WSL-DS: Weakly Supervised Learning with Distant Supervision for Query Focused Multi-Document Abstractive Summarization

In the Query Focused Multi-Document Summarization (QF-MDS) task, a set of\ndocuments and a query are given where the goal is to generate a summary from\nthese documents based on the given query. However, one major challenge for this\ntask is the lack of availability of labeled training datasets. To overcome this\nissue, in this paper, we propose a novel weakly supervised learning approach\nvia utilizing distant supervision. In particular, we use datasets similar to\nthe target dataset as the training data where we leverage pre-trained sentence\nsimilarity models to generate the weak reference summary of each individual\ndocument in a document set from the multi-document gold reference summaries.\nThen, we iteratively train our summarization model on each single-document to\nalleviate the computational complexity issue that occurs while training neural\nsummarization models in multiple documents (i.e., long sequences) at once.\nExperimental results in Document Understanding Conferences (DUC) datasets show\nthat our proposed approach sets a new state-of-the-art result in terms of\nvarious evaluation metrics.\n