公开(公告)号：US12204847B2

公开(公告)日：2025-01-21

申请号：US17938572

申请日：2022-10-06

Applicant: Salesforce, Inc.

Inventor： Alexander R. Fabbri ,  Prafulla Kumar Choubey ,  Jesse Vig ,  Chien-Sheng Wu ,  Caiming Xiong

IPC: G06F17/00 ,  G06F40/166 ,  G06F40/284 ,  G06N20/00

Abstract: Embodiments described herein provide a method for text summarization. The method includes receiving a training dataset having at least an uncompressed text, a compressed text, and one or more information entities accompanying the compressed text. The method also includes generating, using a perturber model, a perturbed text with the one or more information entities being inserted into the compressed text. The method further includes training the perturber model based on a first training objective, and generating, using the trained perturber model, a perturbed summary in response to an input of a reference summary. The method further includes generating, via an editor model, a predicted summary by removing information from the perturbed summary conditioned on a source document of the reference summary, and training the editor model based on a second training objective.

公开(公告)号：US12050855B2

公开(公告)日：2024-07-30

申请号：US17749837

申请日：2022-05-20

Applicant: Salesforce, Inc.

Inventor： Wojciech Kryscinski ,  Alexander R. Fabbri ,  Jesse Vig

IPC: G06F16/00 ,  G06F16/332 ,  G06F16/34 ,  G06F40/166

CPC classification number: G06F40/166 ,  G06F16/3329 ,  G06F16/345

Abstract: Embodiments described herein provide a query-focused summarization model that employs a single or dual encoder model. A two-step approach may be adopted that first extracts parts of the source document and then synthesizes the extracted segments into a final summary. In another embodiment, an end-to-end approach may be adopted that splits the source document into overlapping segments, and then concatenates encodings into a single embedding sequence for the decoder to output a summary.

公开(公告)号：US20220277135A1

公开(公告)日：2022-09-01

申请号：US17749837

申请日：2022-05-20

Applicant: Salesforce, Inc.

Inventor： Wojciech Kryscinski ,  Alexander R. Fabbri ,  Jesse Vig

IPC: G06F40/166 ,  G06F16/34 ,  G06F16/332

Abstract: Embodiments described herein provide a query-focused summarization model that employs a single or dual encoder model. A two-step approach may be adopted that first extracts parts of the source document and then synthesizes the extracted segments into a final summary. In another embodiment, an end-to-end approach may be adopted that splits the source document into overlapping segments, and then concatenates encodings into a single embedding sequence for the decoder to output a summary.

公开(公告)号：US20240370640A1

公开(公告)日：2024-11-07

申请号：US18774375

申请日：2024-07-16

Applicant: Salesforce, Inc.

Inventor： Wojciech Kryscinski ,  Alexander R. Fabbri ,  Jesse Vig

IPC: G06F40/166 ,  G06F16/332 ,  G06F16/34

Abstract: Embodiments described herein provide a query-focused summarization model that employs a single or dual encoder model. A two-step approach may be adopted that first extracts parts of the source document and then synthesizes the extracted segments into a final summary. In another embodiment, an end-to-end approach may be adopted that splits the source document into overlapping segments, and then concatenates encodings into a single embedding sequence for the decoder to output a summary.

公开(公告)号：US20240249082A1

公开(公告)日：2024-07-25

申请号：US18460373

申请日：2023-09-01

Applicant: Salesforce, Inc.

Inventor： Philippe Laban ,  Jesse Vig ,  Wojciech Kryscinski

IPC: G06F40/40 ,  G06F40/166 ,  G06F40/284 ,  G06F40/30

CPC classification number: G06F40/40 ,  G06F40/166 ,  G06F40/284 ,  G06F40/30

Abstract: A method of training a text simplification model is provided. A training dataset including a first set of original textual samples and original revision histories and a second set of simplified textual samples and simplified revision histories is received via a data interface. A training pair including an original textual sample and corresponding original revision history from the first set and a counterpart simplified textual sample and corresponding simplified revision history from the second set are identified. An alignment label for a first revision in the corresponding original revision history and a second revision in the corresponding simplified revision history are generated using a neural network-based alignment model from a score. A revision category label for each of the first revision and second revision is generated using a neural network-based classification model. A neural network-based text simplification model is trained based on the updated training dataset.

公开(公告)号：US20230419017A1

公开(公告)日：2023-12-28

申请号：US17938572

申请日：2022-10-06

Applicant: Salesforce, Inc.

Inventor： Alexander R. Fabbri ,  Prafulla Kumar Choubey ,  Jesse Vig ,  Chien-Sheng Wu ,  Caiming Xiong

IPC: G06F40/166 ,  G06F40/284 ,  G06N20/00

CPC classification number: G06F40/166 ,  G06F40/284 ,  G06N20/00

Abstract: Embodiments described herein provide a method for text summarization. The method includes receiving a training dataset having at least an uncompressed text, a compressed text, and one or more information entities accompanying the compressed text. The method also includes generating, using a perturber model, a perturbed text with the one or more information entities being inserted into the compressed text. The method further includes training the perturber model based on a first training objective, and generating, using the trained perturber model, a perturbed summary in response to an input of a reference summary. The method further includes generating, via an editor model, a predicted summary by removing information from the perturbed summary conditioned on a source document of the reference summary, and training the editor model based on a second training objective.

公开(公告)号：US20230376677A1

公开(公告)日：2023-11-23

申请号：US17880502

申请日：2022-08-03

Applicant: Salesforce, Inc.

Inventor： Prafulla Kumar Choubey ,  Alexander R. Fabbri ,  Jesse Vig ,  Chien-Sheng Wu ,  Wenhao Liu ,  Nazneen Rajani

IPC: G06F40/166 ,  G06N20/00

CPC classification number: G06F40/166 ,  G06N20/00

Abstract: Embodiments described herein provide a document summarization framework that employs an ensemble of summarization models, each of which is a modified version of a base summarization model to control hallucination. For example, a base summarization model may first be trained on a full training data set. The trained base summarization model is then fine-tuned using a first filtered subset of the training data which contains noisy data, resulting in an “anti-expert” model. The parameters of the anti-expert model are subtracted from the parameters of the trained base model to produce a final summarization model which yields robust factual performance.