← all systems · see it in The System
Social media
Online services on which users make and share content (text, photos, video) and connect their profiles to each other, carried by the internet through web and mobile apps: the platforms (Facebook, YouTube, WhatsApp, Instagram, TikTok, WeChat, X and many more), the posts and interactions users generate, the social graph of their connections, and the ranking systems that choose what each user sees. Grown from PLATO's notes and chat (1960), bulletin boards and Usenet (1970s-80s), and profile-and-friend-list sites (SixDegrees, 1997) to platforms used by billions; most are paid for by selling advertising slots targeted with users' data. Communication research reads it as a many-to-many (dialogic) medium beside one-to-many mass media; computer science as feeds ranked by machine-learned relevance models; sociology as networked publics whose studied effects on news, politics, mental health and social ties are disputed.
Properties published scienceour proposal
Sourced fields, and our own readings.
| aliases | Social media platforms, sys:social-media-platform, Social media platform |
| born at | PLATO 1960; Community Memory by 1973; CBBS 16 February 1978; Usenet 1980; SixDegrees 1997 (contested which counts) |
| born precision | contested / decades |
| boundary | Inside: the platforms (their services, software, ranking and recommender systems, rules, moderation and business), users' profiles, the content users post, the connections between profiles (followers, friends, groups) and the interaction data logged on them. Outside: the people themselves and their offline lives and ties, the internet and the devices that carry the services, advertisers, news organisations, governments and courts. The line is porous: users make the content but the terms of service set who owns it, and the platforms 'control access to the content'. |
| nested level | A part of the mass media in the communication field's reading (its dialogic, many-to-many form; contested, C-1140); built from user-generated content, profiles, the social graph, ranking systems and platform companies; carried by the internet's services floor and by people's own output (their posts, taps and talk). |
| scale | technological |
Measures (24) published science
| Social media users worldwide | 4.76 billion (about 59% of the global population) people |
| Social media users worldwide (Statista, 2022) | 3.96 billion people |
| Share of online time on social networks | 22 percent (2015) |
| Share of active Facebook users in the giant component | 99.91 percent |
| Average distance between Facebook users | 4.74 (3.74 intermediaries) hops |
| Stable relationships per Twitter user | 100-200 relationships |
| Algorithmic amplification of parties' tweets | over 200% in some cases (3 times the chronological audience); Canada Liberals 43% vs Conservatives 167% percent over chronological reach |
| Size of Twitter's chronological control group | 1 percent of all global users (since 2016) |
| Diffusion increase per moral-emotional word | 20 percent per word |
| True and false news cascades studied (Twitter 2006-2017) | ~126,000 stories, ~3 million people, more than 4.5 million tweets count |
| Additional votes from one Facebook message (2010) | 340,000 (60,000 direct, 280,000 through friends) votes |
| Direct effect of the social message on voting | 0.39 percentage points |
| Per-close-friend effect on voting | 0.224 percentage points per close friend |
| Variance in adolescent well-being explained by digital technology use | at most 0.4 percent |
| Lag from social media use to life satisfaction in adolescence | one year years |
| Lifetime of a story | 24 hours |
| Experience-sampling rate (Kross) | five a day for two weeks text messages |
| Median daily social media use (sleep study) | 61 minutes per day |
| Median age of starting social media (US teens 16-18) | 14 years |
| Users who regularly get news from social media (Pew 2021) | roughly 70 percent of users |
| Weight of an emoji reaction against a like (Facebook 2017) | 5 times |
| DSA maximum fine | 6 percent of global sales |
| Articles in the digital media and democracy review | 496 (24 causal) articles |
| Posts analysed for reward learning | over one million posts from over 4000 individuals |
Open contentions (6) open contention
Where sources or readings disagree. Both stay on the map.
- C-1127 How strong are online echo chambers?
booked: The map records echo chambers where they are observed, and keeps the experiments that find smaller effects alongside.
the other reading: - C-1140 Are social media platforms still 'mass' media?
booked: The map counts platforms as mass media, and keeps the view that they blur mass and person-to-person communication alongside.
the other reading: - C-2403 Do opposing arguments backfire?
booked: The map keeps the mixed results.
the other reading: - C-3384 Does activism on social media change anything offline?
booked: The map keeps the competing readings; none is ruled.
the other reading: overstated (Gladwell); linked to offline participation (2018 survey; Lorenz-Spreen 2023) - C-3385 What was the first social media?
booked: The map keeps the competing readings; none is ruled.
the other reading: PLATO 1960; Usenet 1980; BBS 1973-1978; SixDegrees 1997; Web 2.0 platforms (2000s) - C-3386 Does social media harm young people's wellbeing?
booked: The map keeps the competing readings; none is ruled.
the other reading: harmful to adolescents, warranting warnings and age limits; small average associations, too small for policy; bidirectional, varying with age and sex
Cross-scale patterns seen here (21) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Social media. Each links to every other system that shows it.
- borrowed-then-held (structural)
Reddit's image hosting was first done for it by an outside service (Imgur) and later held inside (an integrated image hoster, then its own video hosting and galleries); both phases named in the source. - same role, different builds (structural)
The job 'decide which of everything posted reaches each reader' is done by a reverse-chronological list (newest first from followed accounts, Twitter's 1% control group) and by a machine-learned relevance model (Twitter's Home timeline from 2016), and outside social media by journalists' gatekeeping - templating (structural)
One open protocol (ActivityPub) adopted unchanged by Mastodon, GNU social, Diaspora and Friendica, so their servers interoperate as one federation (the Fediverse), while the dominant platforms are not interoperable. - Umwelt: each system carves its own slice of one shared medium (structural)
From one shared pool of posts each platform carves its own slice by its feed algorithm and interaction paradigm (Facebook more segregated than Reddit), and each user's feed is a personal slice of it; the platform's own Umwelt is its channels: posts, engagement, graph, news, bots, experiments, advert - who opens the door (structural)
What reaches a reader's feed passes a door the receiving platform installed ahead of time: a relevance model trained on engagement and tuned by experiments (a port), against the open field of a chronological list. The door's settings change reach measurably (parties amplified up to 400% for some ind - coordination-before-mapping (partial)
Collective attention coordinates without any map: which memes go viral is explained by limited attention and network structure alone, with no intrinsic value assumed; digital media served a coordinating function in hate crimes with many perpetrators. - holding / entrainment (partial)
The platform's reward stream sets the clock of its parts: users space their posts by the rate of likes they receive, posting sooner when rewards come faster, as animals in a Skinner box time their responses. - lightning (partial)
Virality: a post from a little-known user reaches vast numbers within hours and changes what a public talks about; fake news fires in the immediate aftermath of an event, before conventional media publish. A stored difference (a network primed by attention) released at once. Not capped by any held r - outside-to-inside (partial)
People's ties exist offline first (close friends who meet face to face) and are then held inside the platform as a friend list under real names; the strongest online effects on real votes ran along those offline ties. The outside phase comes first; the 'gradient' here is the tie structure, our readi - partner-becomes-part (partial)
Instagram, a separate company, became part of Facebook in 2012 and its content terms were then set by the new owner; the gain/gain loop before the fusion and any inheritance are not shown in the sources read. - red-queen (partial)
Bots and platforms: platforms try to discover and block bots 'with limited success'; bots pose as real people to avoid prohibitions; cyborgs mix human and bot. Both sides keep changing, neither wins outright, as the sources describe it; the stalemate is stated, not measured over time. - runaway from an unstable balance (partial)
Network effects: below critical mass incentives to join are low; at the critical mass point value equals price; past it each adopter raises value for all and the platform runs away ('market tipping': one system pulls away from its rivals once it gains an initial edge). - same shape, opposite sign (partial)
One process, emotional contagion through the News Feed, run with its sign reversed: removing friends' positive posts made people post less positively and more negatively; removing negative posts did the opposite. The same feed-to-post contagion, its direction set by which emotion was reduced. - seed (partial)
A profile with a friend list under one's actual name (SixDegrees, 1997) is a small packet whose arrival opened a combinatorial space: every profile can link to every other, a graph of 721 million users and 69 billion links with an average distance of 4.74. Whether this is a Seed or simply a new medi - symmetry breaking (name the cut) (partial)
Twitter's 2016 cut: before it, every followed account's post was placed by time alone (no preferred content); after it, a relevance model chose a direction, and reach split by party (right above left in six of seven countries) and by outlet. The cut is named and dated; that the before was symmetric - The carry (partial)
The wait and the overflow: each user handles messages from a queue of finite capacity; past a point (out-degree saturating between 200 and 300) new connections can only be reciprocated at a much smaller rate, and messages wait or drop; in the meme model posts survive in a user's memory only for a fi - The Payment to Entropy (partial)
Signal form: each hand-off along the graph passes on only part of what arrived. Readers attend to only a portion of what they receive; a message's effect on voting was 0.39 points direct and 0.224 points per close friend once passed on, with no detectable effect through ordinary friends; few posts m - the whole sets the local reference (partial)
What the crowd has already done sets each user's local reference: adolescents liked photos with many likes more than the same kind with few, risky or neutral; the encyclopedia: what is socially acceptable is now heavily based on social media. - trade one property for another (partial)
Participation bought with commitment (Gladwell): weak-tie networks raise participation by lowering the motivation it needs, against high-risk activism built on strong ties; disputed by a 2018 survey finding online political expression goes with offline participation. - causal ceiling (surface)
Neither user nor platform can see what sets reach: users take on friends' emotions 'without their awareness'; even the platform's own auditors say the causal mechanism of amplification is not known. A limit of view; whether structural or practical is open. - namespace completeness (the map predicts its gaps) (surface)
Kaplan's two-by-two of mobile social media (location-sensitive or not, time-sensitive or not) names four kinds, each with an example; the grid's cells were filled, not predicted ahead. Surface only.
Valence: what it trades, and with whom (19) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Social network via sys:social-network · Social media gain
- Journalism via sys:news · Social media gain
- Human (Homo sapiens) via sys:electromagnetic-field · Social media gain
- Internet via sys:internet · Social media gain
- Human society (from local community up to societies) via sys:human-society · Social media gain
- Mass media via sys:news · Social media mixed
- Democracy via sys:human-society · Social media gain
- Election via sys:human-society · Social media gain
- Attention via sys:electromagnetic-field · Social media gain
- Public opinion via sys:human-society · Social media gain
- Mental health via sys:electromagnetic-field · Social media gain
- Advertisers via sys:market · Social media gain
- Computer law via sys:internet · Social media gain
- Social bots via sys:social-bot · Social media loss
- News via sys:internet · Social media gain
- Marketing via sys:market · Social media gain
- Political science (the field: the practice and the politics it studies) via sys:human-society · Social media even
- Adolescence via sys:internet · Social media gain
- Telegraph via sys:human-society · Social media even
Partners and parts published science
Built from / builds (11)
- Mass media
- Recommender system
- Social bots
- User-generated content (posts, comments, photos, videos, interaction data)
- User profile
- Social graph (friends, followers, groups)
- Social media company (the firm that runs a platform)
- Topics, memes and outlets competing for finite collective attention
- Information sources competing for people's attention (the attention economy)
- Outlets competing for stories and audiences (scoops and copying)
- News media racing to deliver the news first
Sits inside / holds (3)
expressed by (12)
- Templating / Standardization
- Amplification/Scale-Up (Manufacture & Distribution)
- Internal Coordination/Regulation
- Sensing/Scanning
- Uncertainty Reduction/Learning
- Active Experimentation / Probing (Optimal Experimental Design)
- Trail-Marking / Externalized Memory (Stigmergy)
- Social Learning / Public Information Use
- Technique-Based Identification (a.k.a. Technique-Based Detection)
- Self-Monitoring
- Validation of Identity/Intention
- Triage & Prioritized Allocation
floor filled by (24)
- Social network
- Mass media
- Journalism
- Emotional contagion
- Human (Homo sapiens)
- Internet
- Algorithm (the practice of step-by-step procedures and the structure they compute)
- Democracy
- Recommender system
- Election
- Attention
- Public opinion
- Mental health
- Advertisers
- smartphone
- Computer law
- Social bots
- User-generated content (posts, comments, photos, videos, interaction data)
- User profile
- Social graph (friends, followers, groups)
- Social media company (the firm that runs a platform)
- Political science (the field: the practice and the politics it studies)
- News
- Adolescence
Kinds (17)
- Mass media
- News media
- Blogs
- Business networks
- Collaborative projects and wikis
- Enterprise social networks
- Forums
- Microblogs
- Photo sharing
- Products and services reviews
- Social bookmarking
- Social gaming including MMORPGs
- Social network sites
- Video sharing
- Virtual worlds
- Fediverse (decentralised social media)
- Mobile social media
wired (1)
Notes published science
- Since 2016 Twitter kept a random 1% of all its users on the old newest-first feed. Comparing them with everyone else showed the ranking algorithm gives some parties' tweets an audience three times bigger, the mainstream right more than the left in six of seven countries, but not the far left or far right more than moderates, 'contrary to prevailing public belief' (src:huszar-2022).
- One Facebook message on election day 2010 added about 340,000 votes in the US, but only about 60,000 came from the message itself: the other 280,000 were passed on by close friends, who make up just 7% of Facebook friendships (src:bond-2012).
- People post on social media the way animals press a lever in a Skinner box: across more than a million posts on several platforms, users posted sooner when likes came faster, spacing posts to maximise the rate of likes (src:lindstrom-2021-reward). And on Twitter, people still keep only 100 to 200 stable relationships, Dunbar's number from primate brains (src:goncalves-2011-dunbar).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
social media as communication research and computer science describe it: people's lives and posts to the ranked feed, the public conversation and its effects on the world (reader's view)
- L1: People and their lives, events and devices: users reach the services through web and mobile apps carried by the internet; 4.76 billion people
- L2: User-generated content and profiles: text posts, comments, photos, videos, and the data generated by interactions; the profile a user builds
- L3: The social graph: friends, followers, groups and lists connecting profiles (721 million users and 69 billion friendship links on Facebook in 2011, 99.91% in one component)
- L4: The feed: ranking and recommender systems select and order what each user sees from content features, social connectivity and user activity (reverse-chronological feeds before; machine-learned relevance models from 2016 on Twitter)
- L5: The platform company: rules, moderation, terms of service and the business model of selling targeted advertising slots; under law (Section 230, the EU's DSA and DMA, age limits)
- L6: Networked publics and collective attention: virality and cascades, trending, homophilic clusters and echo chambers, emotional and moral contagion
- L7: Effects back on the world: news consumed and shared, political self-expression, votes and protest, moods and well-being, purchases; beneficial and detrimental by context
the boardroom (shared floors)
- L4: The feed: ranking and recommender systems choose and order what each user sees from content features, the graph and engagement; a personalised relevance model or a reverse-chronological list
- L5: The platform company: rules and moderation, terms of service, the advertising business, compliance with law, copying rivals
- L6: Networked publics: virality, cascades, trends, homophilic clusters, emotional and moral contagion; the competition of memes for limited attention
- L7: Effects back on the world: news consumption, participation, votes and protest, trust and polarisation, moods and well-being, purchases
way in: Advertisers' money and targeting requests
- L1: Firms promoting products and seeking customers
- L2: Advertising slots bought
- L3: Matched to data about each user so ads are individually relevant
- L4: Placed in feeds, where users can share and comment on the ad
way in: Automated accounts: bots, social bots and cyborgs
- L1: Automated programs operating on the internet
- L2: Programmed to like, comment and follow like people; some pose as real people
- L3: Mixed into the interaction signals so that bot and human interactions are hard to tell apart; they spread true and false news at the same rate
way in: Law and regulation: legislatures, regulators, courts
- L1: Legislatures act (the US Communications Decency Act, 1996; the EU's DSA and DMA, 2022; Australia's under-16 law, 2024)
- L2: The rule: Section 230 exempts platforms from liability for third-party content
- L3: Duties: very large platforms must redesign systems for privacy and minors, end advertising on personal data, and assess recommender risks
- L4: Compliance machinery: content moderation and age verification, at large cost
way in: News: outlets' stories, politicians' statements and links shared by users
- L1: Events and the news about them; fake news often first, before conventional media publish
- L2: Politicians, political organisations and news outlets post to large audiences
- L3: Published, consumed and shared along the graph; about 70% of users regularly get news this way
way in: Rival platforms and their features
- L1: Other platforms adapting functionality beyond their scope
- L2: A rival's feature taken up (Snapchat's stories by Instagram, then YouTube)
way in: The platform reading itself: randomised experiments and control groups
- L1: A randomly chosen 1% of all Twitter users kept on the old chronological feed since 2016
- L2: Feeds changed for a test group (e.g. emotional content reduced in the News Feed)
- L3: Differences between treatment and control read back and used to improve the ranking
way in: What people do with posts: likes, shares, comments, reactions, time spent
- L1: Users' responses to posts; the like as a signal of approval
- L2: Logged as interaction data; sharing tools (the like option, re-share)
- L3: Read as engagement signals that the ranking tracks; reactions can be weighted (emoji reactions five times a like on Facebook in 2017)
way in: What people post: text, photos, video, live streams
- L1: People's lives, memories, friendships and communities, and the events around them
- L2: Captured and uploaded through web and mobile apps (cameras, camera phones, keyboards)
- L3: Attached to the poster's profile and sent along their connections
way in: Who connects to whom: friend requests, follows, groups
- L1: People's ties, most of the strongest also face to face
- L2: Profiles and friends lists under actual names (SixDegrees, 1997)
- L3: The social graph, held by the platform and capped in use by people's own limits (100-200 stable relationships on Twitter)
way out: Likes, comments and counts returned to the poster
- L1: Shown to the poster as likes: read in the brain's reward regions
- L6: Endorsement accrues on a post
way out: Messages that move offline behaviour (get-out-the-vote, calls to protest)
- L1: Votes cast at polling places (about 340,000 more in 2010)
- L3: Passed on through close friends more than through the message itself
- L7: A platform message to millions, with friends' faces
way out: Moderation: deletion, reduced reach, bans
- L1: Content and accounts removed from the shared space
- L4: Problematic content deleted or its reach reduced; bots blocked with limited success
- L5: Rules for what is allowed and what is promoted
way out: Targeted advertisements
- L1: Ads seen, shared and acted on; peer communication raising purchase intentions
- L3: Targeted with data about each user
- L5: Ad slots sold
way out: The ranked feed shown to each user
- L1: Light and sound from the screen to the user's eyes and ears
- L2: Rendered in the app, curated to keep the user scrolling
- L4: The ranked selection for this user (in some cases an audience three times the chronological one)
way out: Users' data passed on: to advertisers, third parties, governments
- L1: Made available to third parties, including for targeted advertising; used by law enforcement
- L5: Data collected beyond the profile, often without knowing consent