Host: The Japanese Society for Artificial Intelligence
Name : The 39th Annual Conference of the Japanese Society for Artificial Intelligence
Number : 39
Location : [in Japanese]
Date : May 27, 2025 - May 30, 2025
One of the features of commonsense knowledge is the existence of exceptions. For instance, "birds can fly" has the exception "penguins cannot fly." In existing commonsense knowledge graphs (CKGs), this relationship can be captured as a pattern within the graph. However, these graphs cannot handle exceptions concerning extrinsic factors, such as "birds cannot fly in the stratosphere." This paper proposes a CKG that represents exceptions concerning extrinsic factors. The proposed CKG extends the triplet structure of existing CKGs. For example, knowledge is expressed as (Bird,NotCapableOf,Flying,Location:Stratosphere) , where "Location:Stratosphere" is the additional element. This allows the graph to represent knowledge involving extrinsic factors and formalize exceptions concerning such factors as patterns within the graph. Furthermore, this paper proposes a method to generate the extended CKG using LLMs. By applying this method to existing CKGs, the feasibility of the proposed CKG was verified.