Authors: (1) SIMONE CASALE-BRUNET, École Polytechnique Fédérale de Lausanne, Switzerland; (2) MIRKO ZICHICHI, Universidad Politécnica de Madrid, Spain; (3) LEE HUTCHINSON, WhaleAnalytica.com, Switzerland; (4) MARCO MATTAVELLI, École Polytechnique Fédérale de Lausanne, Switzerland; (5) STEFANO FERRETTI, University of Urbino “Carlo Bo”, Italy. Table of Links 1 Introduction 2 State of Art 3 Projects Selection and Data Collection 3.1 Blockchain data 3.2 Twitter data 4 Data Analysis and Results 4.1 Ethereum wallets and Twitter users and 4.2 The communities 4.3 Hashtags 4.4 The role of the social network community 5 Conclusions and References 3.1 Blockchain data The data saved on blockchain, which are public and accessible to all by construction of the Ethereum blockchain, are efficiently accessed when structured in formats such as SQL [4]. For this reason, for the analysis presented in this paper we made use of Google BigQuery which structures this data in an SQL database and provides a very efficient access API [16]. This method of analysis has already been presented in [7] and will be referred to for more technical details about the data extraction techniques used. Briefly, for each transaction we stored the hash of the transaction, the address of the NFT smart contract, the addresses of the wallet(s) or smart contract(s) that sold and purchased an NFT token, the Ethereum block time as milliseconds from the UNIX epoch, the Ethereum block number, the value of ETH and/or WETH (Wrapped ETH) transferred and the number of tokens exchanged during the transaction (can be more than one). Gas fees (i.e., transaction costs) were excluded from the analysis; if the exchange currency is neither ETH nor WETH the transaction was seen as a simple transfer (seen that the vast majority of the transactions happen in ETH or WETH, we concluded that this assumption still produces reasonable estimates of the exchange volumes while significantly reducing the complexity of the queries used). Whenever more than one token was exchanged in the same transaction, the individual price of each token was considered the same, and computed by splitting the total amount of the transaction into equal parts. For each NFT collection, data was extracted from the date of their creation to April 15th, 2022. It must be noted that unlike what was done in [7] where only transactions involving Externally-owed-Accounts (EoAs) were considered, for this work also transactions involving Contract Accounts (CAs) have been considered. In order to eliminate suspicious transactions (e.g., attributable to money laundering activities), for each project we eliminated transactions above the 95th percentile and those where the buyer and seller address are exactly the same. Moreover, only for the Meebits collection we decided to neglect all the transactions made on the LooksRare marketplace because we judged the majority of transactions made within this platform as suspicious: as an example, in mid-January 2021 there were sales of some tokens for millions of dollars each, when the average value traded in the days before was in the tens of thousands of dollars [12]. In Table 2 one can find for each project the number of both Ethereum transactions analysed from each project smart contract deployment date, till April 15th, 2022. The value in USD was calculated using the close spot price value relative to the day of the transaction, and the price provided by [9] has been used. The total number of transactions acquired and analysed in this work is 393,210. 3.2 Twitter data The method that we used to collect tweets from Twitter has been going through the o$cial Search Tweets API [22]. For research purposes, this API allows to programmatically access public tweets from the complete archive dating back to the first Tweet in March 2006, using specific search queries. For each tweet we stored the tweet time as milliseconds from the UNIX epoch, the unique tweet identifier generated by Twitter, the tweet text, the number of replies, likes and retweet count at the time of the query, the list of hashtags used on the tweet, and the unique Twitter username of the author of the tweet. For each project, we queried the API by providing as search points the project official twitter account and specific hashtags used by the project social community (see Table 1). For the scope of this work, only English-language tweets have been collected. In Table 2 one can !nd for each project the number of tweets from the !rst tweet found by the query till April 15th, 2022. The total number of tweets acquired and analysed in this work is 7,747,078. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) SIMONE CASALE-BRUNET, École Polytechnique Fédérale de Lausanne, Switzerland; (2) MIRKO ZICHICHI, Universidad Politécnica de Madrid, Spain; (3) LEE HUTCHINSON, WhaleAnalytica.com, Switzerland; (4) MARCO MATTAVELLI, École Polytechnique Fédérale de Lausanne, Switzerland; (5) STEFANO FERRETTI, University of Urbino “Carlo Bo”, Italy. Authors: Authors: (1) SIMONE CASALE-BRUNET, École Polytechnique Fédérale de Lausanne, Switzerland; (2) MIRKO ZICHICHI, Universidad Politécnica de Madrid, Spain; (3) LEE HUTCHINSON, WhaleAnalytica.com, Switzerland; (4) MARCO MATTAVELLI, École Polytechnique Fédérale de Lausanne, Switzerland; (5) STEFANO FERRETTI, University of Urbino “Carlo Bo”, Italy. Table of Links 1 Introduction 1 Introduction 2 State of Art 2 State of Art 3 Projects Selection and Data Collection 3 Projects Selection and Data Collection 3.1 Blockchain data 3.1 Blockchain data 3.2 Twitter data 3.2 Twitter data 4 Data Analysis and Results 4.1 Ethereum wallets and Twitter users and 4.2 The communities 4.1 Ethereum wallets and Twitter users and 4.2 The communities 4.3 Hashtags 4.3 Hashtags 4.4 The role of the social network community 4.4 The role of the social network community 5 Conclusions and References 5 Conclusions and References 3.1 Blockchain data The data saved on blockchain, which are public and accessible to all by construction of the Ethereum blockchain, are efficiently accessed when structured in formats such as SQL [4]. For this reason, for the analysis presented in this paper we made use of Google BigQuery which structures this data in an SQL database and provides a very efficient access API [16]. This method of analysis has already been presented in [7] and will be referred to for more technical details about the data extraction techniques used. Briefly, for each transaction we stored the hash of the transaction, the address of the NFT smart contract, the addresses of the wallet(s) or smart contract(s) that sold and purchased an NFT token, the Ethereum block time as milliseconds from the UNIX epoch, the Ethereum block number, the value of ETH and/or WETH (Wrapped ETH) transferred and the number of tokens exchanged during the transaction (can be more than one). Gas fees (i.e., transaction costs) were excluded from the analysis; if the exchange currency is neither ETH nor WETH the transaction was seen as a simple transfer (seen that the vast majority of the transactions happen in ETH or WETH, we concluded that this assumption still produces reasonable estimates of the exchange volumes while significantly reducing the complexity of the queries used). Whenever more than one token was exchanged in the same transaction, the individual price of each token was considered the same, and computed by splitting the total amount of the transaction into equal parts. For each NFT collection, data was extracted from the date of their creation to April 15th, 2022. It must be noted that unlike what was done in [7] where only transactions involving Externally-owed-Accounts (EoAs) were considered, for this work also transactions involving Contract Accounts (CAs) have been considered. In order to eliminate suspicious transactions (e.g., attributable to money laundering activities), for each project we eliminated transactions above the 95th percentile and those where the buyer and seller address are exactly the same. Moreover, only for the Meebits collection we decided to neglect all the transactions made on the LooksRare marketplace because we judged the majority of transactions made within this platform as suspicious: as an example, in mid-January 2021 there were sales of some tokens for millions of dollars each, when the average value traded in the days before was in the tens of thousands of dollars [12]. In Table 2 one can find for each project the number of both Ethereum transactions analysed from each project smart contract deployment date, till April 15th, 2022. The value in USD was calculated using the close spot price value relative to the day of the transaction, and the price provided by [9] has been used. The total number of transactions acquired and analysed in this work is 393,210. 3.2 Twitter data The method that we used to collect tweets from Twitter has been going through the o$cial Search Tweets API [22]. For research purposes, this API allows to programmatically access public tweets from the complete archive dating back to the first Tweet in March 2006, using specific search queries. For each tweet we stored the tweet time as milliseconds from the UNIX epoch, the unique tweet identifier generated by Twitter, the tweet text, the number of replies, likes and retweet count at the time of the query, the list of hashtags used on the tweet, and the unique Twitter username of the author of the tweet. For each project, we queried the API by providing as search points the project official twitter account and specific hashtags used by the project social community (see Table 1). For the scope of this work, only English-language tweets have been collected. In Table 2 one can !nd for each project the number of tweets from the !rst tweet found by the query till April 15th, 2022. The total number of tweets acquired and analysed in this work is 7,747,078. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv
