Social Comment Classifier — buying intent & questions · $0.5/1k
Pricing
from $0.43 / 1,000 comment classifications
Classify Instagram, TikTok, Facebook, YouTube comments: buying intent, sentiment — Instagram comments sentiment and TikTok comments analyzer. Use for comments; for reviews use leoworks/korean-review-classifier. datasetId (text auto-detected) or texts. ~100 comments/4 s, pay per comment.
An Apify dataset containing comments — for example the output of an Instagram, TikTok, Facebook or YouTube comment scraper. Use this OR “Comment texts”.
Paste comment texts directly (one per line). Use this OR “Comments dataset”.
Up to 10 extra yes/no labels (each up to 200 characters) written in plain language, e.g. “mentions shipping to Canada”, “asks about a discount code”. Each gets a probability.
[]0 = all. Comments without a like count are kept.
Classify at most this many comments per post or video, in dataset order. 0 = no limit.
Classify only top-level comments, not replies to other comments.
Skip comments flagged by the scraper as written by the channel owner or creator (e.g. YouTube authorIsChannelOwner), plus any account listed in Exclude accounts.
Usernames whose comments are skipped — usually your own brand or support accounts (with or without @).
[]Skip comments with no letters or digits ("🔥🔥", "❤️"). They are mostly praise; skipping them saves cost.
Classify at most this many comments (0 = all).
Add the commenter's public username to each row and save BUYERS.csv (one row per commenter with purchase intent). Off by default. Usernames are only copied from your dataset, never sent for classification. Use them in line with the platform's terms and privacy law.
Datasets that keep replies inside a comment (Instagram replies, Facebook comments) or comments inside a post (Instagram Scraper latestComments) are flattened to one row per comment.
Field holding the comment text. Leave empty to auto-detect (text, comment, content, body, …).
Field holding the title or caption of the post the comment was written under (helps with short comments like “link?”). Leave empty to auto-detect (postTitle, title, …).
Fields copied unchanged from each input item to the output so you can join results back (e.g. id, cid, postUrl). Leave empty to auto-pick the comment ID and post URL.
[]Minimum probability (0–1) for the needs-reply flag and custom labels to be marked as matched. Raise it for fewer, surer matches.
Minimal mode returns only label keys — smaller and easier to aggregate.
Field in your dataset that identifies the post (e.g. postUrl, videoWebUrl, videoId). The run then saves a REPORT record with, per post: comment type mix, sentiment shares, share needing a reply and example purchase-intent, question, complaint and request comments. Leave empty to detect it automatically (overall summary only if none is found). No extra charge.
Comments classified in parallel.
Only comments newer than
commentsNewerThan
Optional
Classify only comments posted after this date — a date like
2026-10-01or a period like7 days,4 weeks,3 months. Relative dates such as "3 days ago" (YouTube) are counted back from the run time. Leave empty for all.