Core Insights - Eastern Airlines' subsidiary, Donghang Data Technology, has developed an intelligent Q&A system named AskData, which has entered the trial operation phase after three iterations of optimization [1][2] - The system utilizes a large model integration scheduling framework and retrieval enhancement technology, creating a specialized semantic parsing engine and high-precision knowledge base tailored for the aviation industry [1] - AskData allows non-technical personnel to quickly access required data using everyday language, reducing the response time for routine data queries to under 3 seconds with an accuracy rate exceeding 98% [1] Industry Application - The system is designed to meet the specific business needs of the aviation sector, incorporating over 2,000 professional terminology entries and more than 500 synonym mappings, covering 87 core business scenarios including marketing, operations, and services [1] - In the marketing domain, AskData integrates core operational data to support route revenue analysis and market strategy formulation, enhancing the efficiency of route revenue conversion [1] - For operations, the system provides real-time insights into key metrics such as aircraft utilization and on-time performance, aiding in capacity allocation and flight scheduling [1] - In the service sector, it focuses on ground support, complaint feedback, and member services, promoting refined service processes and agile responses, thereby improving the overall quality of flight support services [1] Efficiency and Cost Reduction - AskData automates the handling of massive repetitive data query requests, replacing traditional manual statistics, report generation processes, and related maintenance tasks, which reduces inefficient labor consumption [2] - This automation effectively releases human value and achieves dual optimization in data service efficiency and operational cost reduction [2]

目前，检索增强技术正从传统的文本场景逐步拓展至涵盖代码与视觉等多模态数据的应用。然而，相较于文本领域，现有向量模型在代码和视 觉模态中的检索效果仍有待提升。 此次智源研究院发布的三款新模型，为构建更强大的多模态检索增强系统提供了有力的支持。 BGE-Code-v1：新一代代码优化语义向量模型 这些模型取得了代码及多模态检索的最佳效果，并以较大优势登顶CoIR、Code-RAG、MMEB、MVRB等领域内主要测试基准。BGE自2023 年8月发布以来，已成为中国首个登顶Hugging Face榜首的国产AI模型以及Hugging Face2023年度模型全球下载冠军。 目前， BGE-Code-v1、BGE-VL-v1.5、BGE-VL-Screenshot三款模型已向社区全面开放 ，为相关技术研究与产业应用提供助力。 由智源研究院主导研发的通用向量模型系列BGE，旨在为各类数据提供高效一站式向量表征与语义检索方案，已推出覆盖中英文、多语言检 索及重排模型等多个版本，持续刷新MTEB、C-MTEB、BEIR、MIRACL等主流文本向量评测基准。BGE凭借高性能与开源特性备受业界关 注，已广泛应用于RAG、神经搜 ...