Qualitative Pre-Processing for Semantic Search of Unstructured Knowledge

Introduction

There has been a growing interest in treating knowledge as a significant organizational resource. Consistent with the interest in organizational knowledge, knowledge management has been constantly evolving over the last few years (Nonaka, 2005). Identification, acquisition, storage, dissemination, development and utilization of knowledge have been identified as the main tasks of knowledge management (Abecker and Decker, 1999). The advent of high throughput data acquisition technologies as well as digital storage technologies have made it possible to gather and store large amount of data in electronic form resulting in vast amount of information. Accurate and speedy access is required to such repositories, as knowledge is acquired by retrieving information (facts, text, document, images, audio, video etc.).

Knowledge Management Systems have to struggle to overcome the information overload. Finding particular piece of knowledge within large knowledge repositories can be very difficult. Two related problems to do with knowledge retrieval are - a. the issue of finding knowledge again, once it has been stored and b. the problem of retrieving the subset of content from the repository that is relevant to a particular problem. A large section of researchers within knowledge management is working on developing efficient tools for retrieving knowledge from such repositories.

Knowledge is seen as a mixture of various elements, which are sometimes codified and sometimes tacit. Knowledge is communicated through database entities, documents, workflows and e-mails etc (Dustdar, 2005). Such an item that allows knowledge to be communicated independently of its holder is a knowledge artifact. Knowledge artifacts may either be of structured, semi-structured, or unstructured nature (Hahn and Subramani, 2000). For subsequent discussion in this chapter, knowledge is interpreted as pieces of information (serving a specific need). The focus here is to retrieve these pieces of information.

Search and retrieval from unstructured information resources is usually by best match, i.e. to find documents which are most relevant to a given query (e.g. web searching, bibliographic database searches, digital libraries). Majority of unstructured information is either personal or developed through interaction between two or more individuals (Hicks et. al., 2002). Organizing and leveraging content from unstructured information has become a key concern today. There have been several attempts to bring 'structure' to this unstructured information. Ontology is used to capture implicit knowledge of the knowledge workers and to associate it with knowledge artifacts for classification, search, and browsing purposes (Decker et. al., 1999; Wang et. al., 2003). On the basis of ease and representational completeness, (Yang and Oh, 1993; Ounis and Pasca, 1998) argue in favor of general and intuitive knowledge representation formalisms such as conceptual graphs. Flexible and precise knowledge representation and retrieval can be achieved through knowledge representation languages that support logical inference. Many metadata languages (to support such inferences) are being developed to let people index Web information resources with knowledge representations (logical statements) and store them in Web documents (Martin and Eklund, 2000). However, these metadata languages are insufficient to satisfy several requirements necessary to allow precise, flexible, and scalable retrieval.