Facilitating experiential knowledge sharing through situated conversations

This paper proposes a system that facilitates knowledge sharing among people in similar situations by providing audio of past conversations. Our system records all voices of conversations among the users in the specific fields such as tourist spots, museums, digital fabrication studio, etc. and then timely provides users in a similar situation with fragments of the accumulated conversations. For segmenting and retrieving past conversation from vast amounts of captured data, we focus on non-verbal contextual information, i.e., location, attention targets, and hand operations of the conversation participants. All voices of conversation are recorded, without any selection or classification. The delivery of the voices to a user is determined not based on the content of the conversation but on the similarity of situations between the conversation participants and the user. To demonstrate the concept of the proposed system, we performed a series of experiments to observe changes in user behavior due to past conversations related to the situation at the digital fabrication workshop. Since we have not achieved a satisfactory implementation to sense user's situation, we used Wizard of Oz (WOZ) method. That is, the experimenter visually judges the change in the situation of the user and inputs it to the system, and the system automatically provides the users with voices of past conversation corresponding to the situation. Experimental results show that most of the conversations presented when the situation perfectly matches is related to the user's situation, and some of them prompts the user to change their behavior effectively. Interestingly, we could observe that conversations that were done in the same area but not related to the current task also had the effect of expanding the user's knowledge. We also observed a case that although a conversation highly related to the user's situation was timely presented but the user could not utilize the knowledge to solve the problem of the current task. It shows the limitation of our system, i.e., even if a knowledgeable conversation is timely provided, it is useless unless it fits with the user's knowledge level.