Figure 1 — the pipeline from search links through Division 1 (content type) and the supplement screen to Division 2 (primary supplement category).
Figure 2 — the same 2,792 supplement-mentioning posts divided by the marketed target. Per the MIMICS plan this is a separate division, not a sub-level of Division 2, so each post appears once in each figure's final column.
| Stage | Rule | In | Kept | Removed |
|---|---|---|---|---|
| Google search | Daily scrape across 13 search terms; links from YouTube, TikTok, Facebook, Instagram | — | 19,394 | — |
| Download + LLM | Video retrieved by yt-dlp and processed by Nemotron 3 Nano Omni; drops unretrievable links (most Instagram), audio-only files and processing failures | 19,394 | 12,177 | 7,217 |
| Menopause screen | LLM field menopause = True (post targets the supplement toward menopause-related symptoms) | 12,177 | 3,903 | 8,274 |
| Deduplication | Same normalised title (≥ 15 chars) and duration across platforms; copy with most views kept | 3,903 | 3,804 | 99 |
| Division 1 — content type | LLM classification: informative / marketing / other material (personal stories or anecdotes): Informative 2,434 / Marketing 834 / Other material 536 | 3,804 | 3,804 | — |
| Supplement screen | Post names at least one supplement; the 1,012 without are retained separately as symptom-and-experience posts | 3,804 | 2,792 | 1,012 |
| Division 2 — primary supplement category | Modal status category of the post's supplement terms, per the team's nine-category coding framework; unmatched terms go to the manual review queue | 2,792 | 2,792 | — |
| Division 3 — target group | Modal marketing/target group of the post's supplement terms. A separate division of the same 2,792 posts, not a sub-level of Division 2 (second figure) | 2,792 | 2,792 | — |
| From | To | Posts | % of source |
|---|---|---|---|
| Downloaded & LLM-processed | 151 | 2.6% | |
| Not retrievable / not processed | 5,764 | 97.4% | |
| TikTok | Downloaded & LLM-processed | 4,813 | 90.8% |
| TikTok | Not retrievable / not processed | 488 | 9.2% |
| YouTube | Downloaded & LLM-processed | 3,834 | 91.9% |
| YouTube | Not retrievable / not processed | 337 | 8.1% |
| Downloaded & LLM-processed | 3,379 | 84.3% | |
| Not retrievable / not processed | 628 | 15.7% | |
| Downloaded & LLM-processed | Menopause-targeted | 3,903 | 32.1% |
| Downloaded & LLM-processed | Not menopause-targeted | 8,274 | 67.9% |
| Menopause-targeted | Unique posts | 3,804 | 97.5% |
| Menopause-targeted | Cross-platform duplicates | 99 | 2.5% |
| Unique posts | Informative | 2,434 | 64.0% |
| Unique posts | Marketing | 834 | 21.9% |
| Unique posts | Other material | 536 | 14.1% |
| Informative | Mentions supplements | 1,675 | 68.8% |
| Informative | No supplements mentioned | 759 | 31.2% |
| Marketing | Mentions supplements | 726 | 87.1% |
| Marketing | No supplements mentioned | 108 | 12.9% |
| Other material | Mentions supplements | 391 | 72.9% |
| Other material | No supplements mentioned | 145 | 27.1% |
| Mentions supplements | Manual review queue | 794 | 28.4% |
| Mentions supplements | Vitamins & minerals | 735 | 26.3% |
| Mentions supplements | Branded menopause formulations | 442 | 15.8% |
| Mentions supplements | Hormone therapies | 242 | 8.7% |
| Mentions supplements | Protein, amino acids & creatine | 193 | 6.9% |
| Mentions supplements | Herbals & botanicals | 178 | 6.4% |
| Mentions supplements | Gut / microbiome | 68 | 2.4% |
| Mentions supplements | Fatty acids & oils | 65 | 2.3% |
| Mentions supplements | Vague or non-classifiable | 49 | 1.8% |
| Mentions supplements | Prescription & OTC medicines | 26 | 0.9% |
| Mentions supplements | General / multisymptom / unclear | 1,365 | 48.9% |
| Mentions supplements | Bone / musculoskeletal support | 707 | 25.3% |
| Mentions supplements | Phytoestrogenic / hormone-modulating | 283 | 10.1% |
| Mentions supplements | Vasomotor / hot-flush relief | 151 | 5.4% |
| Mentions supplements | Gut / inflammation / detox | 99 | 3.5% |
| Mentions supplements | Sleep / stress / mood / cognition | 98 | 3.5% |
| Mentions supplements | Skin / hair / beauty ageing | 47 | 1.7% |
| Mentions supplements | Metabolic / weight / insulin | 25 | 0.9% |
| Mentions supplements | Sexual / vaginal symptoms | 17 | 0.6% |
Bands are proportional to post counts; nodes smaller than 123 posts are drawn at a 4px minimum so they stay visible. The post-level data behind every node is exported to data/pipeline_stages.xlsx — its recipes sheet states the exact filter that reproduces each number here. Division 2 and 3 assignment gives each post its most frequent code among its supplement terms (ties resolved away from the catch-all group). Generated by src/generate_sankey.py from data/supplements.csv (search links), data/supplements_LLM_results.xlsx (LLM output), src/clean_supplements.py (category and target coding) and src/analyze_data.py (platform filter and deduplication).