A Domain-Specific Conversational AI Framework for Evidence-Grounded Aviation Safety Analysis Using NASA ASRS Data
编号:216 访问权限:仅限参会人 更新:2026-11-20 00:39:48 浏览:55次 Online

报告开始:2026年11月21日 16:00(Asia/Kolkata)

报告时间:15min

所在会场:[S10] Mixed Track [S10-1] Mixed Track

视频 无权播放 演示文件 附属文件

提示:该报告下的文件权限为仅限参会人,您尚未登录,暂时无法查看。

摘要
The aviation safety reports provide useful information about operational incidents; however, extracting relevant evidence from huge sets of narratives is not easy. This paper introduces a domain-dependent, evidence-based conversational artificial intelligence system to process information available in the National Aeronautics and Space Administration’s Aviation Safety Reporting System (ASRS). This framework utilizes the techniques like rule-based intent routing, bounded multi-level planning, semantic retrieval of incidents, deterministic statistics, evidence validation, incremental information acquisition, num ber of report evaluation, and problem-flight phase association analysis. In an evaluation based on 200 questions authored by researchers under eight intent classes, we attained 82.00% of classification accuracy and 81.78% of macro F1-score. All 200 planners took no more than six steps. The evidence verifier demonstrated 100% precision with regard to eight cases in controlled tests without any false acceptances or false rejection. In the retrieval task using the known target with 50 queries, the mean reciprocal rank was 0.5941 and the Recall@10 was 70%. In an incremental ingestion experiment based on the historical states of the AeroSafe AI repository, all 50 records were ingested, the total 6,455 number of records and embedding were recovered, and no duplicates in incident IDs were produced. There are currently 7,319 ASRS incident records in the AeroSafe AI repository. The semantic search is based on 384-dimension embedding from all-MiniLM-L6-v2 Sentence Transformer model. It proves the implementation feasibility but not the operational aviation safety effectiveness and accident prediction capabilities.
关键词
Aviation safety;conversational AI;NASA ASRS;semantic retrieval;evidence verification;large language models;safety analytics
报告人
Janani S
Student Thiagarajar college of engineering

稿件作者
Janani S Thiagarajar college of engineering
Amirtha V Thiagarajar college of engineering
Karthiyayini R Thiagarajar college of engineering
Uma K V Thiagarajar college of engineering
发表评论
验证码 看不清楚,更换一张
全部评论
重要日期
  • 会议日期

    11月20日

    2026

    至

    11月22日

    2026

  • 10月31日 2026

    初稿截稿日期

  • 11月07日 2026

    提前注册日期

  • 11月22日 2026

    注册截止日期

主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询