Building a Media Ecosystem Observatory from Scratch: Infrastructure, Methodology, and Insights

Executive Summary

Understanding the flow of information across today's fragmented digital media landscape requires scalable, cross-platform infrastructure. This paper presents the Canadian Media Ecosystem Observatory (MEO), a national-scale infrastructure designed to monitor political and media discourse across platforms in near real time.

The MEO data infrastructure features custom crawlers for major platforms, a unified indexing pipeline, and a normalization layer that harmonizes heterogeneous schemas into a common data model. Semantic embeddings are computed for each post to enable similarity search and vector-based analyses such as topic modeling and clustering. Processed and raw data are made accessible through an API, dashboards, and a website, supporting both automated and ad hoc research workflows.

The paper illustrates the utility of the observatory through example analyses of major Canadian political events, including Meta's 2023 news ban and the recent federal elections. As a whole, the system offers a model for digital trace infrastructure and an evolving research platform for studying the dynamics of modern media ecosystems. The system tracks 5,515 unique entities across six major platforms.

Key Takeaways

• MEO's seed list comprises 5,515 unique entities across six platforms: politicians (federal and all provincial), news outlets, political influencers, government organizations, civil society organizations, and foreign state actors from the US, Russia, India, and China.
• The system collects from X/Twitter (4,211 accounts), Facebook (3,876), Instagram (3,254), YouTube (1,332), Bluesky (405), and TikTok (554), with platform-specific crawlers on bi-weekly to weekly schedules.
• Each post receives semantic embeddings from transformer-based language models, enabling similarity search, topic modeling, clustering, and cross-platform narrative tracking.
• The architecture stores full raw JSON in Amazon S3 alongside a normalized cross-platform schema in Elasticsearch, enabling both granular platform-specific and integrated multi-platform analysis.
• The paper demonstrates system utility through analyses of Meta's 2023 news ban and the 2025 federal election, and offers the architecture as a replicable model for national-scale media observatories in other countries.