# MemeChain: A Multimodal Cross-Chain Dataset for Meme Coin Forensics and Risk Analysis

Crypto.world

> Meme coins are one of the most active and least studied corners of crypto: thousands launch constantly, most rely on a website and social channels rather than any real technology, and a large share…

**arXiv**: 2601.22185

**Published**: 2026-01-28

**Authors**: Alberto Maria Mongardini, Alessandro Mei

## AI summary

## Overview

**Research area:** Cryptocurrency security and forensics, specifically the meme coin ecosystem; multimodal dataset construction; web forensics; survival analysis.

**Technical level:** Intermediate. The paper is readable without deep cryptographic background, but familiarity with blockchains, DEXs, and token standards (ERC-20, SPL) helps.

**Scope (1 sentence):** The authors build and analyze MemeChain, a 34,988-token cross-chain meme coin dataset that pairs on-chain records with off-chain artifacts (website HTML, logos, social handles) across Ethereum, BNB Smart Chain, Solana, and Base, and use it to quantify how quickly these projects die.

## What This Paper Is About

Meme coins are one of the most active and least studied corners of crypto: thousands launch constantly, most rely on a website and social channels rather than any real technology, and a large share are outright scams. Existing datasets are either single-chain, transaction-only, or built around *successful* tokens, so they miss the failures and the off-chain marketing that promoters use to fake legitimacy. The authors fill that gap by assembling a large cross-chain dataset that also archives website source code, logos, and social links, then analyzing survival, web abandonment, and security signals across it.

## Key Contributions

1. **A large multimodal, cross-chain dataset.** MemeChain covers 34,988 meme coins (31,811 with validated contract addresses) across Ethereum, BSC, Solana, and Base, totaling over 1.46 GB, and links on-chain records to token logos and archived raw HTML source code of project websites.
2. **Lifecycle and extinction analysis.** The paper measures token longevity using transaction timestamps collected until mid-January 2025 (about three months after the mid-October 2024 collection), and defines the "One-Day Meme Coin" — a token that ceases all trading within 24 hours of launch.
3. **First systematic web-infrastructure forensics for meme coins.** The authors resolve shortened URLs, deduplicate domains, pull WHOIS registrars, test HTTP reachability, archive HTML, and run Google Safe Browsing and ChainPatrol over the sites.
4. **A research roadmap.** The paper outlines uses for the data: multimodal scam detection (phishing-kit reuse, logo impersonation via Siamese networks, landing-page/on-chain incongruence), early survival prediction, and cross-chain studies of capital and fraud migration.

## Main Findings

- **Chain distribution is lopsided.** Of the 34,988 tokens, BSC hosts 16,405 (46.88%), Solana 11,829 (33.8%), Ethereum 3,814 (10.9%), and Base 828 (2.36%). These four chains account for 32,876 tokens (93.96%), which is why all later analysis is restricted to them.
- **Nearly half the tokens went silent within two months.** 15,099 of 31,811 tokens (47.46%) had no transaction activity after December 15, 2024. BSC had the highest extinction rate at 80.96% (12,503 of 15,442), followed by Ethereum at 37.32%, Base at 16.20%, and Solana at 9.09%.
- **Short lifespans are common.** 4,939 tokens (15.53%) had a lifetime of 30 days or less, and 3,512 (11.04%) lasted 7 days or less, where lifetime is the interval between first and last recorded transaction.
- **The One-Day Meme Coin is real but unevenly distributed.** 1,801 tokens ceased all trading within 24 hours of launch. BSC accounts for 1,598 of them (10.35% of its tokens), followed by Base (63, 7.62%), Ethereum (129, 3.47%), and Solana (11, 0.09%). Of these ephemeral tokens, 276 recorded only a single transaction, 50% had 7 or more transactions, and 10% recorded more than 68 before going inactive. Note: the abstract states this figure as 5.15% while Section 4.1.2 and Table 4 state 5.66% for the same count of 1,801.
- **Most projects claim a website, few maintain one.** 23,755 of 31,811 tokens (74.80%) listed a website. After removing 2,642 non-dedicated URLs (redirects to social media, crypto services, unrelated pages), 21,113 remained. Only 10,722 (50.78%) were reachable by socket connection, and only 6,766 (32.09%) returned an HTTP 200 — an inaccessibility rate of 67.91%.
- **Domain registration suggests minimal investment.** WHOIS data was retrieved for 11,819 sites (55.98%). The top registrars were Namecheap (3,251), GoDaddy (2,354), and Hostinger (1,287), with Porkbun (186) noted for sub-$1 first-year pricing.
- **Website links often point somewhere else.** 1,761 tokens pointed to social media (Telegram 1,334, X 225), 807 to crypto services (pump.fun 280, poocoin.app 178), 654 to website builders (vercel.app 79, canva.site 73), 136 to other meme coins' domains, and 74 to entirely unrelated pages such as Wikipedia. The domain claudeopusai.com appears 15 times across tokens with different names, and bonefloki.net appears 8 times.
- **Logos are frequently missing.** Logos were retrieved for only 15,095 tokens (43.14%); the authors treat a missing logo as a binary signal of low-effort deployment rather than as missing data.
- **Malicious content is present in the archive.** One HTML snapshot (the BaseAI token page) was flagged by commercial antivirus as containing a potential JavaScript Trojan. Google Safe Browsing flagged 9 sites for social engineering and ChainPatrol blocked 8 as phishing. ChainPatrol caught Bitcoin Dogs but missed Memeinator, a documented scam from the same team.
- **Social presence skews to Telegram.** Telegram is used by 28,291 projects (80.85%), X by 24,029 (68.67%), and Discord by only 2,344 (6.69%). Telegram adoption reaches 91.8% on BSC; Discord is close to absent on Solana (0.7%) and Base (1.2%).
- **Naming trends shifted over time.** TF-IDF on verified meme coin names, refined to 126 meme-specific keywords after the elbow method suggested 150 candidates, shows animal terms dominating (dog 7.24%, cat 6.92%, inu 4.82%) alongside trend terms (AI 4.96%), with a 2024 shift from canine toward AI and feline narratives.

## Methodology in Plain English

The authors built the dataset in stages. First they pulled tokens already tagged as meme coins from CoinMarketCap (2,930), CoinGecko (2,107), and Gecko Terminal (4,735), yielding 8,852 distinct verified tokens after deduplication. To reach emerging tokens they scraped DexScreener (30,771) and CoinSniper (44,611), giving 65,021 unique tokens that needed filtering.

To separate meme coins from utility tokens, they ran TF-IDF over the names of known meme coins to extract distinctive keywords, used the elbow method to pick a 150-keyword cutoff, manually pruned generic terms like SOL and ETH down to 126 keywords, and flagged any DEX-listed token whose name contained one. That found 26,557 tokens. They separately added 4,140 pump.fun tokens, justified by a manual check of 100 randomly sampled pump.fun tokens in which 100% referenced memes or viral trends. Total: 35,220.

Three conservative cleanup passes followed: removing 4 of CoinMarketCap's 82 labeled stablecoins; flagging 525 tokens priced above $0.80 or with market cap above 10^7, querying CoinGecko for their categories, and removing 46; and string-matching away names containing "usd", "wrapped", or "staked" to drop 182 more. Final count: 34,988.

They then validated addresses — format checks and contract-creation lookups via Etherscan, BscScan, and BaseScan for EVM chains, and Solana's RPC API — removing 1,065 invalid or undeployed entries (791 malformed, 254 undeployed, 20 Solana mints) to leave 31,811 valid tokens. Deployment timestamps came from those explorer APIs plus Alchemy for Solana; price and market cap were available for 20,692 tokens (65.05%), with the missing 11,119 explained by the absence of an active liquidity pool at collection time.

For the lifecycle study they pulled every transaction timestamp through mid-January 2025, dropping 14 tokens with zero transactions and analyzing 31,797. For the web study they resolved shorteners, deduplicated to 19,481 domains, resolved WHOIS, tested socket and HTTP reachability, archived raw HTML, and ran two commercial threat-detection services over the results.

## Why This Matters

**Impact on research.** Prior multimodal meme coin datasets (Coin-Meme with 3,751 Solana tokens, CoinVibe with 6,231) are roughly an order of magnitude smaller and single-chain; the largest general scam dataset, TokenScout, covers 214k Ethereum tokens but has no logos, HTML, or social data. MemeChain is the first to combine multi-chain coverage with off-chain web artifacts, and it deliberately includes failed and ephemeral tokens rather than only survivors, which is what a realistic churn model needs.

**Real-world applications:**

- Real-time investor protection tools that warn users before they interact with a contract, using website reachability and social presence as early-warning features.
- Detection of phishing-kit reuse and brand impersonation, by comparing archived HTML and logos across tokens that share templates.
- Automated flagging of incongruence between a landing page's claims and on-chain metrics, a hallmark of fraudulent intent.
- Cross-chain forensics on how manipulation schemes migrate from Ethereum to low-fee chains, and comparative rug-pull prevalence across architectures.

**Industry relevance.** The finding that 80.96% of BSC meme coins went inactive and that one-day tokens are concentrated on BSC (10.35%) is directly useful to exchanges, DEX aggregators, and wallet providers deciding what to list or warn about. The observation that ChainPatrol caught one documented scam but missed another from the same team shows current blacklists have real coverage gaps, and the presence of a JavaScript Trojan in an archived token page shows the risk is not limited to trading losses.

## Future Directions

- **Training survival and early-warning models.** The paper explicitly frames one-day token prediction as a survival analysis problem, arguing that web stability signals should be combined with transaction activity and social presence — but the actual models are left to future work.
- **Multimodal scam classifiers.** Building systems that jointly use logos (e.g., Siamese networks for brand impersonation), HTML structure (phishing-kit detection), and on-chain metrics has been proposed but not implemented here.
- **Longitudinal and migration studies.** Extending beyond the single mid-October 2024 snapshot to track capital and fraud patterns moving between Ethereum, BSC, Base, and Solana over time.
- **Improving detection coverage.** The ChainPatrol miss on Memeinator points to an open question of how well commercial threat-intelligence services actually cover the meme coin long tail, and whether dataset-derived signals close that gap.

## Target Audience

Researchers in blockchain security and financial forensics, practitioners building scam-detection or token-risk systems at exchanges, wallets, and DEX aggregators, and data scientists working on multimodal or survival-analysis problems who need a real-world, high-churn dataset. It is also useful for regulators and analysts trying to characterize the meme coin market, since the paper ties its token definition to the SEC's characterization of meme coins as crypto assets inspired by internet memes, characters, current events, or trends.

## Authors’ abstract

The meme coin ecosystem has grown into one of the most active yet least observable segments of the cryptocurrency market, characterized by extreme churn, minimal project commitment, and widespread fraudulent behavior. While countless meme coins are deployed across multiple blockchains, they rely heavily on off-chain web and social infrastructure to signal legitimacy. These very signals are largely absent from existing datasets, which are often limited to single-chain data or…

[The full abstract on arXiv](https://arxiv.org/abs/2601.22185)

[Read the original paper](https://arxiv.org/abs/2601.22185)

---

Canonical: https://crypto.world/paper/memechain-a-multimodal-cross-chain-dataset-for-meme-coin-forensics-and-2601.22185
