In the just-concluded 2022 World Artificial Intelligence Conference (WAIC), Shanghai Midu Information Technology Co., Ltd. not only hosted the "Data Intelligence and Content Cognition Summit Forum" with the Shanghai Artificial Intelligence Industry Association, inviting data Top domestic and foreign scholars in the field of intelligence jointly discussed in depth technological evolution and future vision; in the exhibition, virtual images such as "Mi Xiaozhi", "Mi Xiaodu" and "Mi Xiaoxiao" were used to lead the audience to immersively experience the cross-border technology. The two major data intelligent applications of modal information retrieval and intelligent proofreading are empowered in specific scenarios. As two innovative applications developed by and with strong scientific research capabilities and profound industry insights,
cross-modal information retrieval and intelligent proofreading can become important ways to deepen the value of data and improve work efficiency. Multi-scenario data intelligence products and solutions based on the two have also provided users in different industries with full-process services from data fusion to intelligent cognition, from policy generation to automatic processing, comprehensively assisting the government and enterprises in digital intelligence. transformation and upgrade.
Cross-modal retrieval: gathering turbulence to find a scoop
Human activities are accompanied by the generation and dissemination of information. With the advancement of technology, the modalities of information have also developed from single text to multi-modal, including pictures, audio and The proportion of multi-modal information including video is increasing. And when turbulent and complex information pours in, cross-modal retrieval can help us get closer to the answers we are looking for.
From a technical perspective, the implementation of cross-modality mainly relies on the following four levels of technology:
The first is cross-modal comparative learning, which refers to first performing data enhancement from the similarity sorting in single-modal data, and then using the corresponding features to calculate the comparison Learning loss ultimately makes the model perform better in multi-modal tasks. The second
is cross-modal semantic fusion, which refers to improving understanding ability and efficiency by integrating models and features between different modalities, realizing automatic error correction of speech and text, and improving recognition accuracy.
third is cross-modal semantic representation, which refers to integrating multi-modal information and combining representations to achieve integrated recognition of video content. The fourth of
is cross-modal semantic retrieval, which refers to using search feature vectors to perform approximate nearest neighbor calculations on massive target high-dimensional vectors to achieve semantic retrieval and recall of TopN similar results, ultimately improving the accuracy of retrieval results.
adheres to efficient and comprehensive cross-modal retrieval capabilities. Midu's cross-modal retrieval platform "Midu Suoji" not only ensures the accuracy of text interpretation and image recognition, but also can intelligently identify text content in images, accurately Analyze the subtitles, background, cover and other characteristic elements in the video.
also continues to optimize the details of the model. At present, Midu Suoji has achieved rapid recognition and extraction of common scenes, and has strengthened training for more than 100 government units and nearly 300 special scenes to achieve special Scene recognition; and output visual and voice multi-dimensional content tags through intelligent recognition capabilities to further improve retrieval efficiency.
Currently, Midu's products such as Midu Suoji, Midu Copyright Tong, and Cheng Gantong have been embedded with advanced cross-modal retrieval capabilities, providing services for social governance, network security, copyright protection, brand decision-making, marketing insights, etc. Scenarios create benchmark applications. In addition to various scenario-based applications of
, the development of cross-modal retrieval has also brought unprecedented potential to AIGC (AI Generated Content). An excellent example is the rapidly developing AI painting - with the cross-modal comprehensive technical capabilities of large models, artificial intelligence can fuse multi-modal information such as images, videos, audios, and semantics through representation learning, and then use Collaborative training of cross-modal data ultimately allows abstract natural language to automatically generate visual images through pre-trained models.
Works drawn by Midu AI painter "Mixiaodu"
With the overall development of artificial intelligence technology, cross-modal retrieval not only improves search efficiency and result quality, but also helps us break through the creative limitations of the human brain and use appropriate Use your imagination to create a more exciting future world.
Intelligent Proofreading: See everything at a glance.
Midu’s intelligent proofreading application capabilities combine the industry’s advanced natural language processing, knowledge graph and optical character recognition technologies to automatically discover and detect errors in Chinese text and semantic relationships. Correction processing can be widely used in government documents, press releases, daily writing and other scenarios. While reducing the probability of errors and improving text quality, it also greatly improves work efficiency. From a technical perspective,
’s implementation of intelligent proofreading mainly has the following features:
is based on “big data + big model” and tens of billions of balanced corpus to capture and identify subtle semantic information. The second
is a dedicated proofreading knowledge graph, driven by knowledge graph technology, depicting entity relationships such as people, institutions, regions, etc., to achieve proofreading of current affairs-related expressions, so that the string has associated semantics. The third aspect of
is proofreading empowerment in professional fields. Through integrated learning technology, proofreading capabilities in different industries can be quickly formed.
Midu's AI intelligent proofreading platform "Midu Proofreading" is a professional software developed based on intelligent proofreading applications. Midu Proofreading Center focuses on Chinese language characteristics and usage habits, and is based on tens of billions of training corpus. It covers three major review and proofreading types: text punctuation errors, knowledge errors, and content-oriented risks, and has 25 types of full-stack review and proofreading capabilities. , able to correct typos, Intelligent review and proofreading of words, errors in too many words or too few words, semantic duplication, word order errors, mixed sentence patterns, errors in quantity and unit; proper nouns and terminology, names of laws and regulations, common sense errors, etc., to effectively solve the standardization of content , safety and legality issues.
Whether it is daily official documents, ideological reports, publicity releases, work summaries and other materials of government units; or books, journals, scientific research reports, papers, media releases, special reports and other documents; or electronic publications such as audio-visual electronics and online games; As well as corporate soft articles, product promotion materials, planning projects and other contents, Midu Proofreading can perform quick error-sensitive proofreading to improve content quality and ensure content security in a one-stop and all-round way.
AI intelligent proofreading platform - Midu Proofreader
At the 2022 World Artificial Intelligence Conference (WAIC), the software and hardware localization intelligent solution - Proofreader AI-Box was also officially released, as the first Huawei Shengteng AI ecological certified localized intelligent proofreading solution, Proofread AI-Box can fully guarantee data privacy under localized deployment; it can also be used as an exclusive edge computing to greatly improve user experience. Work efficiency; while also integrating into Huawei Shengteng In the process of building an AI ecosystem, we work with mainstream domestic systems to create a high-quality digital office experience.
's best respect for data is to intelligently mine their hidden value. In this process, we can not only aggregate human past experience to create a faster algorithm model, but also need artificial intelligence to use different methods than humans. Perceive the world through the brain, thereby opening up a new way to not only understand all things, but also absorb energy, and ultimately create a more exciting future.
artificial intelligence is a powerful tool belonging to this era. It is not only a companion on the long journey of information retrieval, but also the creator of reconstructing the world. As a leader in the field of data intelligence, Midu is committed to using every bit of technological progress to promote In the development of digital intelligence in all walks of life, together with industry partners and users, we are moving forward and exploring the vastness.
