<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Event Extraction | Aqualab - Northwestern University</title><link>https://aqualab.cs.northwestern.edu/tag/event-extraction/</link><atom:link href="https://aqualab.cs.northwestern.edu/tag/event-extraction/index.xml" rel="self" type="application/rss+xml"/><description>Event Extraction</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Mon, 01 Dec 2025 00:00:00 +0000</lastBuildDate><image><url>https://aqualab.cs.northwestern.edu/media/logo_hu_9be9516131db575.png</url><title>Event Extraction</title><link>https://aqualab.cs.northwestern.edu/tag/event-extraction/</link></image><item><title>Sentinel</title><link>https://aqualab.cs.northwestern.edu/project/sentinel/</link><pubDate>Mon, 01 Dec 2025 00:00:00 +0000</pubDate><guid>https://aqualab.cs.northwestern.edu/project/sentinel/</guid><description>&lt;div class="article-style"&gt;
&lt;p&gt;&lt;a class="btn btn-primary" href="https://aqualab.cs.northwestern.edu/sentinel-events/"&gt;View Cable Failure Events &amp;rarr;&lt;/a&gt;&lt;/p&gt;
&lt;h2 id="project-overview"&gt;Project Overview&lt;/h2&gt;
&lt;p&gt;Submarine cable failures — from anchor drags off Mombasa to ship strikes in the Red Sea — affect connectivity for entire countries and regions. Yet learning about these events still depends on ad-hoc monitoring: scanning SubTelForum posts, regional press in local languages, and operator mailing lists. By the time a researcher notices an event, the window for active measurement may have closed.&lt;/p&gt;
&lt;p&gt;Sentinel is a domain-specific event extraction system that automates this process. It watches a tiered set of sources — operator and industry feeds (SubTelForum, TeleGeography), regional press in French, Portuguese, Spanish, and English, and general news — and uses LLM-based extraction to identify cable failure events. A corroboration engine scores confidence by cross-referencing independent reports, distinguishing new incidents from ongoing ones through two-layer deduplication (document-level and event-level).&lt;/p&gt;
&lt;p&gt;The extracted events feed a persistent SQLite database recording the full lifecycle of each incident: which cable, where, when, who is affected, what operators, repair status, and linked source URLs for provenance. This database is designed for two downstream uses:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Active measurement integration&lt;/strong&gt; — Sentinel triggers traceroute-based path-discovery campaigns through a bridge wrapper, targeting countries served by the affected cable. Measurement feedback flows back to confirm or refute the reported failure.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Calypso analysis&lt;/strong&gt; — The structured event database enables longitudinal study of SCN failures: frequency by cable, region, and cause; repair timelines; correlation with routing changes observed in traceroute data.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;h2 id="architecture"&gt;Architecture&lt;/h2&gt;
&lt;p&gt;
&lt;figure &gt;
&lt;div class="d-flex justify-content-center"&gt;
&lt;div class="w-100" &gt;&lt;img src="https://aqualab.cs.northwestern.edu/project/sentinel/sentinel-architecture.svg" alt="Sentinel architecture diagram" loading="lazy" data-zoomable /&gt;&lt;/div&gt;
&lt;/div&gt;&lt;/figure&gt;
&lt;/p&gt;
&lt;p&gt;Sentinel&amp;rsquo;s pipeline flows left to right: a &lt;strong&gt;monitor&lt;/strong&gt; fetches from tiered sources (operator feeds, regional press, general news) and deduplicates at the document level. An &lt;strong&gt;extraction&lt;/strong&gt; stage applies multilingual keyword filtering and LLM-based event extraction, then scores confidence through cross-source corroboration. Confirmed events are persisted in a &lt;strong&gt;SQLite store&lt;/strong&gt; that records the full incident lifecycle — timing, geography, operators, severity, and linked source URLs.&lt;/p&gt;
&lt;p&gt;From there, the pipeline branches. A &lt;strong&gt;notification fan-out&lt;/strong&gt; dispatches alerts (email, webhooks, file drops) and can trigger a &lt;strong&gt;measurement bridge&lt;/strong&gt; that launches traceroute-based path-discovery campaigns targeting affected regions. Measurement feedback flows back to the store, confirming or refuting the reported failure. A &lt;strong&gt;web dashboard&lt;/strong&gt; provides a read-only view of all events grouped by year. Downstream, &lt;strong&gt;Calypso&lt;/strong&gt; reads from the store for longitudinal SCN failure analysis.&lt;/p&gt;
&lt;p&gt;The system runs as a cron job (&lt;code&gt;sentinel check&lt;/code&gt;) given the relatively low rate of cable failures (~100-200/year globally), with a daemon mode available for continuous monitoring.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="notification-and-integration"&gt;Notification and Integration&lt;/h2&gt;
&lt;p&gt;When Sentinel detects a new cable failure event meeting a configurable confidence threshold, it dispatches notifications through a fan-out emitter:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Email alerts&lt;/strong&gt; — configurable recipient list, piped through sendmail&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Webhooks&lt;/strong&gt; — HTTP POST with optional bearer auth for Slack, PagerDuty, or custom endpoints&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;File drop&lt;/strong&gt; — JSON event files for pipeline integration&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Measurement bridge&lt;/strong&gt; — triggers traceroute-based path-discovery campaigns targeting affected regions&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;h2 id="open-problems"&gt;Open Problems&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Extraction accuracy&lt;/strong&gt; — LLM-based extraction from noisy, multilingual news sources requires careful prompt engineering and ground-truth validation against validated historical databases (e.g., from TeleGeography)&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Source coverage&lt;/strong&gt; — Expanding beyond English, French, Portuguese, and Spanish to cover Vietnamese, Arabic, and other languages relevant to cable-connected regions&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Timeliness&lt;/strong&gt; — Reducing the gap between a failure occurring and Sentinel detecting it, particularly for regions with sparse press coverage&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;False positive suppression&lt;/strong&gt; — Distinguishing actual cable faults from planned maintenance, political commentary about cables, and speculation&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;h2 id="related-projects"&gt;Related Projects&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a href="https://aqualab.cs.northwestern.edu/project/scn-criticality/"&gt;Criticality of the Submarine Cable Network&lt;/a&gt; — the broader research agenda that motivates Sentinel&lt;/li&gt;
&lt;li&gt;&lt;a href="https://www.calypso.voyage/" target="_blank" rel="noopener"&gt;Calypso&lt;/a&gt; — submarine cable inference system that uses Sentinel&amp;rsquo;s event database&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;h2 id="people"&gt;People&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;PIs&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a href="https://aqualab.cs.northwestern.edu/author/fabian-e.-bustamante/"&gt;Fabián E. Bustamante&lt;/a&gt; (Northwestern University)&lt;/li&gt;
&lt;/ul&gt;
&lt;/div&gt;</description></item></channel></rss>