Detecting Knowledge Flows in Weblogs

Scanning the internal and external environment of an organisation to trace current and emerging topics of discourse, is an important task for knowledge management. Results of this scanning process can be used to anticipate on new developments and for detecting possible knowledge bottlenecks. This paper addresses the unobtrusive identification of conceptual structures exchanged and possibly shared in weblogs as a case to explore the opportunities for automated support for this scanning process. The research is also motivated by the observation that many domains defy formal conceptualisation in the sense that professionals, in particular knowledge workers, differ on the meaning of the concepts and terminology and the relations between them. In such domains, we use knowledge management itself as an example, detailed formalisation is pointless and the best we can aim for is finding out whether professionals agree and share, or disagree. Informally, we define a knowledge flow to be the communication of some knowledge (represented as a natural language post on a weblog in our case) to a receiver (a reader of the post in question) when the receiver acknowledges the flow by referring back to the sender in a separate post. The research challenges we pursue are to determine what knowledge is being exchanged and to what degree the participants in the communication share conceptualisations. We have implemented the identification of knowledge flows in a tool called BlogTrace. BlogTrace uses semantic web technology to represent weblogs and linkage structures, term extraction to determine concepts in natural language posts, and a statistically founded algorithm which computes a relevance metric for the concepts extracted and their relation to other concepts for any particular weblog author. Based on several assumptions, software was written that had to answer knowledge management relevant questions concerning the frequency, intensity and topics of knowledge flows in a community of bloggers, which act like members of an actual organisation. The software was tested on a set of questions and the results show that this software provides to a large extent the answers.