多代理人推理系統於整合常識知識之研究

Abstract

智慧系統需要常識知識使其更能應付使用者的各種狀況，雖然現今已建立了許多大型的常識知識庫，但這些常識知識庫也都還不完整。應用程式在使用常識知識時，往往因為知識庫選擇的知識表現與推理方式而被侷限於只能使用單一的知識庫做應用。相對於將所有知識庫合併成單一知識庫的作法，本論文提出一個用於常識知識整合之多代理人系統，以及系統中的三個重要機制：(1) 配對含有目標知識的知識庫以做推理 (2) 結合多個推理方法來回答應用程式的查詢 (3) 在有限的人力配置下以群眾外包的技術收集知識以增進知識庫的覆蓋率。最後，本論文以影片編輯與對話輔助的使用者介面作為應用案例，由此兩個案例證實結合此多代理人推理系統的介面代理人可以顯著地增加其處理使用者需求的數量。

Robust intelligent systems require commonsense knowledge. While significant progress has been made in building large commonsense knowledge bases, they are intrinsically incomplete. It is difficult to combine multiple knowledge bases due to their different choices of representations and reasoning techniques, thereby limiting users to one knowledge base and its reasoning methods for any specific task. Instead of merging knowledge bases into a single one, this paper presents a multiagent system for commonsense knowledge integration, and proposes approaches to (1) matching knowledge bases without a common ontology for reasoning, (2) combining different reasoning methods to answer queries from application, and (3) improving coverage of knowledge base via resource-bounded crowdsourcing. Two case studies on video editing and dialog assistance interfaces are also presented to show their improvement in handling user’s actions after incorporating the proposed reasoning system.