Ever-Evolving Memory by Blending and Refining the Past

For a human-like chatbot, constructing a long-term memory is crucial. A naiveapproach for making a memory could be simply listing the summarized dialogue.However, this can lead to problems when the speaker's status change over timeand contradictory information gets accumulated. It is important that the memorystays organized to lower the confusion for the response generator. In thispaper, we propose a novel memory scheme for long-term conversation, CREEM.Unlike existing approaches that construct memory based solely on currentsessions, our proposed model blending past memories during memory formation.Additionally, we introduce refining process to handle redundant or outdatedinformation. This innovative approach seeks for overall improvement andcoherence of chatbot responses by ensuring a more informed and dynamicallyevolving long-term memory.