I have a dataset of long conversations between two users. What i’m trying to accomplish is:
given a message from “user1”, and a label from “user2” (the next message from the other user), retrieve candidate sentences (unchanged!) from the dataset by similarity to the first message. Obviously not retrieving the label and preferably taking the previous context into account.
Could this be thought of as a QA problem with external knowledge retrieval? I realize that i could retrieve by answer sentence (label) similarity from the dataset, but i dont know if will work as expected given that there would be no context and that the question would not be identical to the dataset. Any advice?