度
的前提下，本研究方法僅需1/8 的標註量。此外，當訓練標註率為80%
時，semi-MORE 的Hits@1 表現是主要比較對象Gmatching 的1.45 倍;
而在訓練標註率降低至10% 時，該比值則上升至2.06 倍。這充分說明，
透過增強局部圖資訊，模型性能得到了顯著提升。在半監督式One-shot
Open-set Relation Link Prediction 任務中，採用分群與pseudo-labeling 技
術的架構能夠有效學習局部圖特徵並展現出辨識新類別連接的潛力，雖然
該方法在訓練數據不平衡和高相似度情境下可能仍然存在固有侷限性。

The One-shot Open-set Relation Link Prediction problem focuses on link
prediction in multi-relational graphs, where the model is trained on a large number of links with known relation types. Given only a single link from a previously-unseen relation type, the objective is to identify and establish other valid instances of this new relation within the graph, thereby enhancing its structural completeness. This task has significant benefits for applications that rely on graph-based information storage. An obvious challenge is to recognize and incorporate the new-relation link into the graph. Existing methods predominantly leverage meta-learning, a technique designed for few-shot learning, to address this challenge. Typically, these models employed a neighbor encoder to capture local graph information and a triple encoder to map links into a metric space, enabling similarity comparisons with the given link. In our analysis of Gmatching, the pioneering work that introduced this problem, we observed that its performance degrades significantly for certain test relation types, particularly when tail entities lack sufficient local graph information. Moreover, labeling a large number of known relation types incurs a substantial amount of annotation cost. Motivated by these observations and considerations, we propose semi-MORE, a model specifically designed for the semi-supervised One-shot Open-set Relation Link Prediction task. Semi-MORE integrates pseudo-labeling to leverage unlabeled data while incorporating partial 2-hop neighbor information to enhance samples with insufficient local graph context, thereby improving the effectiveness of the neighbor encoder. Additionally, we employ the CNN-based architecture from the MORE model as the foundation for the triple encoder, integrating clustering approaches to further refine link prediction performance. Furthermore, to facilitate a more comprehensive assessment of model performance, we propose a novel evaluation metric: ranking. Experimental results reveal that, although semi-MORE does not consistently outperform most meta-learning-based methods in terms of overall ranking performance, it demonstrates a pronounced advantage in the Hits@1 metric, which evaluates the model’s capability to accurately identify missing links. Note that semi-MORE achieves comparable Hits@1 accuracy using only 1/8 of the labeled data. In particular, semi-MORE achieves a Hits@1 performance that is 1.45 times higher than Gmatching when 80% of the training data is labeled. This performance gain becomes even more substantial under much lower supervision: at a 10% labeled rate, semi-MORE reaches a Hits@1 score that is 2.06 times that of the same baseline. These findings highlight the pivotal role of leveraging local graph information in improving the model performance. In the semi-supervised one-shot open-set relation link prediction task, the proposed framework that employs clustering and pseudo-labeling techniques effectively captures local graph features and shows promise in distinguishing novel relational links. Nonetheless, this approach remains constrained by certain limitations, particularly under conditions of class imbalance and high inter-class similarity.

[1] Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD international conference on
knowledge discovery & data mining, pages 2623–2631, 2019.
[2] Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Oksana Yakhnenko. Translating embeddings for modeling multi-relational
data. Advances in neural information processing systems, 26, 2013.
[3] Jianghui Cai, Jing Hao, Haifeng Yang, Xujun Zhao, and Yuqing Yang. A review on semi-supervised clustering. Information Sciences, 632:164–200, 2023.
[4] Andrew Carlson, Justin Betteridge, Bryan Kisiel, Burr Settles, Estevam Hruschka, and Tom Mitchell. Toward an architecture for never-ending language learning. In Proceedings of the AAAI conference on artificial intelligence, volume 24, pages 1306–1313, 2010.
[5] Hao Chen, Zhong Huang, Yue Xu, Zengde Deng, Feiran Huang, Peng He,
and Zhoujun Li. Neighbor enhanced graph convolutional networks for node classification and recommendation. Knowledge-based systems, 246:108594, 2022.
[6] Mingyang Chen, Wen Zhang, Wei Zhang, Qiang Chen, and Huajun Chen.
Meta relational learning for few-shot link prediction in knowledge graphs. In Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan, editors, Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 4217–4226, Hong Kong, China, November 2019. Association for Computational Linguistics.
[7] Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. Convolutional 2d knowledge graph embeddings. In Proceedings of the AAAI conference on artificial intelligence, volume 32, 2018.
[8] Kaiwen Dong, Zhichun Guo, and Nitesh V Chawla. Core: Data augmentation for link prediction via information bottleneck. arXiv preprint arXiv:2404.11032, 2024.
[9] Hady Elsahar, Elena Demidova, Simon Gottschalk, Christophe Gravier, and Frederique Laforest. Unsupervised open relation extraction. In The Semantic Web: ESWC 2017 Satellite Events: ESWC 2017 Satellite Events, Portoroˇz, Slovenia, May 28–June 1, 2017, Revised Selected Papers 14, pages 12–16. Springer, 2017.
[10] Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic metalearning for fast adaptation of deep networks. In International conference on machine learning, pages 1126–1135. PMLR, 2017.
[11] Tianyu Gao, Xu Han, Ruobing Xie, Zhiyuan Liu, Fen Lin, Leyu Lin, and Maosong Sun. Neural snowball for few-shot relation learning. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 7772–7779, 2020.
[12] Tianyu Gao, Xu Han, Ruobing Xie, Zhiyuan Liu, Fen Lin, Leyu Lin, and Maosong Sun. Neural snowball for few-shot relation learning. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 7772–7779, 2020.
[13] Xuming Hu, Chenwei Zhang, Yusong Xu, Lijie Wen, and Philip S Yu. Selfore: Self-supervised relational feature learning for open relation extraction. arXiv preprint arXiv:2004.02438, 2020.
[14] Dekang Lin and Patrick Pantel. Dirt@ sbt@ discovery of inference rules from text. In Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining, pages 323–328, 2001.
[15] Yankai Lin, Zhiyuan Liu, Maosong Sun, Yang Liu, and Xuan Zhu. Learning entity and relation embeddings for knowledge graph completion. In Proceedings of the AAAI conference on artificial intelligence, volume 29, 2015.
[16] Bonan Min, Ralph Grishman, LiWan, ChangWang, and David Gondek. Distant supervision for relation extraction with an incomplete knowledge base.
In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 777–782, 2013.
[17] Ndapandula Nakashole, Gerhard Weikum, and Fabian Suchanek. Patty: A taxonomy of relational patterns with semantic types. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1135–1145, 2012.
[18] Maximilian Nickel, Volker Tresp, Hans-Peter Kriegel, et al. A three-way model for collective learning on multi-relational data. In Icml, volume 11, pages 3104482–3104584, 2011.
[19] Guanglin Niu, Yang Li, Chengguang Tang, Ruiying Geng, Jian Dai, Qiao Liu, Hao Wang, Jian Sun, Fei Huang, and Luo Si. Relational learning with gated and attentive neighbor aggregator for few-shot knowledge graph completion.
In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval, pages 213–222, 2021.
[20] Michael Schlichtkrull, Thomas N Kipf, Peter Bloem, Rianne Van Den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph convolutional networks. In The semantic web: 15th international conference, ESWC 2018, Heraklion, Crete, Greece, June 3–7, 2018, proceedings 15, pages 593–607. Springer, 2018.
[21] Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clustering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 815–823, 2015.
[22] Jiawei Sheng, Shu Guo, Zhenyu Chen, Juwei Yue, Lihong Wang, Tingwen Liu, and Hongbo Xu. Adaptive attentional network for few-ahot knowledge graph completion. In Bonnie Webber, Trevor Cohn, Yulan He, and Yang
Liu, editors, Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1681–1691, Online, nov 2020. Association for Computational Linguistics.
[23] Yisheng Song, Ting Wang, Puyu Cai, Subrota K Mondal, and Jyoti Prakash Sahoo. A comprehensive survey of few-shot learning: Evolution, applications, challenges, and opportunities. ACM Computing Surveys, 55(13s):1–40, 2023.
[24] Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research, 15(1):1929–1958, 2014.
[25] Kristina Toutanova, Danqi Chen, Patrick Pantel, Hoifung Poon, Pallavi Choudhury, and Michael Gamon. Representing text for joint embedding of
text and knowledge bases. In Proceedings of the 2015 conference on empirical methods in natural language processing, pages 1499–1509, 2015.
[26] Th´eo Trouillon, Johannes Welbl, Sebastian Riedel, ´Eric Gaussier, and Guillaume
Bouchard. Complex embeddings for simple link prediction. In International conference on machine learning, pages 2071–2080. PMLR, 2016.
[27] A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017.
[28] Denny Vrandeˇci´c and Markus Kr¨otzsch. Wikidata: a free collaborative knowledgebase. Communications of the ACM, 57(10):78–85, 2014.
[29] Hongwei Wang, Jia Wang, Jialin Wang, Miao Zhao, Weinan Zhang, Fuzheng Zhang, Xing Xie, and Minyi Guo. Graphgan: Graph representation learning with generative adversarial nets. In Proceedings of the AAAI conference on artificial intelligence, volume 32, 2018.
[30] Xinshao Wang, Yang Hua, Elyor Kodirov, Guosheng Hu, Romain Garnier, and Neil M Robertson. Ranked list loss for deep metric learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5207–5216, 2019.
[31] YutongWang, Renze Lou, Kai Zhang, Mao Yan Chen, and Yujiu Yang. More: A metric learning based framework for open-domain relation extraction. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 7698–7702. IEEE, 2021.
[32] Zhen Wang, Jianwen Zhang, Jianlin Feng, and Zheng Chen. Knowledge
graph embedding by translating on hyperplanes. In Proceedings of the AAAI conference on artificial intelligence, volume 28, 2014.
[33] Wenhan Xiong, Mo Yu, Shiyu Chang, Xiaoxiao Guo, and William Yang
Wang. One-shot relational learning for knowledge graphs. In Ellen Riloff, David Chiang, Julia Hockenmaier, and Jun’ichi Tsujii, editors, Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1980–1990, Brussels, Belgium, October-November 2018. Association for Computational Linguistics.
[34] Bishan Yang, Wen-tau Yih, Xiaodong He, Jianfeng Gao, and Li Deng. Embedding entities and relations for learning and inference in knowledge bases.
arXiv preprint arXiv:1412.6575, 2014.
[35] Limin Yao, Aria Haghighi, Sebastian Riedel, and Andrew McCallum. Structured relation discovery using generative models. In proceedings of the 2011
conference on empirical methods in natural language processing, pages 1456–1466, 2011.
[36] Limin Yao, Sebastian Riedel, and Andrew McCallum. Unsupervised relation discovery with sense disambiguation. In Proceedings of the 50th Annual
Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 712–720, 2012.
[37] Daojian Zeng, Kang Liu, Siwei Lai, Guangyou Zhou, and Jun Zhao. Relation classification via convolutional deep neural network. In Proceedings of COLING 2014, the 25th international conference on computational linguistics: technical papers, pages 2335–2344, 2014.
[38] Chuxu Zhang, Huaxiu Yao, Chao Huang, Meng Jiang, Zhenhui Li, and
Nitesh V Chawla. Few-shot knowledge graph completion. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 3041–3048, 2020.