Skip to main content

How to proceed to moderation?

GraphComment support team
  • Updated

1) Moderation interface tabs

The Moderation tab in the GraphComment admin has 4 sub-tabs:

  1. Pending
    • Lists all comments awaiting moderation when pre-moderation is enabled.
  2. Approved
    • Lists all comments that have been approved by a moderator in pre-moderation mode and are therefore displayed on the site.
  3. Deleted
    • Lists:
      • pending comments declined in pre-moderation (status “Refused”);
      • comments removed after publication in post-moderation (status “Withdrawn”).
    • In the discussion thread, the removed comment is replaced with the message: “This message has been deleted by the moderator.”
  4. All
    • Lists all comments posted by users on this site, with labels indicating each comment’s status.

mceclip0.png

2) Available moderation actions

Actions available from the interface:

  • Approve: approve a pending comment.
  • Reply and approve: reply to a comment from the admin, immediately publish your reply, and approve the comment.
  • Delete:
    • decline a pending comment (pre-moderation), or
    • remove a comment that is already published (post-moderation).
  • Signal as spam: declines or removes a comment by marking it as spam, in order to enable rules to be created against spammers.

The action bar:

mceclip1.png

3) Pre-moderation and post-moderation (reminder)

  • In pre-moderation, users’ comments are first placed in “Pending”, then a moderator decides whether to approve or decline them.
  • In post-moderation, comments may be visible immediately after publication, and can be removed later by a moderator if necessary.

4) Toxicity filters (add-on, on request)

Objective: detect upfront any comments that create a reasonable doubt, so moderation can be prioritized and/or automated.

Principle:

  • each comment receives an internal toxicity score based on an analysis of language patterns (not a simple dictionary of words that is easy to bypass);
  • thresholds are internal and are not disclosed;
  • the bar is intentionally high: the goal is to avoid human moderation on the safest cases.

Workflow rule (summary):

  • any comment that raises doubt is routed to pre-moderation (the “Pending” queue);
  • only comments with a very favorable score can be published automatically.

Complementarity:

  • GraphComment already provides keyword filters (e.g., insults);
  • the toxicity filter focuses on language and tone, to catch problematic phrasing even without “banned words”.

Availability: on request (add-on, billed separately).

5) Author reputation labels (trend)

Objective: draw the moderator’s attention to authors based on the evolution of their reputation over time, in order to adjust vigilance.

Labels displayed in the moderation interface:

  • freq deleted
    • Triggered when an author’s decline/removal rate exceeds a certain level over a rolling period of roughly one month.
    • Use: flag a potential trolling phase and increase vigilance.
    • The label disappears progressively if the author calms down (trend improves).
  • high rep / low rep
    • Computed from community votes received over a rolling period.
    • Use: provide an immediate indicator to the moderator.

Display: these labels are visible directly on comments during moderation.

6) Contextualized moderation AI (on request)

Objective: help moderate more accurately by taking context into account, rather than evaluating a comment in isolation.

Availability: on request.

Context taken into account (examples, non-exhaustive list):

  • a summary of the article/page where the comment is posted, including themes and moderation watchpoints;
  • the parent comment (and the elements required to understand intent within the thread);
  • any additional information useful to the final verdict, within what is exposed (list intentionally non-exhaustive).

Operating modes:

  1. Full AI
    • The AI can automatically apply the moderation policy according to the configuration.
  2. AI + human (hybrid)
    • The AI makes a decision and, if there is any doubt, it moves the comment to “Pending” so a human can decide in the tool.
  3. AI + outsourced moderation (hybrid) (e.g., Netino, etc.)
    • The AI makes a decision and, if there is any doubt, it sends the case via API to a third-party moderation provider, which moderates it and returns the verdict and the reason via API.

7) Comment distribution per moderator

Objective: distribute the moderation workload across several moderators / internal teams.

What it enables:

  • assign the comments from a given article to a single moderator;
  • or distribute these comments across a team, depending on the newsroom’s organization.

Implementation:

  • assignment can be done manually on a case-by-case basis or via a private API route;
  • then the interface/front can filter the display so each moderator only sees their own queue.

Availability: manual for everyone, and Enterprise plans only via API (on request).

 

Related articles