Fact-checked by Grok 2 weeks ago

EdgeRank

EdgeRank is the name commonly applied to the algorithm that Facebook employed to rank and prioritize content visibility in users' News Feeds prior to 2011.^[1]^[2] The system calculated a score for each potential "edge"—representing interactions such as posts, likes, or comments—using a formula that multiplied three primary factors: user affinity (u_e), which gauged the strength of the relationship between the viewer and the content creator; edge weight (w_e), which assigned higher values to more engaging interaction types like comments over simple likes; and time decay (d_e), which diminished the relevance of older content.^[3]^[2] Introduced as part of News Feed enhancements in the late 2000s and detailed publicly at Facebook's F8 conference in 2010, EdgeRank aimed to deliver personalized, relevant updates by filtering the vast volume of potential stories to those deemed most pertinent.^[1] Although simplified for explanatory purposes and never officially termed "EdgeRank" internally by Facebook, it represented an early deterministic approach to feed curation before the platform transitioned to more complex machine learning models incorporating thousands of variables.^[4]^[5] This evolution reflected ongoing efforts to balance user engagement with algorithmic opacity, amid broader debates on content prioritization's impact on information flow.^[4]

History

Origins and Early Development

Prior to September 2006, Facebook lacked a centralized feed, requiring users to manually navigate to individual profiles to view updates, which became inefficient as the user base expanded beyond college networks to millions, leading to overlooked interactions and stagnant engagement metrics.^[6] The September 5, 2006, rollout of News Feed presented updates in reverse-chronological order, aggregating content from connections into a single stream that boosted daily active users by enabling passive consumption, yet empirical data from early usage patterns revealed overload from irrelevant posts, with users spending more time but reporting dissatisfaction in surveys and feedback loops.^[6]^[7] Engineers led by figures like Chris Cox, one of the key architects of the News Feed launch, initiated internal development of a prioritization mechanism around this period, leveraging the platform's underlying social graph—where "edges" modeled discrete interactions such as likes, comments, or shares between users and objects—to filter content based on relevance signals rather than timestamp alone, driven by first-principles analysis of how interaction density correlated with session depth and return visits.^[6]^[1] Prototype testing employed A/B experiments comparing chronological versus ranked variants, quantifying improvements in click-through rates and time spent on high-affinity content to validate the shift, establishing foundational causal insights that certain edge types (e.g., reciprocal engagements) predicted sustained user value over volume-driven displays.^[8]^[4]

Public Introduction and Initial Rollout

Facebook publicly introduced EdgeRank at its f8 developer conference on April 21, 2010, where engineers including Ari Steinberg provided the first detailed public overview of the News Feed's underlying ranking mechanism.^[1] Presented as a simplified model, EdgeRank framed content prioritization as a function of interactions represented by "edges" between users, content, and actions, emphasizing relevance over pure chronological display.^[9] This disclosure aimed to demystify how the algorithm filtered vast amounts of potential updates into a personalized feed, helping developers and users understand factors influencing visibility without revealing proprietary implementation details.^[10] The rollout of EdgeRank's principles marked a refinement in Facebook's approach to News Feed, building on earlier pilots that tested relevance-based sorting to enhance user retention. Internal experiments around 2009 had demonstrated that algorithmic prioritization could filter chronological noise, focusing on high-affinity interactions to sustain engagement amid rapid platform growth.^[11] By 2010, with Facebook's user base exceeding 400 million monthly active users, the public framing of EdgeRank supported this shift, correlating with observed spikes in session times as users encountered more pertinent content, though exact causal metrics remained undisclosed to safeguard algorithmic edges against competitors.^[1] Facebook intentionally kept the full EdgeRank model opaque post-introduction, disclosing only high-level components like affinity, weight, and decay to guide third-party optimization without exposing core computations. This strategy preserved competitive advantages during a period of intense rivalry from platforms like Twitter, which relied on real-time chronological feeds. Early post-rollout data indicated improved user satisfaction with feed relevance, contributing to sustained growth, but without granular transparency, external analyses relied on inferred behaviors from observed engagement patterns.^[12]

Key Milestones in Refinement

In 2011, Facebook refined EdgeRank by increasing the relative weights assigned to visual content types, such as photos and videos, after internal engagement data revealed these formats consistently outperformed text-based status updates and links in generating user interactions. Analyses from that period indicated photos often received the highest weighting, followed by videos, due to their ability to drive comments, likes, and shares more effectively than other post types.^[13]^[14] These adjustments aimed to optimize feed relevance by amplifying edges associated with high-interaction content, thereby improving overall user retention metrics. A parallel 2011 update integrated real-time elements into the News Feed, modifying the time decay factor to diminish scores for older posts more gradually for timely updates while maintaining rapid devaluation for stale content. This change, announced in September, responded to user feedback favoring dynamic, newspaper-like feeds over static chronological lists, enhancing the algorithm's emphasis on recency for affinity-weighted connections.^[15] By 2012, further iterations addressed evolving user behaviors amid rising mobile adoption, accelerating time decay for non-real-time edges to prioritize immediate interactions and reduce feed clutter from delayed posts. These data-driven tweaks, which favored content eliciting strong responses from close connections, correlated with a 50% rise in News Feed likes and comments year-over-year, validating the refinements' focus on high-affinity scoring to boost engagement without increasing ad reliance.^[16]^[17]

Technical Mechanics

Core Formula and Mathematical Basis

The core formula of EdgeRank calculates a total score for a news feed story object by summing multiplicative contributions from all associated interaction edges between the viewing user and the content's origin. Formally, the score f for an object is f = \sum_{e} u_e w_e d_e, where the summation is over edges e connecting the user to the object's creator or interactions, u_e denotes the affinity score for edge e, w_e the weight assigned to the edge's interaction type, and d_e the time-based decay factor diminishing the edge's relevance with age. This equation, disclosed by Facebook engineers at the f8 developer conference on April 21, 2010, prioritizes computationally efficient aggregation to rank billions of potential daily stories in real-time, avoiding the iterative convergence required in eigenvector methods like PageRank.^[1] The model's mathematical basis rests on a heuristic approximation of social graph dynamics, treating edges as directed signals of potential engagement rather than undirected links in a static web. Affinity u_e quantifies relational closeness via historical interactions (e.g., prior likes or messages), modeled as a user-specific scalar derived from interaction frequency and recency; weight w_e encodes action potency (e.g., shares > comments > likes, with numerical assignments undisclosed but empirically tuned for engagement prediction); and decay d_e applies an exponential or linear diminution, such as d_e = 1 / (1 + \Delta t) where \Delta t is time elapsed since the edge's creation, ensuring recency bias without full historical recomputation. This structure enables causal inference for ranking by weighting edges that historically correlate with downstream user actions like clicks or hides, computed via batch processes on sharded graph data for sub-second latency at scale.^[1]^[10] Unlike probabilistic models in contemporary recommendation systems, EdgeRank's simplicity facilitates deterministic thresholding: stories exceeding a user-tuned visibility cutoff appear in the feed, ordered by descending f, with the summation's sparsity (limited to recent, high-affinity edges) bounding complexity to O(k) per user where k is the sparse edge count.^[1] This first-principles approach, rooted in graph traversal heuristics rather than gradient descent, was engineered for the 2010-era hardware constraints of processing petabyte-scale social data without machine learning overhead.^[18]

Primary Factors: Affinity, Weight, and Decay

Affinity, denoted as u_e, represents a personalized score quantifying the viewer's historical engagement with the content creator, participants, or related entities, derived from interaction frequency and recency logs such as likes, comments, shares, tags, and profile visits.^[19]^[20] For example, a user who regularly comments on posts from a specific friend or page accumulates a higher affinity score for that connection, elevating the deterministic ranking of subsequent content from them over less interactive sources.^[21] This factor operates iteratively through the social graph, incorporating mutual friends' actions to assess proximity, though it remains a unidirectional measure from viewer to edge without reciprocal adjustment.^[2] Empirical logs from user behavior data underscored that repeated deep interactions, like comments over passive views, causally predicted sustained relevance, avoiding over-reliance on shallow signals.^[22] Weight, w_e, applies type-specific multipliers to edges based on engagement differentials observed in aggregated user data, with heavier values assigned to actions and content formats yielding higher interaction rates. Comments and shares receive greater weighting than likes due to their indication of sustained cognitive involvement, while photos and videos outperform status updates or external links, as internal data revealed photos generating up to 2-3 times more engagements per view.^[20]^[23] A/B testing validated these tunings; for instance, prioritizing visual media correlated with 53% higher click-throughs compared to text-only posts in controlled rollouts around 2010-2012.^[24] Negative feedback, such as hides or reports, inversely adjusted weights to suppress low-value edges, ensuring the algorithm favored causally engaging formats without probabilistic predictions.^[25] Decay, d_e, enforces a time-based exponential attenuation on edge scores, with the rate calibrated to empirical evidence linking recency to relevance—posts under one hour old exhibit near-zero decay, while those exceeding 24 hours drop sharply, as session data showed 90% of interactions occurring within 12 hours of creation.^[10]^[2] This factor also varies with user login cadence; frequent logins mitigate decay for recent edges, reflecting active users' preference for timeliness, whereas infrequent access accelerates deprioritization to prevent stale content overload.^[26] Analysis of temporal engagement patterns confirmed the causal mechanism: delayed exposure halved response rates, justifying the deterministic halving or steeper curves applied beyond short windows to maintain feed velocity without machine learning variability.^[27]

Implementation in News Feed Ranking

In the EdgeRank era, News Feed generation began with graph traversals over the social graph to identify candidate stories, pulling recent actions from a user's friends and followed pages, often yielding thousands of potential items per session. Real-time scoring applied the EdgeRank formula to these candidates, filtering to the top approximately 500 highest-ranked stories for display, ensuring computational efficiency on commodity hardware while personalizing content delivery. This pull-based model operated dynamically upon page loads, minimizing latency for hundreds of millions of daily active users.^[1] Advertisements were integrated via sponsored edges subjected to analogous weighting and scoring, inserted into the ranked feed; however, organic content from connections initially received priority in surfacing decisions to emphasize relational interactions over promotional material. Engineering constraints of the period, including limited real-time processing capabilities, relied on pre-computed affinity scores and in-memory caches to avoid full graph recomputations per request.^[1] Scalability demands were met through distributed batch processing of interaction logs—handling petabytes of data daily—using Hadoop clusters for deriving aggregate weights and decay parameters, complementing online systems like TAO for low-latency edge queries. By 2011, migrations of dozens of petabytes across data centers underscored the infrastructure strains, yet enabled EdgeRank's viability amid exponential user growth from 500 million to over 1 billion monthly actives between 2010 and 2012.^[28]^[29]

Evolution and Replacement

Transition to Machine Learning Models

Facebook began transitioning away from the static EdgeRank formula internally around 2010, recognizing its limitations in handling the platform's expanding scale and content variety. By mid-2011, the company ceased using the EdgeRank term for its News Feed algorithm, shifting toward machine learning models that incorporated thousands of dynamic factors derived from user behavior and interactions.^[4]^[30] This evolution was driven by EdgeRank's rigid structure, which relied on predefined weights for affinity, edge type, and decay, proving inadequate for prioritizing content amid growing user networks, emerging features like pages and groups, and diversifying post formats such as images, links, and early multimedia shares.^[30] The primary motivation for the change stemmed from EdgeRank's inability to adapt dynamically to real-time user preferences and the influx of varied content types, which static rules could not effectively rank for relevance across a user base exceeding 1 billion monthly active users by 2012.^[4]^[30] Machine learning enabled more precise engagement predictions by learning from vast datasets of interactions, surpassing the fixed formula's capacity to model non-linear relationships in user interests.^[4] In the hybrid implementation phase, core EdgeRank concepts like affinity scores, content weights, and time decay were retained as input features but augmented with neural networks and additional signals, such as device type and post sub-categories, resulting in models with approximately 100,000 individual weights by 2013.^[4] This approach allowed for personalized ranking without fully discarding proven edge-based mechanics, facilitating a smoother rollout while enhancing overall feed relevance through iterative, data-driven refinements.^[4]

Post-EdgeRank Developments (2013–Present)

In 2013, Facebook discontinued the use of EdgeRank as its primary News Feed ranking system, transitioning to machine learning models capable of incorporating nearly 100,000 weighting factors to better predict user engagement on a personalized basis.^[4] This shift rendered EdgeRank obsolete due to its static formula's inability to scale with the platform's growing data volume and the need for dynamic, user-specific predictions beyond simple affinity, weight, and decay metrics.^[31] By the mid-2010s, these ML systems had evolved to process hundreds of signals, including post type, content freshness, and interaction history, prioritizing relevance over chronological order.^[32] The 2018 Cambridge Analytica scandal, involving the unauthorized harvesting of data from up to 87 million users, intensified public and regulatory scrutiny on Facebook's data practices and algorithmic decision-making.^[33] While it prompted Meta (formerly Facebook) to implement new privacy controls and face a $5 billion FTC fine in 2019 for misleading users on data handling, the core News Feed algorithm retained significant opacity, with internal models comprising billions of parameters trained on vast datasets that were not publicly disclosed to avoid competitive disadvantages.^[34] Empirical analyses post-scandal showed no fundamental increase in algorithmic transparency, as Meta continued to rely on proprietary ML pipelines for ranking, emphasizing aggregated performance metrics like predicted click-through rates over granular explanations.^[35] From 2023 onward, Meta introduced updates to the News Feed algorithm emphasizing AI-driven prioritization of original content from creators, while demoting aggregated reposts and low-value shares to enhance feed quality.^[31] These changes, announced as part of broader efforts to counter short-form video competition, included tests showing a roughly 15% reduction in rage-bait content—posts designed for artificial outrage to boost engagement—through refined ML penalties on manipulative tactics.^[31] In 2025, further refinements incorporated user controls for video feeds and AI-suggested searches, but the system remained centered on predictive modeling of engagement signals such as dwell time (duration of content viewing) and negative feedback like hides.^[36] As of October 2025, the News Feed operates as a fully machine learning and AI ecosystem, utilizing multi-stage ranking pipelines that score content based on over 100,000 signals, including real-time user reactions and session-level behaviors, though foundational concepts like interaction weighting persist in baseline scoring layers.^[37] This evolution reflects causal adaptations to empirical data on user retention, with ML models iteratively trained to maximize long-term value over short-term clicks, amid ongoing challenges in balancing personalization with content diversity.^[32]

Distinctions from Contemporary Algorithms

EdgeRank's architecture relied on a transparent, rule-based computation centered on three core factors—affinity between user and content creator, edge weight reflecting interaction type, and time decay for recency—allowing for deterministic predictions of content visibility based primarily on explicit user actions such as likes, comments, and shares.^[4]^[5] This simplicity facilitated direct causal inference, where adjustments to weights or affinities could be tested and verified through observable changes in feed prioritization without requiring extensive data modeling. In comparison, modern Meta News Feed algorithms, evolved since 2013, employ machine learning frameworks that integrate thousands of weighted signals, including device type, session context, and network effects, to generate probabilistic rankings tailored to individual behaviors.^[4]^[38]^[39] A key distinction lies in the shift from explicit to implicit signal dominance: EdgeRank prioritized verifiable, user-initiated engagements that signaled clear intent, enabling content creators to optimize reliably around measurable interactions.^[5] Contemporary systems, however, incorporate subtle behavioral proxies like post dwell time, scroll speed, and passive views as implicit indicators of interest, allowing for finer personalization but introducing variability from unobservable model interpretations of these cues.^[39] This evolution enhances adaptability to diverse user patterns—evidenced by reported engagement lifts from ML-driven predictions—but heightens risks of overfitting to historical data patterns, where models may amplify noise in training sets over generalizable relevance.^[40]^[41] The rule-based nature of EdgeRank supported iterative, first-principles refinements with minimal computational overhead, as tweaks to its limited parameters yielded predictable outcomes across users. Modern black-box models, by contrast, demand vast datasets for training and retraining, prioritizing scalability and real-time adaptation over interpretability, which obscures direct linkages between specific inputs and ranking decisions.^[4]^[38] While this complexity has driven higher retention through hyper-personalization, it trades off the verifiable causality of simpler systems, complicating empirical validation of algorithmic efficacy beyond aggregate metrics like click-through rates.^[39]^[40]

Impact

Effects on User Behavior and Engagement

EdgeRank's affinity factor, which scored edges based on historical user interactions such as likes, comments, and tags with specific individuals or pages, incentivized repeated engagement with familiar connections, thereby reinforcing and densifying subgraphs within users' social networks by elevating content from high-affinity sources.^[2] This mechanism created feedback loops where prior behaviors determined future content exposure, amplifying interactions with akin content and fostering habitual checking patterns among users.^[42]^[43] By assigning higher weights to substantive actions like comments relative to passive likes, the algorithm promoted deeper rather than superficial engagement, correlating with extended session durations as users encountered prioritized, interaction-eliciting posts.^[44] Such prioritization of recency via decay factors further encouraged frequent logins to view timely, relevant updates, contributing to sustained daily activity during EdgeRank's prominence from approximately 2006 to 2013.^[5] Personalization through affinity and weight components raised early concerns about filter bubbles, as articulated by Eli Pariser in his 2011 analysis, which posited that Facebook's feed algorithm shielded users from ideologically divergent views by favoring resonant content.^[45] Initial 2012 examinations, however, indicated that cross-cutting exposures via friends' shares mitigated extreme isolation, though the system's bias toward reinforcing interactions laid groundwork for polarized consumption patterns observable in subsequent data.^[46]

Business and Marketing Consequences

EdgeRank's algorithmic prioritization of high-engagement content led to a marked decline in organic reach for business pages, falling from around 16% of fans in February 2012 to 10.15% by November 2013 and further to 6.51% by March 2014, as documented in analyses of median reach per fan.^[47] This reduction, driven by tweaks emphasizing user affinity and content weight over chronological distribution, compelled brands to allocate budgets toward paid boosts and promoted posts to sustain visibility, marking a pivot from predominantly free exposure to a hybrid model reliant on advertising spend.^[48] The framework's affinity scoring mechanism, which gauged user-page interactions to rank edges, paralleled advancements in ad targeting by enabling more precise audience segmentation based on behavioral signals, thereby improving return on ad spend for marketers.^[2] This underpinned explosive revenue growth, with Facebook's advertising income rising from $1.87 billion in 2010 to $6.99 billion in 2013, reflecting heightened advertiser adoption of platform tools amid diminishing unpaid reach.^[49] Algorithmic changes, validated through internal A/B testing and external studies, demonstrated that reach reductions yielded net gains in per-post engagement—such as a reported doubling in interaction rates despite a 38% visibility drop—positioning the shifts as engagement-efficient rather than solely revenue-extractive, as they aligned feed quality with advertiser goals for qualified impressions.^[50] For businesses, this fostered strategic adaptations like content optimization for weight factors (e.g., favoring photos and questions over links) to maximize organic remnants while scaling paid campaigns, ultimately embedding EdgeRank's logic into broader performance marketing ecosystems.^[51]

Broader Societal Influences

EdgeRank's emphasis on engagement weight and temporal decay amplified network effects within Facebook, enabling the rapid propagation of high-interaction content across user connections. During its primary implementation from approximately 2006 to 2013, this mechanism accelerated viral phenomena, including the widespread dissemination of internet memes and trends that characterized early 2010s social media culture, as content generating quick likes, shares, and comments received prioritized visibility, compounding reach through iterative interactions.^[52]^[20] The affinity factor, derived from users' past interactions with specific connections, systematically favored content from close social ties, which often aligned with preexisting homophily patterns in networks. Analyses of news feed dynamics during this era revealed that such prioritization reinforced ideological segregation, as users encountered predominantly in-group perspectives; for instance, a study of over 10 million U.S. users found that while the algorithm marginally increased cross-ideological exposure compared to chronological feeds, user-driven homophily in affinities limited diverse content to about 15-20% less than potential network diversity, contributing to echo chamber formation.^[53]^[54] By weighting stories based on measurable engagement rather than editorial quality or factual accuracy, EdgeRank created incentives for content that maximized reactions, allowing both substantive and sensational material to spread based on empirical user responses. This approach facilitated access to underrepresented viewpoints challenging institutional narratives but also enabled misinformation to thrive when it provoked outsized interactions, as evidenced by subsequent examinations of algorithmic legacies showing engagement optimization correlates with higher virality of low-quality or polarizing items over verified information.^[55]^[56]

Criticisms and Controversies

Reduced Organic Reach and Pay-to-Play Dynamics

In September 2012, Facebook implemented an algorithm update that deprioritized "promotional" content from brand pages in users' news feeds, classifying posts with external links or sales-oriented language as lower priority compared to those fostering direct engagement.^[17] This adjustment resulted in a sharp decline in organic reach for brand posts, with averages falling from around 16% of followers in early 2012 to approximately 6% by early 2014, according to analyses by social media monitoring firms.^[57]^[58] The changes drew public criticism from entrepreneur Mark Cuban in November 2012, who described Facebook's promoted post feature—requiring payment to boost visibility to fans—as a "shakedown," citing an example where reaching 1 million Dallas Mavericks followers would cost $3,000.^[59]^[60] Facebook responded by emphasizing that promoted posts were optional tools to amplify high-engagement content, arguing that organic reach limitations encouraged quality over quantity and that paid boosts correlated with higher user interaction rates.^[61] These algorithm tweaks aligned with Facebook's business model evolution, shifting reliance from free organic distribution to paid advertising; ad revenues grew from $4.28 billion in 2012 to $6.99 billion in 2013 and $11.49 billion in 2014, more than doubling overall and reflecting increased adoption of sponsored content by brands seeking visibility.^[62] This transition incentivized marketers to invest in promoted posts and targeted ads, as organic alternatives became insufficient for broad audience reach.^[63]

Amplification of Engagement-Driven Content

The EdgeRank algorithm's weighting factor (w_e) assigned higher values to interactions such as shares and comments compared to likes, incentivizing content that elicited strong, immediate responses rather than passive approval or thoughtful analysis.^[5] Shares, in particular, received the highest weight due to their indication of active dissemination, while comments—often sparked by controversy—outweighed simple likes by a factor that prioritized discussion volume over depth.^[20] This structure systematically amplified "rage-bait" posts designed to provoke outrage, as empirical analyses of millions of social media interactions demonstrated that divisive content slamming political opponents garnered significantly higher engagement rates than neutral or positive posts.^[64] The time decay factor (d_e), which exponentially diminished the relevance of older content, further exacerbated this by rewarding rapid engagement bursts, fostering cycles of urgency where users rushed to react before posts faded from feeds.^[21] This mechanic causally linked to the proliferation of outrage-driven discourse, as content requiring prolonged consideration lost priority to sensational items generating quick shares and comments; studies on platform virality confirmed that out-group animosity and emotionally charged negativity drove up to several-fold increases in shares and overall propagation compared to substantive topics.^[65] While this approach boosted short-term user retention by aligning feeds with habitual scrolling patterns, internal Facebook research later revealed engagement plateaus and normalization of low-quality, polarizing material, undermining claims of neutral prioritization in favor of unfiltered metric maximization.^[66]^[67]

Allegations of Bias and Transparency Issues

Allegations of conservative suppression emerged prominently after the 2016 U.S. presidential election, with critics claiming Facebook's news feed algorithm—initially formalized as EdgeRank and later evolved—systematically downranked right-leaning content. A May 2016 Gizmodo report alleged that Facebook contractors suppressed conservative stories in the trending topics section, prompting conservative outlets and figures to accuse the platform of ideological bias favoring liberal narratives.^[68] These claims intensified amid broader scrutiny of social media's role in elections, though empirical analyses of the period indicate that engagement metrics, rather than explicit political targeting, primarily drove visibility disparities.^[69] Subsequent audits and studies largely refuted intentional political skew in favor of engagement primacy. A 2019 independent review commissioned by Facebook, led by former U.S. Senator Jon Kyl, examined anti-conservative bias allegations and found no systemic evidence of algorithmic discrimination against right-leaning pages, attributing differences to varying content strategies that boosted conservative interactions.^[70] Independent research, including a 2021 study using neutral bots on Twitter (with parallels drawn to Facebook's feed mechanics), detected no consistent political bias, emphasizing user interactions like likes and shares as the dominant ranking factors over ideological filters.^[71] Data from a 2023 analysis revealed right-leaning pages generating 47% of interactions from political content despite comprising only 26% of posts, underscoring how provocative tactics—common in conservative media—amplified reach independent of platform intent.^[72] Transparency deficits in EdgeRank's formulation exacerbated distrust, as Facebook withheld full details on weighting factors like affinity (u_e), content type (w_e), and decay (d_e), citing competitive risks and potential gaming.^[73] This opacity persisted into machine learning successors, where partial disclosures via A/B testing reports and developer tools offered glimpses but fueled skepticism, particularly from right-leaning critics arguing it enabled unchecked normalization of left-leaning narratives.^[73] While EdgeRank's simpler rule-based era (pre-2013) exhibited less politicization than later models, valid concerns over over-censorship arose in audits highlighting human moderation's influence on edge weights, though quantitative evidence prioritizes causal engagement loops over deliberate ideological engineering.^[69]^[70]

References

[1]
EdgeRank: The Secret Sauce That Makes Facebook's News Feed Tick
Apr 22, 2010 · Each Edge has three components important to Facebook's algorithm: First, there's an affinity score between the viewing user and the item's ...Missing: uwd | Show results with:uwd
[2]
Facebook Edge, EdgeRank algorithm - Affinity, Weight, Decay
Where ue is user affinity, we is content weight and de is the decay. This algorithm, which was called EdgeRank, is no longer in use since 2011. According to ...
[3]
EdgeRank- A Facebook's newsfeed algorithm | Myprogworld
Jan 10, 2015 · EdgeRank- A Facebook's newsfeed algorithm ; u_e is user affinity ; w_e is how the content is weighted ; d_e is a time-based decay parameter.Missing: formula | Show results with:formula
[4]
EdgeRank Is Dead: Facebook's News Feed Algorithm Now Has ...
Aug 16, 2013 · EdgeRank, Facebook's original News Feed ranking system, is dead. Facebook hasn't used the word internally for about two-and-a-half years.
[5]
EdgeRank Algorithm - Algo behind Facebook News Feed
Jul 23, 2025 · The EdgeRank algorithm not only selects the posts to be displayed on your newsfeed but it also sorts them in order to be displayed on your newsfeed.
[6]
Everyone Hated News Feed. Then It Became Facebook's Most ...
Sep 6, 2016 · Two of the three engineers who helped launch News Feed are still working at Facebook. Chris Cox is now the chief product officer, where he ...
[7]
Facebook News Feed Changed Everything
Sep 12, 2013 · In The Facebook Effect, David Kirkpatrick's history of Facebook's early years, Chris Cox, who's now the company's vice president of product ...Missing: origins | Show results with:origins
[8]
Facebook Keeps Getting More Addictive. Here's How.
Apr 24, 2014 · The news feed has gone from being a simple chronological list to a machine learning product, with posts ranked in your timeline according to the likelihood ...Missing: early | Show results with:early
[9]
How to Optimize Your Facebook News Feed Presence | Inc.com
During the f8 conference in April, Ari Steinberg, a Facebook engineer, introduced the public to EdgeRank, an algorithm Facebook uses to ...
[10]
Facebook EdgeRank Explained: How it Works, and How to Beat It
Facebook revealed the factors that are at work in Facebook's EdgeRank algorithm during the 2010 F8 conference. They are Affinity, Edge Weight, and Time Decay.Missing: explanation | Show results with:explanation
[11]
The Evolution of the Facebook Algorithm: What it Means for Your ...
Aug 28, 2015 · August 2013 - What once was a simple algorithm known as "EdgeRank" has become a lot more sophisticated and complex. In fact, Facebook no longer ...Missing: rollout engagement boost
[12]
What Is EdgeRank? Facebook's Feed Prioritisation Algorithm - LCN
How does EdgeRank work? Back in 2010 at the 2010 F8 conference, Facebook revealed the three main factors of the algorithm: Affinity Score, Edge Weight and Time ...
[13]
Is Measuring Your Facebook Edgerank Useful? - Beth Kanter
Jun 10, 2011 · Weight = typically, photos receive the highest weight, followed by videos, links, status updates and apps. Manual posts receive more weight ...
[14]
How To Reach More People on Facebook - Codeless
Sep 27, 2011 · Weight determines what type of update it is. More weight is given to multimedia over simple text status updates. In order of importance: Videos ...
[15]
Facebook redesigns its News Feed, adds real-time updates
Sep 20, 2011 · 'Now, News Feed will act more like your own personal newspaper,' Mark Tonkelowitz, an engineering manager at Facebook, said in a blog post.
[16]
Surprisingly, Facebook's News Feed Ads Aren't Scaring Away Users
Jan 30, 2013 · Its news feed quality improvements around showing more relevant posts helped increase Likes and comments roughly 50% during 2012. That was only ...
[17]
Dealing With Facebook's Unfriendly New Algorithm - Forbes
Nov 30, 2012 · It's simple, in September Facebook made a change to their EdgeRank algorithm that limits the reach of posts from businesses on Facebook to their ...
[18]
Deconstructing Facebook's EdgeRank - Teach To Fish Digital
May 12, 2010 · One important discussion centered around EdgeRank, Facebook's formula for determining what items show up in the Facebook news feed. To explain ...
[19]
Demystifying Facebook's News Feed Algorithm - Revinate
Oct 21, 2022 · EdgeRank was an algorithm developed by Facebook to govern what is displayed – and how high it's displayed – on the News Feed. Facebook defined ...
[20]
Facebook News Feed Algorithm. Overview | by Dilip Kumar - Medium
Aug 10, 2024 · Edgerank is old algorithm used by Facebook to rank the feed for user. What is Edge? An Edge is defined as any activity that occurs in Facebook and has a ...Missing: explanation | Show results with:explanation
[21]
EdgeRank: How Facebook Determines What Appears in the…
Let's unpack the formula above. An “edge” refers to any piece of content you share on Facebook ... u, the w and the d. Affinity score is gauged by one's ...Missing: w_e d_e
[22]
6 Tips to Increase Your Facebook EdgeRank and Exposure
Apr 28, 2011 · An object's EdgeRank is based on three factors: affinity or the relationship between the creator and user, interaction with the object (likes, ...
[23]
(PDF) Back to Basics: Facebook's EdgeRank Algorithm for SEO
May 20, 2013 · Three factors make up the EdgeRank algorithm: affinity score, edge weight, and time decay. Affinity score measures how connected you are with ...Missing: refinements | Show results with:refinements
[24]
Facebook Edgerank: what marketers need to know
May 5, 2025 · EdgeRank is the name of the algorithm which Facebook uses to determine what appears in their users' news feeds.Missing: refinements | Show results with:refinements
[25]
Facebook EdgeRank Tell-All - Jon Loomer Digital
Jan 15, 2013 · Facebook continues to tweak the EdgeRank algorithm and include new objects into the news feed such as GraphRank. GraphRank is an algorithm ...Missing: graph | Show results with:graph
[26]
5 Things You Need to Know About Facebook's Edgerank Algorithm
Jul 17, 2012 · Consistency is imperative: The algorithm's time-decay factor means that if your last Facebook post, compelling or not, was a week ago, it's ...Missing: details | Show results with:details
[27]
What Facebook's EdgeRank Algorithm Means for Your Brand
statuses, links, photos, videos ...Missing: u_e w_e d_e<|control11|><|separator|>
[28]
Moving an Elephant: Large Scale Hadoop Data Migration at Facebook
Jul 27, 2011 · The data infrastructure team finished our largest data migration ever – moving dozens of petabytes of data from one data center to another.Missing: Feed scalability 2010-2013
[29]
Presto: Interacting with petabytes of data at Facebook
Nov 6, 2013 · Facebook's warehouse data is stored in a few large Hadoop/HDFS-based clusters. Hadoop MapReduce and Hive are designed for large-scale, reliable ...Missing: scalability 2010-2013
[30]
How Facebook went all in on AI | MIT Technology Review
Nov 14, 2023 · As Chris Cox would later note at a press event, News Feed was an instant success at boosting activity on the platform and connecting users.Missing: origins | Show results with:origins
[31]
Facebook News Feed Algorithm History | 2025 Update
Aug 7, 2025 · Facebook's algorithm has shifted from reverse chronological to popularity-based, with a 2022 overhaul to compete with TikTok, and the "Like" ...
[32]
How machine learning powers Facebook's News Feed ranking ...
Jan 26, 2021 · We are sharing new details of how we designed an ML-powered News Feed ranking system for content that's personally relevant to billions ...Missing: 2010-2013 | Show results with:2010-2013
[33]
Revealed: 50 million Facebook profiles harvested for Cambridge ...
Mar 17, 2018 · Cambridge Analytica spent nearly $1m on data collection, which yielded more than 50 million individual profiles that could be matched to electoral rolls.
[34]
FTC Imposes $5 Billion Penalty and Sweeping New Privacy ...
Jul 24, 2019 · The FTC also alleges that Facebook misrepresented users' ability to control the use of facial recognition technology with their accounts.
[35]
How does News Feed predict what you want to see? - Tech at Meta
The system determines which posts show up in your News Feed, and in what order, by predicting what you're most likely to be interested in or engage with.Missing: 2010-2013 | Show results with:2010-2013
[36]
Facebook updates its algorithm to give users more ... - TechCrunch
Oct 7, 2025 · The update includes features that offer users greater control over the videos they see, AI-powered search suggestions, and new friend bubbles.Missing: 2023-2025 rage- bait reduction
[37]
How Facebook Algorithm Ranks News Feed Content
Feb 3, 2021 · Facebook's news feed algorithm is a machine learning ranking system. It's not just one algorithm though. It's a combination of multiple algorithms that work ...Missing: 2013 onwards dwell
[38]
Facebook News Feed Algorithm - GeeksforGeeks
Jul 11, 2025 · Now, the news feed algorithm has been revamped into a machine learning approach that takes into account more than 10000 weights. The ...
[39]
Inside the Facebook Algorithm in 2025: All the Updates You Need to ...
Apr 8, 2025 · The Facebook algorithm was created so that the Meta-owned platform shows users more interesting, relevant content that's tailored to them and keeps them ...
[40]
TOP FACEBOOK ALGORITHM STATISTICS 2025 - Amra & Elma
Mar 25, 2025 · 1. Introduction of EdgeRank (2009): Facebook introduced the EdgeRank algorithm to determine the visibility of posts in users' News Feeds.<|control11|><|separator|>
[41]
Facebook's Algorithms: Understanding Meta's AI In Action
Feb 1, 2025 · In the context of social media, algorithms are mathematical ... EdgeRank primarily focused on three main factors: affinity, weight, and decay.
[42]
Feedback loops and echo chambers: How algorithms amplify ...
Feb 4, 2019 · Algorithms used by social media networks expose users to divisive content separating them into bubbles. But the ways in which they are ...
[43]
Trapped In Social Media's Validation Feedback Loop - Ben Oofana
Jun 13, 2023 · Social media uses a feedback loop where dopamine hits from likes/comments encourage more content, creating a cycle of seeking validation and ...<|separator|>
[44]
EdgeRank: Give your Facebook page the edge - Wordtracker
EdgeRank is Facebook's algorithm that determines what content users see in their news feeds, based on factors like Affinity, Weight, and Decay.Missing: u_e w_e d_e
[45]
Eli Pariser: Beware online "filter bubbles" | TED Talk
May 1, 2011 · As web companies strive to tailor their services (including news and search results) to our personal tastes, there's a dangerous unintended ...Missing: EdgeRank | Show results with:EdgeRank
[46]
Is there a filter bubble? - Will Rinehart
Feb 28, 2012 · In the most recent example, The Filter Bubble, Eli Pariser questioned the benefits of personalized content like Facebook's EdgeRank algorithm, ...
[47]
Why Is Facebook Page Reach Decreasing? More Competition And ...
Apr 3, 2014 · ... organic reach per fan (median) has steadily declined: Feb 2012 = 16%; Sep 2013 = 12.60%; Nov 2013 = 10.15%; Dec 2013 = 7.83%; Mar 2014 = 6.51 ...
[48]
Facebook organic reach began decline before algorithm change
Apr 14, 2014 · However, a new study from EdgeRank Checker shows that the decline of organic reach began in February 2012 and saw its biggest drops in November ...
[49]
Facebook Ad Revenue From 2009-2025 (Stats & Data) - Yaguara
Jan 2, 2025 · Facebook Ad Revenue: Year-By-Year ; 2010, $1,868 million ; 2011, $3,154 million ; 2012, $4,279 million ; 2013, $6,986 million ; 2014, $11,492 million.
[50]
Changes in Facebook's Edgerank reduce reach, but double ...
Nov 20, 2012 · GLOBAL – Facebook's changes to its EdgeRank algorithm have indeed resulted in 38 per cent lower reach, but a recent study by GroupM finds ...Missing: initial rollout 2009
[51]
Which content will get the best visibility in Facebook in 2013?
Dec 27, 2012 · ... Edgerank algorithm in recent weeks. It's clear that many brands have experienced a significant drop in organic reach (although to be fair I ...
[52]
Facebook EdgeRank, Promoted Posts, And The Connection - Baekdal
EdgeRank used to be a great concept for finding relevance in a stream of noise. But today, it's a tool for Facebook to force people to actively engage (and thus ...
[53]
Exposure to ideologically diverse news and opinion on Facebook
Friends shared substantially less cross-cutting news from sources aligned with an opposing ideology. People encountered roughly 15% less cross-cutting content ...
[54]
Political science. Exposure to ideologically diverse news ... - PubMed
Jun 5, 2015 · We examined how 10.1 million US Facebook users interact with socially shared news. We directly measured ideological homophily in friend networks.
[55]
Frances Haugen says Facebook's algorithms are dangerous. Here's ...
Oct 5, 2021 · The machine-learning models that maximize engagement also favor controversy, misinformation, and extremism: put simply, people just like ...
[56]
Analyzing news engagement on Facebook: tracking ideological ...
Sep 30, 2025 · While a low-quality score does not equate to the presence of misinformation in every piece of content from that outlet, we interpret low domain ...
[57]
The Decline of Organic Facebook Reach & How to Adjust to the ...
Jan 7, 2020 · According to a recent study by quintly, native Facebook videos have an 186% higher engagement rate and are shared more than 1000% more than ...
[58]
Facebook Organic Reach: 5 Facts For Businesses - InformationWeek
According to a Social@Ogilvy report, organic reach dropped from 16% of followers engaging with a brand page post in 2012 to just 6% in February of this year.Missing: EdgeRank | Show results with:EdgeRank
[59]
Mark Cuban Blasts Facebook Promote Fees - Inc. Magazine
Nov 14, 2012 · Cuban's feud with Facebook began when he posted a screenshot of an offer from Facebook asking him to pay $3,000 to reach 1 million followers on ...
[60]
Mark Cuban Is Fed up With Facebook - Business Insider
Nov 14, 2012 · Cuban thinks the social network's promotion feature for businesses isn't a good thing, and he isn't alone. Last month Facebook wanted to charge ...<|separator|>
[61]
Facebook's Mobile Revenues Strategy and "Promote Posts" in ...
Facebook said it wasn't a shakedown, Cuban obviously disagrees. Facebook Revenues? Facebook may see increased revenues from campaign spending when it ...
[62]
Meta Ad Revenue (2009–2023) [Updated Jul 2024] | Oberlo
Meta ad revenue breakdown: table ; 2012. $4.28 billion. +35.7% ; 2013. $6.99 billion. +63.3% ; 2014. $11.49 billion. +64.5% ; 2015. $17.08 billion. +48.6%.
[63]
Facebook Revenue and Usage Statistics (2025) - Business of Apps
Facebook revenue 2010 to 2024 ($bn) ; 2012, 5.08 ; 2013, 7.87 ; 2014, 12.46 ; 2015, 17.92.
[64]
Slamming political rivals may be the most effective way to go viral
Study of almost 3 million Facebook and Twitter posts from US media and politicians shows divisive posts dunking on opponents drive engagement on social media.Missing: EdgeRank | Show results with:EdgeRank
[65]
Out-group animosity drives engagement on social media - PNAS
Jun 23, 2021 · We sought to investigate whether out-group animosity was associated with increased virality on two of the largest social media platforms: ...
[66]
https://www.wsj.com/podcasts/the-journal/the-facebook-files-part-4-the-outrage-algorithm/b2069914-3bd4-4d3d-8061-b9630f520696
[67]
Engagement, User Satisfaction, and the Amplification of Divisive ...
Jan 3, 2024 · Our study reveals that the engagement- based algorithm tends to amplify emotionally charged content, particularly that which expresses anger and ...
[68]
Conservatives Accuse Facebook of Political Bias
May 9, 2016 · Facebook responded to a report by the website Gizmodo that accused the social network of suppressing stories from conservative news sources.Missing: EdgeRank | Show results with:EdgeRank
[69]
The Facts Behind Allegations of Political Bias on Social Media | ITIF
Oct 26, 2023 · Facebook claims that conservatives' advantage on social media comes down to the tactics right-wing populist figures use being best suited to ...Missing: EdgeRank | Show results with:EdgeRank
[70]
Facebook commissioned a study of alleged anti-conservative bias ...
Aug 20, 2019 · The audit came at Facebook's request last year, following repeated claims of anti-conservative bias, including from President Donald Trump. The ...
[71]
Neutral bots probe political bias on social media - PMC - NIH
We find no strong or consistent evidence of political bias in the news feed. Despite this, the news and information to which U.S. Twitter users are exposed ...
[72]
[PDF] Facebook tweaked its News Feed algorithm, and right-leaning ...
Mar 27, 2023 · Even after Facebook slightly tweaked the algorithm last year, our findings suggest that the change has not affected right-leaning pages, which ...Missing: EdgeRank | Show results with:EdgeRank
[73]
Rising Through the Ranks: News Feed Ranking - New America
This section of the report focuses on how internet platforms deploy algorithmic curation and ranking practices in order to shape and operate their news feeds.