Knowledge trees and protoforms in question‐answering systems

Abstract

We point out that question-answering systems differ from other information-seeking applications, such as search engines, by having a deduction capability, an ability to answer questions by a synthesis of information residing in different parts of its knowledge base. This capability requires appropriate representation of various types of human knowledge, rules for locally manipulating this knowledge, and a framework for providing a global plan for appropriately mobilizing the information in the knowledge to address the question posed. In this article we suggest tools to provide these capabilities. We describe how the fuzzy set-based theory of approximate reasoning can aid in the process of representing knowledge. We discuss how protoforms can be used to aid in deduction and local manipulation of knowledge. The idea of a knowledge tree is introduced to provide a global framework for mobilizing the knowledge base in response to a query. We look at some types of commonsense and default knowledge. This requires us to address the complexity of the nonmonotonicity that these types of knowledge often display. We also briefly discuss the role that Dempster-Shafer structures can play in representing knowledge. IntroductionThe widespread availability of the Internet has generated considerable interest in electronic sources of information. Various types of software applications have been developed to support this interest. We can generically refer to these as applications useful for seeking information (AUSIN). One widely used class of these AUSINs are search engines. These applications retrieve pointers to pages or files based on their matching some keywords specified by the user.Another widely used class ofAUSINs are database systems. In some ways database applications and search engines are opposite extremes of these AUSINs. Database applications give precise information; however, they only work in highly structured environments. The search engines, although only giving imprecise responses, pointers, are capable of functioning in highly unstructured, almost chaotic environments. However, neither of these two applications typically has reasoning capacity.Another class of AUSINs are question-answering systems (Chaudri & Fikes, 1999;Clark, Thompson, & Porter, 1999;Maybury, 2003;Zadeh, 2004b). An important dimension along which question-answering systems differ from search engines and most databases is reasoning ability (Zadeh, 2004b). A fundamental characteristic of question-answering systems is their ability to reason over an information base. This facility leads to a fundamental difference in the nature of the response to query. In particular, the response from a search engine is a pointer to a document (file or Web page) resident in its library.1 The database responds essentially by providing some value already resident in the database. A question-answering system, because of its reasoning capacity, can construct new knowledge that is not resident in its knowledge base in response to a query. Although currently less commo...

Relevance in systems having a fuzzy-set-based semantics

“…Let U is A be some available knowledge of U where A is normal. Using this with the appropriate approximate reasoning protoform 3 we can infer V is D where D~y! ϭ Max x @H~x, y!…”

“…7,8 As discussed in Ref. 3, a knowledge tree arranges and mobilizes the user's local knowledge to address the question being asked. One important feature of the knowledge tree approach is that it can point us in directions for seeking additional information, not in our local knowledge base, that may help in getting a better answer than locally available.…”

Exploring the learning mechanism of web‐based question‐answering systems and their design

“…With the deepening of understanding, people come to realise that QA systems differ from common information‐seeking applications, such as search engines, by having a deduction capability (Yager, 2006). QA systems can provide answers to user questions, while web‐based search engines, such as Google and NorthernLight, usually only return documents that are relevant to a user query.…”