DeepSeekers

Overview & Problem Statement

With the recent years' uptake in short-form content consumption, TikTok seeks to grow its revenue by incentivising content creators through programmes/features like Creator Rewards, TikTok Pulse, TikTok Live coins. However, there exists serveral issues with the current value-sharing system: both its reward mechanism and currency system.

1. Lack of ad revenue for smaller content creators

At least 96% of short form (<60s) TikToks do not qualify for monetisation from ad revenue (Tik Tok Pulse & Creator Rewards Program).
86% of all TikToks are less than 60 seconds. Short form content dominates the platform yet are excluded from Creator Rewards Program (more than 60 seconds)
Eligibility criteria for TikTok Pulse and Creator Rewards Program, such as 100k and 10k followers respectively creates a high barrier of entry for smaller content creators in terms of generating ad revenue
Since smaller content creators will not receive ad revenue despite quality TikToks, high-engagement content creators are prioritised in the short term at the expense of long-term growth via new creators

2. Lack of transparency in revenue splits

While creators can view raw video analytics data, the ad revenue payout doesn't provide a clear breakdown of intermediate metrics contributing to the final amount, leaving creators in the dark on how to improve video performance
Likewise, TikTok does not disclose the cut of revenue from live gift rewards (coins) it keeps.

3. Money Laundering Operations

TikTok Live's coin gift rewards is vulnerable to money laundering attempts
Allegations that TikTok is not doing enough to monitor suspicious transactions and establish Know Your Customer verification for users (Utah lawsuit)

4. Limited payment access and fragmented in-platform spending

Users spend money in multiple separate ways (TikTok coins, TikTok Shop, memberships, etc.) leading to friction and inefficiency
Payout options (e.g. bank transfer, Paypal, certain wallets) are not accessible in all countries, so creators struggle to receive earnings easily or cheaply

Our Solution

1. Multi-Objective Optimisation (Ad Revenue Distribution)

We propose allocating the ad revenue pool back to the TikToks that generated it through a multi-objective optimisation framework.
This balances:

Efficiency: payouts are linked to effective impressions (views × quality score Q), rewarding content that drives advertiser value.
Fairness: a logarithmic utility enforces diminishing returns so that rewards are not overly concentrated in just the largest creators.
Sustainability: a platform revenue floor ensures TikTok retains sufficient resources for infrastructure, safety, and growth.

Importantly, TikTok retains direct control over the platform cut: before optimisation begins, TikTok decides what fraction of the ad pool to keep. The remaining creator pool is then distributed across eligible videos using the optimisation model.

How it works:

Set the pool – TikTok first keeps its platform share; the remainder becomes the creator pool P.
Score content – For each video, a quality score Q_v is computed from watch completion, engagement rate/diversity, rewatch behaviour, and NLP safety. Effective impressions are calculated as M_v = Views_v × Q_v.
Optimise payouts – The model chooses payouts p_v subject to:
- Budget constraint: ∑ p_v = P
- Fairness weights: w_v = s_v^α where s_v is baseline share and 0<α≤1 flattens distribution
- Floors/caps and compliance gates to enforce policy.
  The objective maximises a weighted sum of fairness (∑ w_v log(p_v)) and efficiency (∑ s_v (p_v / P)), with tunable λ-parameters.
Dynamic updates – The optimiser queries real video stats from the database and updates projected earnings in real time (adrev_opti.py).

Why this matters:

For creators: rewards are tied to both reach and quality, giving smaller or niche creators a fairer path to monetisation.
For TikTok: ensures platform sustainability while providing levers (λ-values, α, platform cut) to tune outcomes.
For advertisers: links spend to quality-adjusted engagement, strengthening trust in the platform.
For the ecosystem: creates a transparent, principled, and auditable system for sharing value across millions of short-form videos.

Note that the parameters (λ-values and α) act as policy levers. Their values need to be tuned on large volumes of TikTok data to ensure the framework achieves fairness, efficiency, and sustainability at scale.

For detailed formulations, case studies, and interpretations, see the PDF in the ZIP folder or GitHub titled "Ad Revenue Distribution Formulation & Interpretation".

2. Multi-Objective Optimisation (Donations and Rewards from Viewers)

For viewer donations (coins, gifts, tips), we optimise how much goes to creators versus the platform.

Normal coins (from unverified users) have lower weight.
Premium coins (ByteCoin) (from verified wallets) are given higher creator fractions to discourage bots and promote trust.
A video’s quality score (Q) scales its utility, ensuring genuine, high-quality videos earn more.

The intention of the model is to prioritise and promote ByteCoin. Since ByteCoin transactions are tied to verified wallets and integrated with KYC/AML checks, encouraging its use not only rewards creators more generously but also reduces fraud and money-laundering risks.

How it works:

Platform Cut Control – TikTok still decides how much of donations to retain upfront (for infra and compliance). The remainder becomes the creator pool for this video.
Optimisation Variables – For each video, the model chooses creator fractions x_n (Normal coins) and x_p (ByteCoin). Policy constraints ensure x_p >= x_n + δ, so premium donations always reward more than normal donations.
Objective Function – The optimisation maximises a weighted sum of:
- Platform sustainability (R): TikTok keeps enough to cover costs and fraud monitoring.
- Creator utility (U): scaled by quality score Q and donation amount, with diminishing returns (log utility) to prevent one video from taking all.
- ByteCoin incentive (I): rewarding higher gaps between x_p and x_n to make premium donations more attractive.
Dynamic Updates – The model queries actual video stats from the DB and updates payouts in real-time (donate_opti.py).

Why this matters:

For creators: higher trust donations (ByteCoin) give them larger payouts, motivating them to encourage verified supporters.
For viewers: verified fans feel safer knowing their donations are transparent and harder to spoof.
For TikTok: ByteCoin adoption is promoted, fraud risks are reduced, and the donation ecosystem becomes more sustainable.

This ensures donations are not just raw transactions, but part of a structured, fraud-resistant, and quality-driven ecosystem.

Similarly, the parameters here must be tuned using large amounts of TikTok video data.

Implementation details are available in our prototype donate_opti.py. For more detailed formulation and interpretation, refer to the PDF titled "Donation Coin Split Formulation & Interpretation".

1.1 / 2.1 Quality Score Formulation

The quality score Q_v is used as an input variable in both optimisation models. It combines trust-weighted engagement signals into a single measure:

Q_v = C_v * min(1, W_v^φ_W * E_v^φ_E * D_v^φ_D * R_v^φ_R * S_v^φ_S)

where:

W_v = watch completion
E_v = engagement rate
D_v = engagement diversity
R_v = rewatch/retention
S_v = NLP-based safety/quality
C_v ∈ {0,1} = compliance gate
weights φ_W, φ_E, φ_D, φ_R, φ_S ≥ 0 with ∑ φ = 1

Why it matters:

Rewards not just views but genuine, diverse, and safe engagement.
Penalises low-quality or policy-violating content through compliance gates.
Harder to game since all components must align for a high score.
Provides a transparent, formula-driven signal that directly feeds into fairer revenue and donation splits.

3. ByteCoin

ByteCoin is a proposed blockchain-based stablecoin that serves as the central currency within the ByteDance ecosystem.
It enables transparent, formula-driven value sharing, helps verify users to prevent viewbotting and fraudulent accounts, and allows creators — especially those in regions with limited banking access — to monetise more easily.

Key Features:

Stable & trusted: operates as a stablecoin pegged to the US dollar.
Fast & secure: built on the high-performance SUI blockchain for scalable transactions.
Compliance-first: all transactions comply with global AML/KYC regulations via integration with Notabene.
Smart contract flexibility: ByteCoin can dynamically manage multiple pools, each with tailored allocation rules (e.g., gifting, ad-revenue distribution, TikTok Shop transactions, reserve management).
Unified ecosystem currency: functions across all ByteDance platforms (TikTok, Lemon8, CapCut, etc.), promoting seamless in-ecosystem spending and cross-platform payments.
Verification layer: user crypto wallets double as identity verification, discouraging fraudulent bot accounts and improving trust.

Challenges

Since SUI SDKs were quite new, it could not work with Lynx. The unfamiliarity with web3 and blockchain also had a steep learning curve to get used to.

What it does

Slush wallet was meant to act as a verification to separate verified users and normal users, which will affect the coin splits. ByteCoin acts like a premium coin, where only the verified users can buy and gift to creators, which has a higher value compared to normal tiktok coins. By using ByteCoins, we also increases the transparency of TikTok's splits and the value-sharing system.

How we built it

The solution was built on the Lynx framework, leveraging its dual-threaded architecture and main-thread scripting for speed and performance. We sought to replicate TikTok's design with its numerous interactive components, but with newly improved features like ad revenue breakdown.

In the backend, multi-objective optimisation was implemented via Gurobi to optimise for the ideal ad revenue and coin revenue splits. An optimisation process would be triggered by each video upload.