Social Comment Classifier — purchase intent, questions & sentiment

For brands, social media managers, agencies and AI agents that already scrape comments — from an Instagram, TikTok, Facebook or YouTube comment scraper, any other Apify dataset, or pasted as text — and need every comment labelled by what it is, for $0.50 per 1,000 comments, with no prompt writing or LLM key.

• Comment type — purchase_intent · question · complaint · request · praise · spam · other (one main type, with a probability for every type)
• Sentiment — positive / neutral / negative
• Needs reply — whether the brand or creator should answer (questions, problems, "how do I buy?")
• Your own labels — up to 10 yes/no criteria in plain language (e.g. "mentions shipping or delivery", "asks about a discount code")

Use it to: find buyers in your comments ("price?", "link?", "when is the restock?") · answer unanswered questions · catch complaints before they spread · hide spam and fake-shop promotion · collect product requests · Instagram, TikTok and YouTube comment sentiment analysis at scale · compare comment mix across posts or competitors.

Output sample

Real rows from run PgvMClRhIRV9gKmp7 (2026-10-09): public comments from the four comment scrapers below, with one custom label ("mentions shipping or delivery").

comment	source	type (probability)	sentiment	needs reply	custom: shipping
When will you restock Generation G Fuzz?	Instagram	purchase_intent (1.00)	neutral	yes	no
Link	Facebook (live shopping)	purchase_intent (0.56)	neutral	no	no
How much do they hold?	Instagram	question (1.00)	neutral	yes	no
Temu, you guys are up over here making tiktoks. When im still am waiting for my package…	TikTok	complaint (1.00)	negative	yes	yes
I would like a refund for the 2 weeks the towers were down in my area.	Facebook	complaint (0.94)	negative	yes	no
can y'all make the resurfacing retinol serum in a bigger bottle ?	TikTok	request (1.00)	neutral	yes	no
obsessed with every single one of these!	Instagram	praise (1.00)	positive	no	no
For me personally… I do buy quality fakes/knockoffs at (shop name)	YouTube	spam (0.72)	neutral	no	no

Each row also keeps the comment ID and post fields you choose (id, cid, postUrl, videoWebUrl, …), and in full mode typeScores with the probability of every type.

Input example

The form default — two pasted comments, no dataset needed (about $0.001, 2 seconds):

{
  "texts": [
    "Where can I buy this in Canada? Need it!!",
    "Ordered 3 weeks ago and still nothing. Is anyone answering messages?"
  ]
}

To classify a comment scraper's output, pass its dataset instead — the text, post title and ID fields are detected automatically:

{
  "datasetId": "YOUR_COMMENTS_DATASET_ID",
  "customLabels": ["mentions shipping or delivery"]
}

Pricing

Pay only for classified comments — no subscription.

Event	Price	When
comment-judged	$0.0005	One comment classified (type such as purchase intent, question or complaint; sentiment; needs-reply flag and any custom labels).

That is $0.50 per 1,000 comments. First run with the form defaults: about $0.001 (2 comments, 2 seconds).

Cost examples

Comments	Cost
100	$0.05
1,000	$0.50
10,000	$5.00
100,000	$50.00

With the free $5 monthly Apify credit you can classify about 10,000 comments.

Items without comment text are skipped and not charged. Comments that fail after retries are reported with an error field and not charged. If you set a maximum cost per run, the Actor stops cleanly when it is reached.

Works with

The most used comment scrapers on Apify Store — run one, then pass its dataset to this Actor. Field detection was checked against real output of each (2026-10-09, 20 comments per scraper).

Platform	Scraper on Apify Store	Text field	Post title used	IDs kept
Instagram	Instagram Comments Scraper (apify)	text	—	id, commentUrl, postUrl
TikTok	TikTok Comments Scraper (clockworks)	text	—	cid, videoWebUrl
Facebook	Facebook Comments Scraper (apify)	text	postTitle	commentId, id, commentUrl, facebookUrl
YouTube	YouTube Comments Scraper (streamers)	comment	title (video title)	cid, videoId

Also: the Apify API and JavaScript/Python clients · Apify Schedules (e.g. classify new comments every morning) · Claude, Cursor and Claude Code through the Apify MCP server (next section) · for product reviews, our Korean, Japanese and AliExpress review classifiers.

Use with Claude, Cursor or Claude Code (MCP)

Add the Apify MCP server with this Actor as a tool and ask your agent in plain language — for example "Which of these comments are from people who want to buy? …" or "Classify dataset abc123 from my TikTok comments run and list the unanswered questions." The agent calls the tool leoworks--social-comment-classifier and reads the labels with get-dataset-items.

Claude Desktop or Cursor (mcp.json):

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com?tools=leoworks/social-comment-classifier",
      "headers": { "Authorization": "Bearer YOUR_APIFY_TOKEN" }
    }
  }
}

Claude Code: claude mcp add --transport http apify "https://mcp.apify.com?tools=leoworks/social-comment-classifier" --header "Authorization: Bearer YOUR_APIFY_TOKEN". Leave out the header to sign in with OAuth in the browser instead. Your Apify token is in Console → Settings → API & Integrations. We verified this setup with the Apify MCP server (v0.17.4) on 2026-10-09: the agent classified a pasted comment in 6 seconds end to end (run XTdACdD0TchL5eQ34).

Output (one row per comment)

{
  "cid": "7694457506550088461",
  "videoWebUrl": "https://www.tiktok.com/@temu/video/7694358584565583117",
  "text": "Temu, you guys are up over here making tiktoks. When im still am waiting for my package like, give me my freaking package, please.",
  "labels": {
    "type": { "label": "complaint", "probability": 1, "confidence": 1 },
    "sentiment": { "label": "negative", "probability": 1, "confidence": 1 },
    "needsReply": { "probability": 0.84, "matched": true },
    "custom": [{ "label": "mentions shipping or delivery", "probability": 0.98, "matched": true }],
    "typeScores": [
      { "label": "complaint", "probability": 1 },
      { "label": "spam", "probability": 0 }
    ]
  }
}

typeScores lists all seven types (shortened here). Minimal mode returns label keys only ("type": "complaint", "sentiment": "negative", "needsReply": true).

How types are chosen: each comment gets one main type. When a comment fits several, the first in this order wins: spam → purchase_intent → complaint → question → request → praise → other. So "Love it! Where can I buy it in the UK?" is purchase_intent, and "It stopped working after a week, any tips?" is complaint. Use typeScores or Needs reply when you want every comment that asks something.

Summary by post (REPORT)

Each run also saves a REPORT record (Output tab → Summary by post) at no extra charge: for every post, the comment type mix, sentiment shares, the share of comments that need a reply and up to 3 example comments for purchase intent, questions, complaints and requests — plus the same for all comments together. Posts are grouped by summaryGroupField (detected automatically from fields such as postUrl, videoWebUrl, facebookUrl or videoId when empty).

{
  "groupField": "videoWebUrl",
  "groups": [{
    "group": "https://www.tiktok.com/@brand/video/1",
    "comments": 120,
    "classified": 118,
    "types": [{ "label": "praise", "count": 51, "share": 0.432 }, { "label": "purchase_intent", "count": 22, "share": 0.186 }],
    "sentiment": { "positive": 0.55, "neutral": 0.36, "negative": 0.09 },
    "needsReplyRate": 0.31,
    "examples": { "purchase_intent": ["price?", "Do you ship to Canada?"], "question": ["Is it waterproof?"] }
  }]
}

Accuracy

Measured on 205 hand-labelled public English comments from Instagram, TikTok, Facebook and YouTube (beauty brands, a mobile carrier, a retailer's live shopping and product review videos; 2026-10-09). The questions were adjusted on one part, then checked once on a separate part that was not used for adjusting.

Set	Comments	Main type	Main type (either label where two fit)	Sentiment
Check set — not used for adjusting	100	90%	95%	87%
Adjusting set	105	88%	93%	84%

On the check set, complaints were found 97% of the time, questions 100%, spam 86% and purchase intent 71% (7 comments). Short comments without context ("No", "Vibes") are the hardest. Probabilities are calibrated — use typeScores and Label threshold to trade coverage for certainty. Automated labels can be wrong; check samples before making big decisions.

Limits

Item	Limit
Comment length	First 4,000 characters are used (post title: 300)
Custom labels	Up to 10, each up to 200 characters
Dataset size	Any — datasets are read in pages of 1,000
Language	English (measured). Other languages are accepted and usually work, but accuracy has not been measured — treat them as beta
Replies	Each row is classified on its own. Replies nested inside a comment (e.g. an Instagram replies array) are not classified — use the scraper's option to output replies as rows
Speed	About 100 comments in 3–4 seconds
Data	Only the comment text, the post title and your custom labels are sent for classification; usernames and profile fields are not used or output

FAQ

Which AI makes the judgments? Jev, TypeSafe's decision model (version jev-1.13.0, pinned). Jev answers each label with a calibrated probability instead of generated text, so the same input gets the same answer from run to run. Only the comment text, the post title and your custom labels are sent to Jev; usernames and other fields are not.

Does it scrape Instagram, TikTok, Facebook or YouTube? No. It classifies comments you already have — run a comment scraper first (see Works with) or paste texts.

Does it write replies? No. It only assigns labels with probabilities — fast, cheap and consistent. Use Needs reply to decide which comments to answer.

Why request as well as complaint? Product requests ("please make a bigger bottle") are feedback for product teams, not problems for support. They are kept apart so each team gets its own list.

Disclaimer

Independent tool — not affiliated with, endorsed by or sponsored by Meta (Instagram, Facebook), TikTok, Google (YouTube), or by the authors of the scrapers listed above. Names are used only to describe compatible data sources. The comments shown are public comments used as examples.

Reviews and support

If this Actor saved you time, a short review on Apify Store helps others find it. Questions or a dataset whose fields are not detected? Open an issue in the Issues tab — we answer within a day.

Changelog

See the Changelog tab.