Discussion Mining: Knowledge Discovery from Sets of Minutes
Discussion Mining is a preliminary study for recording human activities in a meetings, creating a structurization minute semi-automatically and extracting reusable knowledge from the minute.A set of minutes are represented structurally by using similarity calculation and semantic information (annotation), because there are same or similar themes between stored minutes.
In this paper, we describe a method for knowledge discovery from a set of minutes.We represent the set of stored minutes structurally, discover important minutes and related mintes by using the structure of minutes, and integrate the set of minutes.
2 Method to Make Multimedia Minutes
Related works about discussion support system use automatical recognition techniques for audio and vidual information, such as Meeting BrowzerIn a discussion mining system, we record human activities in the real world with some cameras, microphones, and tools based a Web browzer.A presenter of the meeting transmit her/his presenatation data (mainly slides generated by Microsoft PowerPoint) and timings of operation of slide. These data are recoded in the discussion mining system automatically.A secretariat record a statement in the meeting with a secretariat tool based Web browzer.The meeting participants support the automatic generation of minutes by transmitting their user IDs and types of their statements at the meeting via InfraRed signals via a special device called a ``discussion tag.'' Each participant has three types of discussion tag, colored green, yellow, and red. The green one is shown to all participants when the participant wants to make a comment. The red one is used for asking a question, and the yellow one is for answering one. All tags are also used to evaluation the entire0 discussion.
The content of the minutes is represented in the XML data format and stored in an XML database, while the audio-visual content is accumulated in the multimedia database. These databases are connected to the network. The XML data include pointers to the multimedia data. Figure shows an example of XML data of minutes.
gIBIS appears to represent groupware by the structural approach. It can display the structure of a discussion graphically to facilitate the understanding of the content of minutes and to encourage effective statements.
We discover the important statement from the minutes with the structure of discussion.The statements which have links from and to the important statement is also important, we use spreading activation model by the folloing formula for calculating importance of statement.
We repeated to calculate importance of statements until the activation value is converged. We determined the converged activation value as the importance of statement.
3 Structurization of Sets of Minutes
A set of minutes are represented structurallybecause there are same or similar themes between stored minutes.Then, we generate the minutes map (Fig.).Users can easily understand the structure of sets of minutes by using the minutes map.