(Comet-) Atomic 2020: On Symbolic and Neural Commonsense Knowledge Graphs
نویسندگان
چکیده
Recent years have brought about a renewed interest in commonsense representation and reasoning the field of natural language understanding. The development new knowledge graphs (CSKG) has been central to these advances as their diverse facts can be used referenced by machine learning models for tackling challenging tasks. At same time, there remain questions quality coverage resources due massive scale required comprehensively encompass general knowledge. In this work, we posit that manually constructed CSKGs will never achieve necessary applicable all situations encountered NLP agents. Therefore, propose evaluation framework testing utility KGs based on how effectively implicit representations learned from them. With goal, Atomic 2020, CSKG general-purpose containing is not readily available pretrained models. We evaluate its properties comparison with other leading CSKGs, performing first large-scale pairwise study resources. Next, show 2020 better suited training generate accurate, representative new, unseen entities events. Finally, through human evaluation, few-shot performance GPT-3 (175B parameters), while impressive, remains ~12 absolute points lower than BART-based model trained despite using over 430x fewer parameters.
منابع مشابه
Neuro-symbolic representation learning on biological knowledge graphs
Motivation Biological data and knowledge bases increasingly rely on Semantic Web technologies and the use of knowledge graphs for data integration, retrieval and federated queries. In the past years, feature learning methods that are applicable to graph-structured data are becoming available, but have not yet widely been applied and evaluated on structured biological knowledge. Results: We deve...
متن کاملWebBrain: Joint Neural Learning of Large-Scale Commonsense Knowledge
Despite the emergence and growth of numerous large knowledge graphs, many basic and important facts about our everyday world are not readily available on the Web. To address this, we present WebBrain, a new approach for harvesting commonsense knowledge that relies on joint learning from Web-scale data to fill gaps in the knowledge acquisition. We train a neural network model to learn relations ...
متن کاملCommonsense Knowledge Representation I
Significant advances in artificial intelligence, including machines that play master level chess, or make medical diagnoses, highlight an intriguing paradox. While systems can compete with highly qualified experts in many fields, there has been much less progress in constructing machines that exhibit simple commonsense, the kind expected of any normally intelligent child. As a result, commonsen...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence
سال: 2021
ISSN: ['2159-5399', '2374-3468']
DOI: https://doi.org/10.1609/aaai.v35i7.16792