< Back
Posted: 22 Sep 2021 01:00

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

Brevi Assistant
Brevi Assistant

Business performance assistant

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


Arxiv - summary generated by Brevi Assistant


Summing up conversations through neural strategies has been gaining research study grip recently, yet it is still testing to get functional services. In this work, we investigate different strategies to clearly integrate coreference information into neural abstractive dialogue summarization models to deal with the previously mentioned difficulties. Dialogue is an important part of human communication and teamwork. Existing research studies primarily concentrate on short dialogue scenarios in a one-on-one style. In a client service system, dialogue summarization can improve service efficiency by automatically developing summaries for lengthy talked dialogues in which customers and agents attempt to attend to problems about certain topics. To effectively carry out topic modeling on discussions and capture multi-role info, in this work we suggest a novel topic-augmented two-stage dialogue summarizer collectively with a saliency-aware neural topic version for topic-oriented summarization of customer care discussions. Dialogue summarization intends to generate a recap that indicates the bottom lines of a given dialogue. The reality regularization urges the created recap to be factually regular with the ground-truth recap during model training, which helps to boost the factual correctness of the created summary in reasoning time. With the quick boost in the volume of dialogue data from every day life, there is an expanding demand for dialogue summarization. In particular, we exploit large in-domain non-summary data to independently pretrain the dialogue encoder and the summary decoder.


Source texts:



Astrophysics Data System - summary generated by Brevi Assistant


Dialogue summarization helps viewers record salient info from lengthy discussions in conferences, meetings, and TELEVISION series. In this work, we performed a thorough research study on lengthy dialogue summarization by investigating 3 techniques to take care of the lengthy input issue and locate appropriate information: expanded transformer designs such as Longformer, retrieve-then-summarize pipe versions with a number of dialogue articulation retrieval techniques, and ordered dialogue inscribing versions such as HMNet. Especially in the customer service domain, agents might make use of dialogue summaries to aid boost their works by swiftly knowing client's issues and service progression. We really hope that this research study can benchmark Chinese dialogue summarization and benefit more studies. Dialogue summarization aims to generate a summary that suggests the key points of a given dialogue. The truth regularization urges the created summary to be factually constant with the ground-truth summary during model training, which aids to boost the accurate accuracy of the produced summary in reasoning time. With the quick increase in the volume of dialogue data from every day life, there is a growing demand for dialogue summarization. In particular, we manipulate large in-domain non-summary data to separately pretrain the dialogue encoder and the recap decoder. Unlike well-structured text, such as news reports and encyclopedia short articles, dialogue content frequently comes from two or more dialogists, exchanging details with each other. To catch the numerous topic information of a conversation and overview significant truths for the recorded subjects, this work suggests 2 topic-aware contrastive learning objectives, specifically coherence discovery and sub-summary generation goals, which are expected to implicitly model the subject adjustment and handle information spreading challenges for the dialogue summarization task.


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, reports, reviews, feedback, etc.

Partners:

© All rights reserved
2021 made by Brevi Technologies Inc.