{"status":"public","author":[{"full_name":"Demir, Caglar","last_name":"Demir","first_name":"Caglar","id":"43817"},{"full_name":"Moussallem, Diego","last_name":"Moussallem","id":"71635","first_name":"Diego"},{"full_name":"Heindorf, Stefan","last_name":"Heindorf","first_name":"Stefan","orcid":"0000-0002-4525-6865","id":"11871"},{"last_name":"Ngonga Ngomo","full_name":"Ngonga Ngomo, Axel-Cyrille","first_name":"Axel-Cyrille","id":"65716"}],"date_updated":"2022-10-17T15:06:40Z","_id":"29287","language":[{"iso":"eng"}],"year":"2021","type":"conference","publication":"The 13th Asian Conference on Machine Learning, ACML 2021","date_created":"2022-01-12T10:21:10Z","title":"Convolutional Hypercomplex Embeddings for Link Prediction","external_id":{"arxiv":["2106.15230"]},"department":[{"_id":"574"}],"oa":"1","citation":{"bibtex":"@inproceedings{Demir_Moussallem_Heindorf_Ngonga Ngomo_2021, title={Convolutional Hypercomplex Embeddings for Link Prediction}, booktitle={The 13th Asian Conference on Machine Learning, ACML 2021}, author={Demir, Caglar and Moussallem, Diego and Heindorf, Stefan and Ngonga Ngomo, Axel-Cyrille}, year={2021} }","ama":"Demir C, Moussallem D, Heindorf S, Ngonga Ngomo A-C. Convolutional Hypercomplex Embeddings for Link Prediction. In: <i>The 13th Asian Conference on Machine Learning, ACML 2021</i>. ; 2021.","chicago":"Demir, Caglar, Diego Moussallem, Stefan Heindorf, and Axel-Cyrille Ngonga Ngomo. “Convolutional Hypercomplex Embeddings for Link Prediction.” In <i>The 13th Asian Conference on Machine Learning, ACML 2021</i>, 2021.","short":"C. Demir, D. Moussallem, S. Heindorf, A.-C. Ngonga Ngomo, in: The 13th Asian Conference on Machine Learning, ACML 2021, 2021.","ieee":"C. Demir, D. Moussallem, S. Heindorf, and A.-C. Ngonga Ngomo, “Convolutional Hypercomplex Embeddings for Link Prediction,” 2021.","apa":"Demir, C., Moussallem, D., Heindorf, S., &#38; Ngonga Ngomo, A.-C. (2021). Convolutional Hypercomplex Embeddings for Link Prediction. <i>The 13th Asian Conference on Machine Learning, ACML 2021</i>.","mla":"Demir, Caglar, et al. “Convolutional Hypercomplex Embeddings for Link Prediction.” <i>The 13th Asian Conference on Machine Learning, ACML 2021</i>, 2021."},"user_id":"11871","abstract":[{"lang":"eng","text":"Knowledge graph embedding research has mainly focused on the two smallest\r\nnormed division algebras, $\\mathbb{R}$ and $\\mathbb{C}$. Recent results suggest\r\nthat trilinear products of quaternion-valued embeddings can be a more effective\r\nmeans to tackle link prediction. In addition, models based on convolutions on\r\nreal-valued embeddings often yield state-of-the-art results for link\r\nprediction. In this paper, we investigate a composition of convolution\r\noperations with hypercomplex multiplications. We propose the four approaches\r\nQMult, OMult, ConvQ and ConvO to tackle the link prediction problem. QMult and\r\nOMult can be considered as quaternion and octonion extensions of previous\r\nstate-of-the-art approaches, including DistMult and ComplEx. ConvQ and ConvO\r\nbuild upon QMult and OMult by including convolution operations in a way\r\ninspired by the residual learning framework. We evaluated our approaches on\r\nseven link prediction datasets including WN18RR, FB15K-237 and YAGO3-10.\r\nExperimental results suggest that the benefits of learning hypercomplex-valued\r\nvector representations become more apparent as the size and complexity of the\r\nknowledge graph grows. ConvO outperforms state-of-the-art approaches on\r\nFB15K-237 in MRR, Hit@1 and Hit@3, while QMult, OMult, ConvQ and ConvO\r\noutperform state-of-the-approaches on YAGO3-10 in all metrics. Results also\r\nsuggest that link prediction performances can be further improved via\r\nprediction averaging. To foster reproducible research, we provide an\r\nopen-source implementation of approaches, including training and evaluation\r\nscripts as well as pretrained models."}],"main_file_link":[{"url":"https://papers.dice-research.org/2021/ACML2021_HyperConv/public.pdf","open_access":"1"}]}