Backers100
MethodologyCurrent algorithm version v1.0.1

How we measure

Backers100 Score is a partitioned sum of six indicators. Every algorithm change is published in the changelog.

1. What Backers100 is

Backers100 is a data infrastructure that measures the influence of K-celebrities and K-brands using AI algorithms. It is not just a ranking — the basis of every score is published alongside it.

  • Backers100 analyzes the influence of K-celebrities — both individuals and groups.
  • The measurement combines news coverage, search trends, fan signals, positive impact, and brand momentum.
  • Scores are produced on a 0–100 scale once per day.
  • Every score is published alongside its per-indicator breakdown.

Three differentiators

Data assets

We accumulate proprietary datasets across fan engagement, giving history, and issue lifecycle. Depth compounds over time.

Explainability

Every score answers "why this score." The algorithm version is public so AI search engines can cite the result.

Trust

Zero donation fees, an explicit correction workflow, and a permanent algorithm changelog. The operation is built to be verifiable by media, B2B, and legal teams alike.

2. What we measure

Backers100 Score is the direct sum of six indicators. Each indicator contributes up to its own maximum — no separate multiplicative weights are applied.

IndicatorMaxInput
Media Influence30Source-weighted coverage volume and reach
Positive Impact25Public-good coverage, donations, and campaigns
Momentum20Day-over-day and week-over-week change and spread
Search Interest15External search volume and on-site views
Fan Signal5Weighted fan support and giving
Brand Power5Endorsements and brand-celebrity matching
Media Influence30Positive Impact25Momentum20Search Interest15Fan Signal5Brand Power5
The six maxima sum to 100 points. Bar length shows each indicator's weight.

Scores are reported with two decimal places (e.g., 92.43).

3. Data sources

All sources used for measurement are disclosed; per-outlet credibility scores are reviewed on a regular cadence.

  • Naver News

    Primary source for Korean-language coverage.

  • Google News

    Primary source for English-language coverage.

  • Naver search volume

    Used to measure changes in external search interest.

  • On-site fan signals

    On-site support and giving activity reflected as a fan-engagement signal.

Every article is weighted by its outlet's credibility score.

Source articles are preserved as recorded and are not modified after the fact.

Coverage is initially Korean- and English-language outlets; other languages are added in phases.

4. Backers100 Score formula

Backers100 Score is the plain sum of the six indicator scores. There are no multiplicative weights on top.

  1. 01
    Ingest

    Raw signals are stored immutably in a 30-day archive window.

  2. 02
    Normalize

    Normalize across indicators so signals at different scales can be compared fairly.

  3. 03
    Map

    Each indicator's maximum (Media 30 / Positive 25 / Momentum 20 / Search 15 / Fan 5 / Brand 5) maps proportionally.

  4. 04
    Sum

    The six indicator scores sum directly to a 0–100 final score.

Total = Media + Positive Impact + Momentum + Search + Fandom + Brand (30 + 25 + 20 + 15 + 5 + 5)

Typical distribution — rank 1 around 95–96, rank 10 around 93, rank 100 around 85.

Every score is stored with its 6-indicator breakdown so the reasoning is transparent.

5. Article pipeline (7 steps)

Each article passes through seven steps before it influences a score. Every step is traceable for operator review.

  1. 01
    Collect

    Articles are collected from major news outlets on a regular cadence and stored separately.

  2. 02
    Deduplicate

    Duplicate articles are filtered out by URL so the same piece is counted only once.

  3. 03
    Self-mention check

    Only articles whose title mentions the celebrity by name, English name, group, or official alias enter scoring.

  4. 04
    Outlet credibility

    Outlets are auto-classified into tiers (major dailies / wires / broadcast, entertainment / sports / portal, others) with different weights.

  5. 05
    Content classification (AI)

    AI analyzes article content and flags positive, neutral, or risk signals. Auto-classification is reviewed on a regular cadence to maintain quality.

  6. 06
    Daily scoring

    Each day at dawn (KST) the previous 30 days of coverage are aggregated into six indicator scores.

  7. 07
    Final ranking

    Indicator scores are mapped to their maxima and summed for a 0–100 final score with the day's rank.

Each step is traceable by the operations team, and AI classification is regularly sampled and audited for quality.

6. Fan Engagement Index

Fan Engagement Index is a 100-point user-side metric capturing the cumulative engagement of an individual fan across five dimensions. It measures depth and consistency rather than raw support counts.

AxisMaxMeaning
Participation period20How long the fan has been active since signing up
Support depth25Concentration on the most-supported celebrities
Consistency20Regularity and frequency of activity
Participation diversity15Breadth across categories
Activity mix20Balance of support, giving, and campaign participation
Participation period20Support depth25Consistency20Participation diversity15Activity mix20
The five maxima sum to 100 points. Bar length shows each item's weight.

Tiers are operated in four levels: Bronze, Silver, Gold, Diamond.

Formula non-disclosure

The detailed formula for Fan Engagement is not published — a deliberate decision to deter abuse. The measured dimensions and tier thresholds remain public.

7. Monitoring policy

Risk monitoring detects negative signals. Public surfaces only ever show the normal state or the "Under monitoring" badge.

Risk levelInternal labelPublic display
0NormalNo display
1MonitoringGray badge "Under monitoring"
2CautionNo display (B2B only)
3AlertNo display (B2B only)

Forbidden wording (public surfaces)

controversyscandalcrisisdisgraceshocking

We avoid assertive wording. Our copy is grounded in verifiable public information and neutral phrasing.

8. Limits of AI analysis

Backers100 uses AI algorithms, and every auto-classified result is operated on the assumption that misclassification is possible.

  • All AI-generated text is labeled "AI-generated."
  • AI auto-classification results are regularly sampled and audited for quality.
  • If anomalies are detected, an entity may be temporarily made non-public.
  • Anyone can file a correction request if the input data appears wrong.
  • Feedback about the algorithm itself can be submitted via the correction request menu.

9. Corrections and feedback

Channels available when you disagree with data, scores, or wording.

Input-data correction

Article-matching errors, entity-matching errors, sentiment-classification errors are handled in the correction channel.

Correction policy
Algorithm feedback

Opinions about the algorithm itself are collected in a separate channel. Responses are not guaranteed, but meaningful feedback is considered for the next patch.

Send feedback
Media-credibility dispute

Disagreements with per-outlet credibility scoring are handled in the correction channel.

Correction policy

Anonymous submissions are accepted, but if no reply channel is provided, follow-up may not be possible when more material is needed.

10. Collection frequency

Data collection, score calculation, and update cadence work as follows.

  • News archive: daily + every 5 minutes (Naver News + Google News RSS)
  • Search trends: daily (Naver DataLab + Google Trends)
  • Fan signals: real-time (recorded on cheer submission)
  • Donations: real-time (Backersby campaign ledger)
  • Daily orchestrator: 01:50 KST every day
  • Daily aggregation: 02:30 KST every day
  • Realtime trending ticker: every 5 minutes

11. How AI summaries are generated

Per-celebrity AI summaries are auto-generated daily through the following pipeline.

  • Input: 30-day archive articles + 6-indicator scores + day-over-day change
  • Model: OpenAI gpt-4o-mini (cost/accuracy balance)
  • Output policy: fact-based, no negative assertion, no speculation about unknown facts
  • Human review: Risk Watch L2+ signals trigger auto-private + human-review queue
  • Correction: user-submitted corrections receive first response within 24h, fact-checked and revised immediately
  • Source label: every summary explicitly states 'AI-generated'

12. Fan signal abuse prevention

Policies to prevent abuse of fan signals (cheers, engagement).

  • Identity: SSO (Google) sign-in required, anonymous cheering disabled
  • Rate limit: max 10 cheers per IP/account per celebrity per hour
  • Duplicate prevention: identical repeated messages auto-invalidated
  • Pattern detection: rapid bursts raise the AI bot suspicion score
  • Engagement weight: new sign-ups weighted at 0.5 (returns to 1.0 after 60 days)
  • Reporting: user reports of inappropriate cheers reviewed within 24h
  • Forced invalidation: confirmed abuse = weight 0 + account warning

10. Algorithm change log

Every algorithm change is permanently published on the changelog page. Notification timing depends on the change type.

Change scaleExampleNotice timing
SmallMinor weight adjustmentWithin 7 days after activation
MediumFormula or distribution change7 days before activation
MajorIndicator structure overhaul30 days before activation

Frequently asked questions

What is Backers100?
Backers100 is a data infrastructure that measures the influence of K-celebrities and K-brands using AI algorithms. News, search trends, fan signals, positive-impact activity, and brand momentum are synthesized into six indicators that sum to a 100-point Backers100 Score.
How is Backers100 Score calculated?
It is the sum of six indicators — Media Influence 30, Positive Impact 25, Momentum 20, Search 15, Fan Signal 5, Brand Power 5. Each indicator is computed from the last 30 days of coverage, search, and fan-activity data and normalized so that signals at different scales can be compared fairly.
Which data sources are used?
Naver News, Google News, Naver search volume, and on-site fan activity. Every outlet has a regularly reviewed credibility score that is applied as a weight.
What is Monitoring?
Risk monitoring tracks negative signals. Public surfaces only show the normal state or a gray "Under monitoring" badge. Higher-severity states are disclosed only to B2B subscribers.
How do I request a correction?
Input-data errors and feedback about the algorithm itself can be submitted via the correction request menu. Anonymous submissions are accepted.
How are algorithm changes announced?
Small changes are announced within seven days of activation. Medium changes are announced seven days in advance. Major changes are announced thirty days in advance. The audience varies by change scale and the user's Fan Engagement tier.
What are the limits of AI analysis?
AI auto-classification can be wrong. Every AI-generated text is labeled "AI-generated," and regular random-sample audits maintain quality. If input data appears incorrect, anyone may file a correction request.

Backers100 rankings are calculated from public media coverage, search trends, fan signals, positive impact, and momentum data. Some entries may be adjusted through the data review process.