| Task-oriented human-machine dialogue aims to build a computer system that uses natural language to interact with humans,and is a research focus of AI and NLP.Current studies are still based on structured KB.However,to build structured KB often requires a lot of manpower and is extremely costly.Unstructured documents are a kind of natural knowledge.How to introduce large-scale unstructured document knowledge into task-oriented dialogue is worth of study.We focus on a multi-document driven task-oriented dialogue technology including the following works.First of all,limited to the difficult of accurately representing the largescale unstructured documents,we build a kind of MLTC task and introduce a label-wise contrasitve learning mechanism to pre-train documents,in order to learn document representation in several different dimensions.Experiments show that the LW-PT model proposed in this paper has greatly improved the performance on multiple datasets.Secondly,to address the current lack of research on MD3,we propose a new MD3 task,guess the target document that the user is interested in by leading a dialogue.To benchmark progress,we introduce a new dataset of GuessMovie,which contains 16,881 documents,each describing a movie,and associated 13,434 dialogues.Finally,we propose the MD3 model.Keeping guessing the target document in mind,it converses with the user conditioned on both document engagement and user feedback.In order to incorporate large-scale external documents into the dialogue,it pretrains a document representation which is sensitive to attributes it talks about an object.Then it tracks dialogue state by detecting evolvement of document belief and attribute belief,and finally optimizes dialogue policy in principle of entropy decreasing and reward increasing,which is expected to successfully guess the user’ s target in a minimum number of turns.Experiments show that our method significantly outperforms several strong baseline methods and is very close to human’s performance. |
PDF Full Text Request