Method for Building a Community-Oriented Agent With a Dialogue Ability Adapted to the Community

Autonomous dialogue agents are expected to be applied in multi-party dialogue where three or more agents or people converse. In multi-party dialogue, autonomous dialogue agents that perform one-to-many dialogue with interlocutors assembled for specific purposes, such as communities, are useful, because these agents can be applied to provide information and entertainment to communities. For such applications, autonomous dialogue agents need to be acceptable and attractive to community members, but no method has been established to achieve this. Therefore, we aim to realize a “Community-Oriented Agent” (COA) that is acceptable and attractive to community members through dialogue. We proposed a method for building an autonomous dialogue function of COA, which consists of two essential steps: 1) Collect dialogue examples among the community members by creating a situation in which multiple members of the community converse with other members through operating the agent; 2) Build the agent’s autonomous dialogue function by utilizing the collected examples. To validate this method, we developed a “COA system” to implement the approach and conducted the experiment. In the experiment, using this system, we collected actual dialogue examples among community members, and then asked the members to evaluate the autonomous dialogue agents built using these examples. The results showed that the members evaluated an autonomous dialogue agent built using dialogue examples collected within their community as acceptable and attractive, but it was also shown that it is difficult to build such an agent if there are not enough participants in the dialogue when collecting dialogue examples.