✨ Add reporter_stats materialized view and endpoint to fetch reporter stats

[foysalit]

No description provided.

[matthieusieben]

Would it make sense to split the "accounts reports stats" and "record reports stats" in two distinct objects, to be more consistent with the reportee stats ?

[matthieusieben]

Doing this could actually rely on a shared sub-type. Something like:

defs:
  # ...

  reportsCountAggregate:
    description: Aggregated statistics
    type: object
    required:
      - totalCount
      - distinctCount
      - takendownCount
      - labeledCount
    properties:
      totalCount:
        description: The total count of reports in the aggregate (including duplicate reports)
      subjectCount:
        description: The count of distinct subjects
      takendownCount:
        description: The count of reported subjects that were taken down
      labeledCount:
        description: The count of reported subjects that were labeled

  reporterStats:
    type: object
    required: ['did']
    properties:
      did:
        type: string
        description: The DID of the user
      reportedRecordsStats:
        type: ref
        ref: '#reportsCountAggregate'
        description: Aggregated statistics on the user's reported records
      reportedAccountsStats:
        type: ref
        ref: '#reportsCountAggregate'
        description: Aggregated statistics on the user's reported accounts

[matthieusieben]

This will have no effect because this is already present in the global WHERE clause bellow.

[matthieusieben]

After some experimentation with production like data in a database, there might be a more efficient way of doing this. It involves creating 4 partial indexes instead of a materialized view, and splits the query in 4 different queries:

Migration

CREATE INDEX "moderation_event_account_reports_idx" ON "public"."moderation_event"("createdBy","subjectDid") where "subjectUri" IS NULL AND "action" = 'tools.ozone.moderation.defs#modEventReport';
CREATE INDEX "moderation_event_record_reports_idx" ON "public"."moderation_event"("createdBy","subjectDid","subjectUri") where "subjectUri" IS NOT NULL AND "action" = 'tools.ozone.moderation.defs#modEventReport';

CREATE INDEX "moderation_event_account_actions_ids" ON "public"."moderation_event"("subjectDid","action") where "subjectUri" IS NULL AND "action" IN ( 'tools.ozone.moderation.defs#modEventTakedown', 'tools.ozone.moderation.defs#modEventLabel');
CREATE INDEX "moderation_event_record_actions_ids" ON "public"."moderation_event"("subjectDid","subjectUri", "action") where "subjectUri" IS NOT NULL AND "action" IN ( 'tools.ozone.moderation.defs#modEventTakedown', 'tools.ozone.moderation.defs#modEventLabel');

Queries

-- uses "moderation_event_account_reports_idx" index
SELECT
    -- Count total number of reports for accounts (including duplicates)
    COUNT(*) AS "accountReportCount",
    -- Count unique accounts reported
    COUNT(DISTINCT reports."subjectDid") AS "reportedAccountCount"
FROM "moderation_event" AS reports
WHERE reports."action" = 'tools.ozone.moderation.defs#modEventReport'
AND reports."subjectUri" IS NULL
AND reports."createdBy" IN ('did:plc:foo','did:plc:bar')
GROUP BY reports."createdBy";

-- uses "moderation_event_record_reports_idx" index
SELECT
    -- Count total number of reports for records (including duplicates)
    COUNT(*) AS "recordReportCount",
    -- Count unique records reported
    COUNT(DISTINCT reports."subjectUri") AS "reportedRecordCount"
FROM "moderation_event" AS reports
WHERE reports."action" = 'tools.ozone.moderation.defs#modEventReport'
AND reports."subjectUri" IS NOT NULL
AND reports."createdBy" IN ('did:plc:foo','did:plc:bar')
GROUP BY reports."createdBy";

-- uses "moderation_event_account_reports_idx" index
-- uses "moderation_event_account_actions_ids" index
SELECT
    -- Count unique accounts taken down by moderators
    COUNT(DISTINCT actions."subjectDid") FILTER (
        WHERE actions."action" = 'tools.ozone.moderation.defs#modEventTakedown'
    ) AS "takendownAccountCount",
    -- Count unique accounts labeled by moderators
    COUNT(DISTINCT actions."subjectDid") FILTER (
        WHERE actions."action" = 'tools.ozone.moderation.defs#modEventLabel'
    ) AS "labeledAccountCount"
FROM "moderation_event" AS reports
LEFT JOIN "moderation_event" AS actions ON
      actions."subjectDid" = reports."subjectDid"
  AND actions."subjectUri" IS NULL -- explicited to match index 
  AND actions."action" IN ( 'tools.ozone.moderation.defs#modEventTakedown', 'tools.ozone.moderation.defs#modEventLabel')
WHERE reports."action" = 'tools.ozone.moderation.defs#modEventReport'
AND reports."subjectUri" IS NULL
AND reports."createdBy" IN ('did:plc:foo','did:plc:bar')
GROUP BY reports."createdBy";

-- uses "moderation_event_record_reports_idx" index
-- uses "moderation_event_record_actions_ids" index
SELECT
    -- Count unique records taken down by moderators
    COUNT(DISTINCT actions."subjectUri") FILTER (
        WHERE actions."action" = 'tools.ozone.moderation.defs#modEventTakedown'
    ) AS "takendownRecordCount",
    -- Count unique records labeled by moderators
    COUNT(DISTINCT actions."subjectUri") FILTER (
        WHERE actions."action" = 'tools.ozone.moderation.defs#modEventLabel'
    ) AS "labeledRecordCount"
FROM "moderation_event" AS reports
LEFT JOIN "moderation_event" AS actions ON
      actions."subjectDid" = reports."subjectDid"
  AND actions."subjectUri" IS NOT NULL -- explicited to match index definition
  AND actions."subjectUri" = reports."subjectUri"
  AND actions."action" IN ( 'tools.ozone.moderation.defs#modEventTakedown', 'tools.ozone.moderation.defs#modEventLabel')
WHERE reports."action" = 'tools.ozone.moderation.defs#modEventReport'
AND reports."subjectUri" IS NOT NULL
AND reports."createdBy" IN ('did:plc:foo','did:plc:bar')
GROUP BY reports."createdBy";

There does not seem to be any noticeable effect on insert latencies due to the new indexes. Note that any insert in the moderation_event table triggers at an update of at most one of the newly added indexes.

[matthieusieben]

That being said, I think that the strategy that is applied here can be abused by a user to artificially increase their statistics, and subsequently, their perceived reporter credibility by moderators.

Indeed, it is very easy for a user to just send reports on subjects already labelled / takendown. This behavior should probably actually be punished, which will not be the case with these changes as they are.