< Back
Posted: 14 Sep 2021 20:00

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

Brevi Assistant
Brevi Assistant

Business performance assistant

“Text Summarization” September 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

 

Encoder-decoder designs have attained exceptional success in abstractive text summarization, which intends to compress several records into a shorter version without the loss of the essential content. During the training phase, with instructor requiring these versions are maximized to take full advantage of the likelihood of the gold summary offered the gold summary tokens as input to the decoder, while at reasoning the offered tokens are replaced by the generated tokens. In this paper, we present a denoising sequence-to-sequence autoencoder through contrastive learning for abstractive text summarization. These two parts, seq2seq autoencoder and contrastive learning, are jointly trained through fine-tuning, which boosts the performance of text summarization when it comes to ROUGE scores and human evaluation. In spite of significant progress has been made in text summarization, factual disparity in generated summaries still seriously restricts its functional applications. Due to these difficulties, we suggest a novel statistics to evaluate the valid uniformity in text summarization via counterfactual evaluation, which develops the causal relationship among the source document, the created recap, and the language prior. Extractive text summarization targets at removing one of the most depictive sentences from a given document as its summary. Recent studies have leveraged graph semantic networks to catch the inter-sentential relationship to learn contextual sentence embedding. Contrastive learning designs have achieved fantastic success unsupervised visual representation learning, which maximizes the resemblances in between feature representations of various views of the very same picture, while minimizing the resemblances between attribute depictions of sights of various photos. In this paper, we recommend a contrastive learning version for monitored abstractive text summarization, where we check out a document, its gold recap and its model created recaps as various sights of the exact same mean representation and make best use of the resemblances in between them throughout training.

 

Source texts:

 

Arxiv - summary generated by Brevi Assistant

 

Encoder-decoder designs have achieved remarkable success in abstractive text summarization, which intends to press one or more papers into a much shorter variation without the loss of the essential content. During the training stage, with the teacher forcing, these versions are enhanced to make best use of the chance of the gold recap offered the gold recap tokens as input to the decoder, while at reasoning the offered tokens are replaced by the generated tokens. When supervised information is limited, Contrastive Learning has arised as a powerful representation learning method and facilitates numerous downstream tasks especially. We utilize these examples for supervised contrastive learning to obtain much better text depictions which substantially benefit text category tasks with limited annotations. In this paper, we provide a denoising sequence-to-sequence autoencoder by means of contrastive learning for abstractive text summarization. We carried out experiments on 2 datasets and demonstrate that our design outperforms many existing criteria and attains equivalent efficiency to the cutting edge abstractive systems trained with more complex styles and substantial computation resources. In spite of substantial development has been attained in text summarization, valid variance in generated summaries still seriously limits its sensible applications. Due to these difficulties, we recommend novel statistics to assess the accurate uniformity in text summarization using counterfactual estimate, which develops the causal relationship among the resource document, the created summary, and the language prior. Extractive text summarization targets at removing one of the most depictive sentences from a given document as its summary. To extract an excellent summary from a lengthy text document, sentence embedding plays an essential duty.

 

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.