DATA DRIVEN DESIGN GROUP        UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN
The rise of social media has changed the nature of the fashion industry. Influence is no longer concentrated in the hands of an elite few: social networks have distributed power across a broader set of tastemakers. To understand this new landscape of influence, we created FITNet — a network of the top 10k influencers of the larger Twitter fashion graph. To construct FITNet, we trained a content-based classifier to identify fashion-relevant Twitter accounts. Leveraging this classifier, we estimated the size of Twitter’s fashion subgraph, snowball sampled more than 300k fashion-related accounts based on the following relationships, and identified the top 10k influencers in the resulting subgraph. We use FITNet to perform a large-scale analysis of fashion influencers, and demonstrate how the network facilitates discovery, surfacing influencers relevant to specific fashion topics that may be of interest to brands, retailers, and media companies.
PDF (29.2MB)
Jinda Han, Qinglin Chen, Xilun Jin, Weikai Xu, Wanxian Yang, Suhansanu Kumar, Li Zhao, Hari Sundaram and Ranjitha Kumar. FITNet: Identifying Fashion Influencers on Twitter. Proceedings of ACM Human-Computer Interaction, 5, CSCW1, Article 153 (April 2021), CSCW 2021.
Bibtex

The Dataset
FITNet is a network of 10,000 fashion-related Twitter accounts that heavily influence the larger Twitter fashion graph. FITNet comprises 4,188 (42%) influencers, 2,066 (20%) brands, 1,356 (14%) retailers, and 3,105 (31%) media. Among 4,188 influencer accounts, there are 2,248 bloggers (54%), 593 editors (14%), 458 designers (11%), 485 stylists (12%), 307 celebrities (7%), 291 models (7%), 137 photographers (3%), and 79 casting directors (2%).

FITNet number of influencers by subcategories
Data Collection
To explore what FITNet accounts to post about and how they interact with each other, we used Twitter’s API to mine all of their tweets between January 1st, 2018, and February 1st, 2019. On average, we captured 2,923 posts per account. We index this repository of tweets by hashtags, mentions, and retweets. This allows us to identify FITNet accounts that post about specific topics, and understand how accounts interact with each other’s content by visualizing mention and retweet networks.

FITNet structure
1. FITNet Dataset: Top 10k FITNet Accounts w/ Simple Metadata (full)
  • FITNet 2021 re-modified JSON full version (10,000): Download (2.5 MB)
  • FITNet 2023 triple-cleaned CSV version (8,930): Download (509 KB)
It contains the simple metadata of 10k FITNet users, such as Twitter ID, Twitter screen name, following ranking, human labels, etc., and the file is named 'fitnet_recleaned_2021.json' which was modified in 2021. All the following list from Twitter API: GET friends/ids.

              file: fitnet_recleaned_2021.json
              
              {
                "_id":"136361303",
                "twitter_user_id":"136361303",
                "twitter_screenname":"voguemagazine",
                "labels":"['magazine']",
                "follow_ranking":"1",
                "followers_count":"13596800",
                "friends_count":"526"
              }
                                
2. FITNet Accounts Metadata Hierarchies
Contains 10,000 fashion-related Twitter accounts that heavily influence the larger Twitter fashion graph. For each account, we present a detailed view hierarchy (JSON file).

A sample view hierarchy file is shown below. The twitter_profile contains the metadata of the user, such as geo-location (location), profile description (description), website details (entities), etc. All the detail metadata from Twitter API: GET users/show.

{
  "_id": "19212009",
  "twitter_user_id": "19212009",
  "twitter_screenname": "WhoWhatWear",
  "twitter_name": "Who What Wear",
  "twitter_profile": {
    "location": "Los Angeles, California",
    "description": "Your download on what matters in fashion.",
    "entities": {
      "url": {
        "urls": [
          {
            "url": "http://t.co/9ggzV7QsB7",
            "expanded_url": "http://whowhatwear.com",
            "display_url": "whowhatwear.com",
            "indices": [
              0,
              22
            ]
          }
        ]
      },
      "description": {
        "urls": []
      }
    },
    "followers_count": 2199173,
    "friends_count": 1645,
    "listed_count": 11143,
    "created_at": "Tue Jan 20 00:57:14 +0000 2009",
    "favourites_count": 7664,
    "geo_enabled": true,
    "verified": true,
    "statuses_count": 136104
  }
}
                
3. FITNet Following Graph
Contains every user's following graph* in a single file as "user_id.json". All the following list from Twitter API: GET friends/ids.

                file: 19212009.json
                
                {
                "user_id": "19212009", "screen_name": "WhoWhatWear", "followers_ids": [3327720838, 
                82455213, 70740986, 842520310226198528, 17446621, 715522322510299136, 339838202, 
                389052298, 181561712, 268414482, 27260086, 14459746, 367285520, 704091486, 17565514, 
                941782988819521536, 487718304, 1888149452, 342327238, 185149826, 956207887784722432, 
                16963828, 495512390, 189362860, 346353711, 2418100549, 709540733456224257, 307426028, 
                24032232, 42655526, 44620271, 14230524, 19989038, 104299103, 1128950437, 119922340, 
                18815234, 276575042, 50725573, 47216804, 193929442, 233906161, 14790966, 124003770, 
                78366205, 18278629, 78525538, 214622133, 708528918, 266697434, 62513246, 37865243, 
                4863187106, 1027746071743627264, 46241729, 26239721, 383031509, 3384776182, 
                2294733114, 111935889, 21199743, 2275481522, 2317790305, 26589987, 56506548, 
                256573865, 18924291, 89898430, 249514855, 15088579, 19637579, 15692086, 27167403, 
                89563994, 74669397, 987662683, 3320478908, 4164998908, 323376858, 634734888, 
                938424931230060544, 451131659, 19409508, 757695091, 1512446430, 963631905878036482, 
                2812768561, 1020383864, 304185486, 167428167, 200889851, 53933574, 834609484781723648, 
                16573941, 137557877, 243165626, 756271337585319936, 17812282, 266336410, 10506152, 
                  .
                  .
                  .]
                }
                                  
4. FITNet Rankings
We then run PageRank over this 300k graph to identify the most important accounts based on the fashion subgraph based on follow relationships, and defined FITNet as the subgraph comprised of the top 10k accounts under this ranking. In addition, we also run PageRank on the mention and retweet relationships that extracted from the tweets between accounts in FITNet.

{
  "_id": "136361303",
  "twitter_user_id": "136361303",
  "twitter_screenname": "voguemagazine",
  "follow_ranking": 1,
  "mention_ranking": 2,
  "retweet_ranking": 1
}
                  
5. FITNet Accounts Labelling and Categories
To validate and further categorize the top-ranked fashion accounts, we recruited 55 undergraduate students majoring in Textile and Apparel Management at a large, public university in the United States. We built a web interface that facilitated the validation and labeling process. Starting from the top of the ranked list, the interface presented accounts one at a time to participants until 10k fashion accounts had been validated and categorized.

The following 22 options were provided to participants for categorizing Twitter accounts:
Influencers (8): celebrity, model, stylist, blogger, photographer, editor, designer, casting director
Brands (4): luxury fashion brand, premium fashion brand, mass-market/fast fashion brand (e.g. Zara, Gap, Uniqlo), beauty (e.g. Estee Lauder, Clinique)
Retailers (3): department store (e.g. Nordstrom, Macy’s), e-commerce site (e.g. Farfetch, 6pm), other retailers (e.g. Target, Kohls)
Media (7): blog, newspaper, magazine, PR/marketing agency, e-zine (digital magazine only and social media website), fashion week or other fashion events, social networking site (e.g. Instagram, Twitter, Pinterest)

{
  "_id": "136361303",
  "twitter_user_id": "136361303",
  "twitter_screenname": "voguemagazine",
  "labels": {
    "magazine": {
      "count": 2
    }
  }
}
                  
6. FITNet Tweets Metadata (mention/retweet and hashtags)
We provide a sample amount of extracted metadata from each FITNet users' tweets. Every user's extracted metadata contains in a singel file as "user_id.json". All the origional tweets from Twitter API: Get Tweet timelines.

There are two different relationships: "retweet" and "mention". For example, if user A mention user B in A's single tweet, then "relation" will be "mention", and the B's name will appear in the "relation_node" list. In addition, all the hashtags will be contained in the list of "hashtags".


file: 14791162.json

[{
  "_id": 1092807609302896600,
  "user_id": "14791162",
  "relation": "mention",
  "relation_node": [
    "the_spacewitch"
  ],
  "hashtags": []
},{
  "_id": 1092804048565256200,
  "user_id": "14791162",
  "relation": "",
  "relation_node": [],
  "hashtags": []
},{
  "_id": 1092800114043879400,
  "user_id": "14791162",
  "relation": "",
  "relation_node": [],
  "hashtags": [
    "#feud",
    "#americancrimestory"
  ]
},
...
]