Contrastive Language-Knowledge Graph Pre-training

IF 1.8 4区计算机科学 Q3 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

ACM Transactions on Asian and Low-Resource Language Information Processing Pub Date : 2024-02-09 DOI:10.1145/3644820

Xiaowei Yuan, Kang Liu, Yequan Wang

{"title":"Contrastive Language-Knowledge Graph Pre-training","authors":"Xiaowei Yuan, Kang Liu, Yequan Wang","doi":"10.1145/3644820","DOIUrl":null,"url":null,"abstract":"Recent years have witnessed a surge of academic interest in knowledge-enhanced pre-trained language models (PLMs) that incorporate factual knowledge to enhance knowledge-driven applications. Nevertheless, existing studies primarily focus on shallow, static, and separately pre-trained entity embeddings, with few delving into the potential of deep contextualized knowledge representation for knowledge incorporation. Consequently, the performance gains of such models remain limited. In this paper, we introduce a simple yet effective knowledge-enhanced model, College (Contrastive Language-Knowledge Graph Pre-training), which leverages contrastive learning to incorporate factual knowledge into PLMs. This approach maintains the knowledge in its original graph structure to provide the most available information and circumvents the issue of heterogeneous embedding fusion. Experimental results demonstrate that our approach achieves more effective results on several knowledge-intensive tasks compared to previous state-of-the-art methods. Our code and trained models are available at https://github.com/Stacy027/COLLEGE.","PeriodicalId":54312,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing","volume":null,"pages":null},"PeriodicalIF":1.8000,"publicationDate":"2024-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"ACM Transactions on Asian and Low-Resource Language Information Processing","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1145/3644820","RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}

引用次数: 0

Abstract

Recent years have witnessed a surge of academic interest in knowledge-enhanced pre-trained language models (PLMs) that incorporate factual knowledge to enhance knowledge-driven applications. Nevertheless, existing studies primarily focus on shallow, static, and separately pre-trained entity embeddings, with few delving into the potential of deep contextualized knowledge representation for knowledge incorporation. Consequently, the performance gains of such models remain limited. In this paper, we introduce a simple yet effective knowledge-enhanced model, College (Contrastive Language-Knowledge Graph Pre-training), which leverages contrastive learning to incorporate factual knowledge into PLMs. This approach maintains the knowledge in its original graph structure to provide the most available information and circumvents the issue of heterogeneous embedding fusion. Experimental results demonstrate that our approach achieves more effective results on several knowledge-intensive tasks compared to previous state-of-the-art methods. Our code and trained models are available at https://github.com/Stacy027/COLLEGE.

查看原文本刊更多论文

对比语言知识图谱预培训

近年来，学术界对知识增强型预训练语言模型（PLMs）的兴趣激增，这些模型结合事实知识来增强知识驱动型应用。然而，现有的研究主要集中在浅层、静态和单独预训练的实体嵌入上，很少有人深入研究深度语境化知识表征在知识整合方面的潜力。因此，此类模型的性能提升仍然有限。在本文中，我们介绍了一种简单而有效的知识增强模型--College（对比语言-知识图谱预训练），它利用对比学习将事实知识纳入 PLM。这种方法将知识保持在原始图结构中，以提供最可用的信息，并避免了异构嵌入融合的问题。实验结果表明，与之前最先进的方法相比，我们的方法在一些知识密集型任务中取得了更有效的结果。我们的代码和训练有素的模型可在 https://github.com/Stacy027/COLLEGE 上获取。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

ACM Transactions on Asian and Low-Resource Language Information Processing Computer Science-General Computer Science

CiteScore

3.60

自引率

15.00%

发文量

241

期刊介绍： The ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP) publishes high quality original archival papers and technical notes in the areas of computation and processing of information in Asian languages, low-resource languages of Africa, Australasia, Oceania and the Americas, as well as related disciplines. The subject areas covered by TALLIP include, but are not limited to: -Computational Linguistics: including computational phonology, computational morphology, computational syntax (e.g. parsing), computational semantics, computational pragmatics, etc. -Linguistic Resources: including computational lexicography, terminology, electronic dictionaries, cross-lingual dictionaries, electronic thesauri, etc. -Hardware and software algorithms and tools for Asian or low-resource language processing, e.g., handwritten character recognition. -Information Understanding: including text understanding, speech understanding, character recognition, discourse processing, dialogue systems, etc. -Machine Translation involving Asian or low-resource languages. -Information Retrieval: including natural language processing (NLP) for concept-based indexing, natural language query interfaces, semantic relevance judgments, etc. -Information Extraction and Filtering: including automatic abstraction, user profiling, etc. -Speech processing: including text-to-speech synthesis and automatic speech recognition. -Multimedia Asian Information Processing: including speech, image, video, image/text translation, etc. -Cross-lingual information processing involving Asian or low-resource languages. -Papers that deal in theory, systems design, evaluation and applications in the aforesaid subjects are appropriate for TALLIP. Emphasis will be placed on the originality and the practical significance of the reported research.