ナビゲーションにスキップ

How many usable TikTok comments did I actually collect, and is my budget sufficient?

The number of TikTok comments you collected is the number of records returned successfully, after deduplicating by comment ID. Two other numbers are not the collected count: the request’s count (default 20) is the number you asked for, and the video’s comment_count is the total the video reports. Record the three numbers in separate columns.

For example, you request 20 records and receive 16, two of which are duplicates: you collected 14 distinct comments. Then mark which of the 14 comments address your research question.

What do the three numbers mean?

NumberWhat it establishesWhat to avoid
Requested countThe page quantity you requestedDo not report count × pages as completed collection.
Video comment_countThe video statistic returned at observation timeThe statistic does not establish visible top-level comments, complete replies or unique people.
Successfully retrieved unique commentsThe distinct records actually obtainedDistinct records are not automatically valid needs; relevance, text and context still require review.

For a hypothetical request of 20 that returns 12 records, two of which already appeared on an earlier page, the page adds only ten distinct records. Irrelevant records and records without text reduce the usable theme evidence further. The example shows how to count; the example does not promise a response size.

Check usefulness before buying another page

After one page, check four things: the target video, the relevance of the comments, readable text and whether replies are needed. If the discussion mostly concerns editing or background music, more pages may not answer the product question; reconsider the video selection first.

A response with fewer records than count does not by itself prove data loss, and a larger displayed video total does not reveal the uncollected text. When actual records, IDs or continuation evidence are missing, record the uncertainty instead of asking AI to fill the gap.

Report top-level and reply coverage separately. A reply that appears both in a preview and on a separate reply page counts once. Deduplicate by verified identifiers and sources, not by matching text or display names. Automatic output mapping for the comment action is not verified yet; see the comment research guide.

How should I budget and stop?

Set the collection budget as at most P approved pages × the current comment-action price, and list search, link resolution, details, replies and retries as separate costs. P limits calls; P does not guarantee a sample size. Stop at the page or budget limit, on a failure, on a repeated page or when no usable continuation is returned; do not keep paging toward the displayed total.

Report the planned pages, the successful pages, the unique records, the human decisions (included, excluded, unverified) and the scope you did not collect. Human relevance review supports observations about the sample, not proportions of the whole audience. Rereading the originals, changing labels and recalculating results need no new collection.

How do video metrics provide context?

The checked share-detail path data.aweme_details[].statistics provides digg_count (likes), collect_count (saves), share_count (shares), play_count (plays) and comment_count (the comment statistic). Keep each value with the observation time. A missing value is not zero, plays are not unique people, and saves are not purchases. The difference between two observations is a net counter change, not a list of events.

Video counters give context and contain no comment text. To save the counters with sources and presence states, follow the video snapshot workflow.

Free Inspect references: data:tiktok.app.v3.fetch_video_comments, data:tiktok.app.v3.fetch_video_comment_replies and data:tiktok.app.v3.fetch_one_video_by_share_url_v2. Sources: the local input contracts and observed response structures.