Search Youtube

Search filters for YouTube caption chunks.

The query field is evaluated against the chunks' text_search_fts column via websearch_to_tsquery('english', ...) — stemming on, websearch syntax (quoted phrases, OR, leading - to exclude).

Institution linkage uses BOTH routes signals uses, results unioned: • Channel route: raw.channel_id → youtube_links.channel_id → unique_id • Video route: raw.video_institution_unique_id (direct per-video tag)

The signals retriever uses phraseto_tsquery because its API takes programmatic keyword groups. Our user-input API uses websearch to match the rest of the chunk endpoints. For tight phrase matching, consumers quote the phrase: query='"school board meeting"'. Cover-density ranking (ts_rank_cd) already prefers chunks where query terms appear close together.

At least one of institution_ids or query must be supplied — the chunks table is large and unbounded scans are rejected.