< Back
Posted: 30 Nov 2021 04:00

“Dialogue Summarization” November 2021 — summary from Astrophysics Data System and Arxiv

Brevi Assistant
Brevi Assistant

Business performance assistant

“Dialogue Summarization” November 2021 — summary from Astrophysics Data System and Arxiv main image

The content below is machine-generated by Brevi Technologies’ NLG model, and the source content was collected from open-source databases/integrate APIs.


Astrophysics Data System - summary generated by Brevi Assistant


Dialogue summarization aids visitors capture prominent details from long discussions in meetings, interviews, and TV collection. In this work, we perform thorough research on long dialogue summarization by checking out 3 techniques to manage the extensive input issue and situate pertinent details: extended transformer models such as Longformer, retrieve-then-summarize pipe models with several dialogue utterance access approaches, and hierarchical dialogue inscribing models such as HMNet. In this paper, we propose a controlled neural generation framework that can flexibly guide dialogue summarization with individual called entity preparation. With the rapid rise in the volume of dialogue data from every day life, there is an expanding need for dialogue summarization. In medical dialogue summarization, recaps have to be coherent and must catch all the clinically appropriate information in the dialogue. In in-depth experiments, we reveal that this method creates excellent quality training information that can better be combined with human labeled data to obtain recaps that are highly better than those generated by models trained on human data alone both in terms of clinical precision and coherency.

Previous dialogue summarization datasets primarily concentrate on open-domain chitchat dialogues, while summarization datasets for the generally made use of task-oriented dialogue have not been checked out yet. We establish a thorough benchmark for TODSum and propose a state-aware organized dialogue summarization model to incorporate dialogue state information and dialogue background. Unlike well-structured text, such as reports and encyclopedia posts, dialogue content typically originates from 2 or even more interlocutors, trading info with each other. To catch the different subject info of a discussion and summary of prominent facts for the captured topics, this work proposes 2 topic-aware contrastive learning objectives, particularly comprehensibility discovery and sub-summary generation goals, which are expected to implicitly model the subject change and take care of information spreading obstacles for the dialogue summarization job.


Source texts:



Arxiv - summary generated by Brevi Assistant


Dialogue summarization helps visitors record salient details from long conversations in conferences, interviews, and TV series. In this work, we do a comprehensive research study on long dialogue summarization by exploring 3 approaches to take care of the lengthy input issue and locating pertinent info: extended transformer models such as Longformer, retrieve-then-summarize pipe models with a number of dialogue utterance retrieval techniques, and hierarchical dialogue inscribing models such as HMNet. In this paper, we recommend a manageable neural generation framework that can flexibly direct dialogue summarization with personal named entity preparation. During training, we exploit event planning of individuals called entities and coreference info to boost temporal coherence and to lessen hallucination in neural generation. Summarizing conversations by means of neural techniques has been obtaining research grip lately, yet it is still challenging to get sensible solutions. In this work, we check out different methods to clearly include coreference details in neural abstractive dialogue summarization models to tackle the aforementioned difficulties.

Dialogue is an important part of human interaction and teamwork. Existing research primarily concentrates on short dialogue scenarios in an individual style. Dialogue summarization has been thoroughly studied and used, where the prior works mostly concentrated on exploring exceptional model frameworks to line up the input dialogue and the outcome recap. A cutting-edge end-to-end dialogue recap generation structure is proposed with two auxiliary tasks: Expectant Factual Aspect Regularization and Missing Factual Entity Discrimination.


This can serve as an example of how to use Brevi Assistant and integrated APIs to analyze text content.


Source texts:


logo

The Brevi assistant is a novel way to summarize, assemble, and consolidate multiple text documents/contents.

Partners:

© All rights reserved 2022 made by Brevi Technologies