公开(公告)号：US20210192364A1

公开(公告)日：2021-06-24

申请号：US17124030

申请日：2020-12-16

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Haifeng WANG ,  Wenbin JIANG ,  Yajuan LV ,  Yong ZHU ,  Hua WU

IPC: G06N5/02 ,  G06F40/30 ,  G06F40/279 ,  G06K9/62

Abstract: The present application discloses a text processing method and device based on natural language processing and a knowledge graph, and relates to the in-depth field of artificial intelligence technology. A specific implementation is: an electronic device uses a joint learning model to obtain a semantic representation, which is obtained by the joint learning model by combining knowledge graph representation learning and natural language representation learning, it combines a knowledge graph representation learning and a natural language representation learning, compared to using only the knowledge graph representation learning or the natural language representation learning to learn semantic representation of a prediction object, factors considered by the joint learning model are more in quantity and comprehensiveness, so accuracy of semantic representation can be improved, and thus accuracy of text processing can be improved.

公开(公告)号：US20210192142A1

公开(公告)日：2021-06-24

申请号：US17024756

申请日：2020-09-18

Applicant: Beijing Baidu Netcom Science Technology Co., Ltd.

Inventor： Zhifan FENG ,  Haifeng WANG ,  Kexin REN ,  Yong ZHU ,  Yajuan LYU

IPC: G06F40/30 ,  G06N3/08 ,  G06N3/04

Abstract: The present disclosure discloses a multimodal content processing method, apparatus, device and storage medium, which relate to the technical field of artificial intelligence. The specific implementation is: receiving a content processing request of a user which is configured to request semantic understanding of multimodal content to be processed, analyzing the multimodal content to obtain the multimodal knowledge nodes corresponding to the multimodal content, determining a semantic understanding result of the multimodal content according to the multimodal knowledge nodes, a pre-constructed multimodal knowledge graph and the multimodal content, the multimodal knowledge graph including: the multimodal knowledge nodes and an association relationship between multimodal knowledge nodes. The technical solution can obtain an accurate semantic understanding result, realize an accurate application of multimodal content, and solve the problem in the prior art that multimodal content understanding is inaccurate.

公开(公告)号：US20230016403A1

公开(公告)日：2023-01-19

申请号：US17934876

申请日：2022-09-23

Applicant: Beijing Baidu Netcom Science Technology Co., Ltd.

Inventor： Zhaoji WANG ,  Fang HUANG ,  Ye JIANG ,  Yabing SHI ,  Chunguang CHAI ,  Yong ZHU

IPC: G06F16/9537 ,  G06F40/226 ,  G06F40/30

Abstract: The present disclosure provides a method of processing triple data, a method of training a triple data processing model, an electronic device, and a storage medium. A specific implementation solution includes: performing a triple data extraction on text data to obtain a plurality of field data; normalizing the plurality of field data to determine target triple data, wherein the target triple data contains entity data, entity relationship data, and association entity data; and verifying a confidence level of the target triple data to obtain a verification result.

公开(公告)号：US20230013796A1

公开(公告)日：2023-01-19

申请号：US17866104

申请日：2022-07-15

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Wenbin JIANG ,  Zhifan FENG ,  Xinwei FENG ,  Yajuan LYU ,  Yong ZHU

IPC: G06N20/00 ,  G06K9/62

Abstract: The present disclosure provides a method and apparatus for acquiring a pre-trained model, an electronic device and a storage medium, and relates to the fields such as deep learning, natural language processing, knowledge graph and intelligent voice. The method may include: acquiring a pre-training task set composed of M pre-training tasks, M being a positive integer greater than 1, the pre-training tasks including: N question-answering tasks corresponding to different question-answering forms, N being a positive integer greater than 1 and less than or equal to M; and jointly pre-training the pre-trained model according to the M pre-training tasks.

公开(公告)号：US20230008897A1

公开(公告)日：2023-01-12

申请号：US17932598

申请日：2022-09-15

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Wenbin JIANG ,  Yajuan LYU ,  Yong ZHU ,  Hua WU ,  Haifeng WANG

IPC: G06F16/735

Abstract: An information search method includes: obtaining search words at least including a question to be searched and obtaining an initial text vector representation of the search words; obtaining a video corresponding to the search words, and obtaining multi-modality vector representations of the video; starting from the initial text vector representation, performing N rounds of interaction between the video and the search words based on the multi-modality vector representations and a text vector representation of the search words of a current round, to generate a target fusion vector representation, where N is an integer greater than or equal to 1; and obtaining target video frames matching the question to be searched by annotating the video based on the target fusion vector representation.