Chapter 1: Recommender systems in action
The introduction illustrates how user behavior and recommender systems can reinforce each other and potentially steer people toward more extreme content.
After this chapter you will be able to
- User recommendation feedback loop. Trace how a user’s interactions with recommendations become signals used to generate further recommendations.
- Extremity and shared influence. Explain how the cousin example illustrates a possible path toward extreme content and why the section considers both user behavior and recommendations.
- Relevance and ranked choices. Explain how recommender systems help people find relevant options among many possibilities.
- Consequences of ranking attention. Explain how directing attention to highly ranked options can have real-world consequences for those ranked lower.
- Finding Relevant Content. Explain how recommender systems help users find appealing content and why this is especially useful on platforms with large, continuously growing collections.
- User and Platform Incentives. Explain how recommendations can benefit both users and platforms, and describe the observations the section offers in support of that possibility.
- Personalization and Societal Tradeoffs. Describe the societal downsides the section associates with personalization, alongside its potential to bring people with similar interests together.
- Ranking and personalization. Explain how recommender systems rank items toward a goal, and distinguish personalized from non-personalized ranking using the section’s examples.
- Machine learning and opacity. Explain the role of past data and interactions in modern recommendations and why particular recommendations can be hard to explain.
- Social media defining features. Describe the features that define social media in this book and distinguish it from purely private messaging.
- Public sharing and virality. Explain how public sharing enables content to go viral.
- Traditional and social media constraints. Contrast traditional media’s production oversight and physical constraints with social media’s more open, screen-based publishing.
- Early and Mainstream Platforms. Distinguish early topic-based online platforms from the mainstream social media that followed.
- Social Media as News. Explain how social media became a source of real-time news and gave a voice to people outside traditional reporting.
- Speed and Information Risks. Explain the risks that accompany rapid, unreviewed social media news and the appeal of extreme statements.
- Personalized ranking and preferences. Explain why Facebook introduced algorithmic ranking and how EdgeRank used post recency and interaction frequency to personalize feeds.
- Engagement and platform incentives. Explain the shift toward predicted engagement and why platforms have an incentive to keep users spending time on their sites.
- Social media distribution models. Distinguish the subscription, network, and algorithmic models by where feed content comes from and who enables it to reach beyond its creator's followers.
- Ranking Benefits and Creator Pressure. Explain how ranking creates winners and influences content creators’ decisions.
- Open Social Media Effects. Describe how social media’s lack of gatekeepers can support collective action while also enabling misinformation and disinformation.
- Engagement and Intermittent Reward. Explain how public engagement can encourage repeated checking and longer platform use.
- Similar Content and Vulnerability. Explain how recommendations for similar content can worsen a vulnerable person’s situation and why supervision raises a freedom-of-speech issue.
- Network Effects and Concentration. Explain how network size attracts users and makes it difficult for new platforms to compete with established ones.
- Corporate Influence on Regulation. Explain how the section says technology companies use arguments and lobbying to resist meaningful regulation.
- Concentrated Recommendation Power. Explain how concentrated ownership of recommendation systems gives platform owners consequential influence over what people see and potentially over public opinion.
- Recommendations and limited attention. Explain how recommendations allocate limited attention, especially during passive consumption.
- Social influences on opinion. Explain how social media can shape opinions through both consumed content and other people's reactions.
- Amplification and measurement stakes. Explain why measuring recommender systems matters for culture and democracy, and describe the concern about overamplifying niche views.
- Value and engagement measures. Explain why platforms use engagement to evaluate recommendations and why it is an imperfect measure of user value.
- Popularity feedback loop. Trace how prominent display can make already popular content even more popular, and explain why popularity alone does not establish value.
- Population impact and attribution. Explain why assessing societal impact requires considering relationships among users and algorithms at the population level.
- Evolving recommendation feedback. Trace how user interactions can shape later versions of a recommender system and encourage further interactions.
- Algorithmic amplification and exposure. Explain algorithmic amplification in terms of content exposure, and describe why measuring it is not straightforward.
- Amplification reports and limits. Describe the nutrition-label analogy for amplification reports and its limits.
- Power to define measures. Explain why deciding how amplification is measured is an issue of power.
- Selection and ranking amplification. Explain why amplification should be considered in both content selection and ranking.
Section by section
Introduction
The introduction illustrates how user behavior and recommender systems can reinforce each other and potentially steer people toward more extreme content.
- User recommendation feedback loop: Trace how a user’s interactions with recommendations become signals used to generate further recommendations.
Why it matters: This cycle helps explain why user actions and algorithmic recommendations cannot be understood in isolation. - Extremity and shared influence: Explain how the cousin example illustrates a possible path toward extreme content and why the section considers both user behavior and recommendations.
Why it matters: The example shows why judging the effects of recommendations requires looking at both the person and the system.
Key terms: feedback loop, recommendations, signals, algorithmic recommendations, extreme content, users’ behavior, algorithmic input
1.1 Recommender systems are everywhere
This section explains how recommender systems help people navigate abundant choices, where they appear, and how their rankings can influence the world.
- Relevance and ranked choices: Explain how recommender systems help people find relevant options among many possibilities.
Why it matters: Ranking helps explain how recommendations make an overwhelming set of choices easier to navigate. - Recommendations across everyday activities: Describe how recommender systems are used beyond social media, using examples of the activities they mediate.
Why it matters: Recognizing recommendations beyond social media reveals how broadly they affect everyday choices. - Consequences of ranking attention: Explain how directing attention to highly ranked options can have real-world consequences for those ranked lower.
Why it matters: A ranking is not just a convenient list: it can affect who gets attention and business.
Key terms: relevant, top-ranked content, recommender systems, e-commerce, travel platforms, social media, dating apps, top-ranked results, funneling of attention, market power
1.1.1 A world without RecSys
This section uses a world without recommendations to explain their benefits for users and platforms and their potential societal downsides.
- Finding Relevant Content: Explain how recommender systems help users find appealing content and why this is especially useful on platforms with large, continuously growing collections.
Why it matters: Recommendations make it easier to find things you like without searching an entire catalog or feed. - User and Platform Incentives: Explain how recommendations can benefit both users and platforms, and describe the observations the section offers in support of that possibility.
Why it matters: The same recommendations can serve both a user's interests and a platform's business goals, though that does not eliminate downsides. - Personalization and Societal Tradeoffs: Describe the societal downsides the section associates with personalization, alongside its potential to bring people with similar interests together.
Why it matters: Personalization can help people find communities while also affecting the shared culture and reality they have in common.
Key terms: recommender systems, inferred preferences, user-generated content, incentives, business objectives, algorithmic feeds, engagement, hyper-personalization, filter bubbles, shared sense of reality
1.1.2 What exactly are RecSys?
This section defines algorithms and recommender systems, distinguishes personalized ranking, and introduces how past interactions can influence recommendations.
- Algorithms as task instructions: Explain what an algorithm is and illustrate it with a non-computer or computer example.
Why it matters: This definition provides a starting point for understanding recommender systems as algorithms. - Ranking and personalization: Explain how recommender systems rank items toward a goal, and distinguish personalized from non-personalized ranking using the section’s examples.
Why it matters: Recognizing the goal behind an ordering helps you distinguish ranking itself from personalization. - Cross-app influence on feeds: Describe how activity in one part of an app can affect a social media feed, including a possible benefit or risk.
Why it matters: A feed can reflect activity beyond the feed itself, with helpful or harmful consequences. - Machine learning and opacity: Explain the role of past data and interactions in modern recommendations and why particular recommendations can be hard to explain.
Why it matters: Knowing that individual recommendations can be opaque helps explain the need to measure their impact.
Key terms: algorithm, input, output, recommender systems, algorithmic ranking, personalized algorithmic ranking, broader design, feed, rabbit hole, machine learning
1.2 Social media: From bulletin boards to doom scrolling
This section defines social media, explains how public sharing enables virality, and contrasts its reach and production constraints with those of traditional media.
- Social media defining features: Describe the features that define social media in this book and distinguish it from purely private messaging.
Why it matters: This definition sets the boundaries for the book’s discussion of social media. - Public sharing and virality: Explain how public sharing enables content to go viral.
Why it matters: Virality shows how a post can reach well beyond the people who first saw it. - Traditional and social media constraints: Contrast traditional media’s production oversight and physical constraints with social media’s more open, screen-based publishing.
Why it matters: The contrast helps explain why social media produces so much content and reaches audiences differently.
Key terms: social media, direct connection, public sharing, viral, original audience, traditional media, editors, fact-checked
1.2.1 The explosive success of social media
This section traces social media’s growth from early online communities to mainstream platforms and examines its benefits and risks as a source of real-time information.
- Communities Beyond Geographic Boundaries: Explain how the internet enabled people to form communities around shared interests.
Why it matters: Online communities can connect people who might otherwise struggle to find others with the same interests. - Early and Mainstream Platforms: Distinguish early topic-based online platforms from the mainstream social media that followed.
Why it matters: The shift in audience, technology, and funding helps explain how social media became mainstream. - Social Media as News: Explain how social media became a source of real-time news and gave a voice to people outside traditional reporting.
Why it matters: Social media can change both who reports an event and which perspectives reach the public. - Speed and Information Risks: Explain the risks that accompany rapid, unreviewed social media news and the appeal of extreme statements.
Why it matters: The same speed that makes social media useful during events can also make its information harmful or misleading.
Key terms: communities, common interests, BBS, bulletin board system, Usenet, social media platforms, monetize, breaking news, backchannel, traditional news sources
1.2.2 RecSys in social media
This section traces the evolution of social media feeds from chronological updates to personalized ranking and compares three models of content distribution.
- Personalized ranking and preferences: Explain why Facebook introduced algorithmic ranking and how EdgeRank used post recency and interaction frequency to personalize feeds.
Why it matters: This shows why social feeds moved beyond simply displaying the newest updates. - Interaction feedback loop: Describe the feedback loop that can arise when a feed shows more updates from a frequently contacted friend.
Why it matters: A feed can influence the interactions it later treats as evidence of interest. - Ranking formula transparency: Distinguish the transparency of EdgeRank's weighting formula from that of contemporary machine-learning recommendations as described in the section.
Why it matters: Knowing a ranking outcome is different from being able to explain why it occurred. - Engagement and platform incentives: Explain the shift toward predicted engagement and why platforms have an incentive to keep users spending time on their sites.
Why it matters: What a platform optimizes can reflect its incentives, not just what users say they value. - Social media distribution models: Distinguish the subscription, network, and algorithmic models by where feed content comes from and who enables it to reach beyond its creator's followers.
Why it matters: These models show how platforms gained a greater role in determining which content reaches a reader.
Key terms: EdgeRank, personalized relevance, reverse-chronological order, feedback loop, weighting formula, counterfactual questions, engagement, behavioural data, surveillance capitalism, subscription model
1.3 Societal impact
This section describes how social media and recommender systems affect people, creators, and society, from enabling collective action to increasing exposure to harmful content.
- Ranking Benefits and Creator Pressure: Explain how ranking creates winners and influences content creators’ decisions.
Why it matters: Ranking can affect both creators’ livelihoods and what they choose to make. - Open Social Media Effects: Describe how social media’s lack of gatekeepers can support collective action while also enabling misinformation and disinformation.
Why it matters: Social media’s societal effects are not all caused by recommendation algorithms, and its openness has both benefits and risks. - Engagement and Intermittent Reward: Explain how public engagement can encourage repeated checking and longer platform use.
Why it matters: Public feedback can shape how people feel about themselves and how long they spend on a platform. - Similar Content and Vulnerability: Explain how recommendations for similar content can worsen a vulnerable person’s situation and why supervision raises a freedom-of-speech issue.
Why it matters: Recommendations that follow apparent interests can intensify an existing vulnerability. - Hateful Content Reach: Describe the section’s account of how emotionally charged hateful content can spread and its example involving the Rohingya genocide.
Why it matters: The reach of hateful content can have serious consequences beyond the platform.
Key terms: ranking algorithm, content creators, gatekeepers, mis- and disinformation, social media, intermittent reward, public engagement, recommender systems, harmful content, content moderation
1.3.1 The influence of platforms
This section explains how network effects, regulatory influence, and concentrated ownership give major platforms power over society and public opinion.
- Network Effects and Concentration: Explain how network size attracts users and makes it difficult for new platforms to compete with established ones.
Why it matters: Network effects help explain why a few platforms can become so dominant. - Corporate Influence on Regulation: Explain how the section says technology companies use arguments and lobbying to resist meaningful regulation.
Why it matters: Corporate influence can help explain why powerful platforms face limited constraints. - Regulation as Entry Barrier: Explain how a regulation can impose relatively greater burdens on smaller entrants and thereby reinforce established companies.
Why it matters: A company's support for a rule does not necessarily mean the rule will reduce its market power. - Concentrated Recommendation Power: Explain how concentrated ownership of recommendation systems gives platform owners consequential influence over what people see and potentially over public opinion.
Why it matters: Choices made by a small number of platform owners can have consequences far beyond individual feeds.
Key terms: network effect, network, monopolies, regulation, lobbying, legislation, barrier to entry, General Data Protection Regulation, GDPR, compliance obligations
1.3.2 Recommendation as amplification
This section explains how recommendations allocate attention, shape opinions, and make measuring their societal effects important.
- Recommendations and limited attention: Explain how recommendations allocate limited attention, especially during passive consumption.
Why it matters: Recommendations affect what gets noticed even when we do not actively choose each item. - Social influences on opinion: Explain how social media can shape opinions through both consumed content and other people's reactions.
Why it matters: A feed can influence thinking through both its content and the visible reactions around it. - Amplification and measurement stakes: Explain why measuring recommender systems matters for culture and democracy, and describe the concern about overamplifying niche views.
Why it matters: Measurement can help us question whether recommendations distort public discourse or normalize dangerous views.
Key terms: attention, passive consumption, infinite scrolling, autoplay, social media, opinions, peers, amplification, overamplifying, measuring recommender systems
1.3.3 Measuring recommender systems
This section explains why measuring recommender systems and their societal impact is difficult, highlighting imperfect metrics and feedback among users, content, and algorithms.
- Value and engagement measures: Explain why platforms use engagement to evaluate recommendations and why it is an imperfect measure of user value.
Why it matters: This distinction helps explain why a recommendation can perform well by a platform’s measure without being valuable to you. - Popularity feedback loop: Trace how prominent display can make already popular content even more popular, and explain why popularity alone does not establish value.
Why it matters: This shows how the way a platform presents content can help produce the popularity it measures. - Population impact and attribution: Explain why assessing societal impact requires considering relationships among users and algorithms at the population level.
Why it matters: A population-level view helps us see why an individual recommendation’s effect is only part of the societal picture. - Evolving recommendation feedback: Trace how user interactions can shape later versions of a recommender system and encourage further interactions.
Why it matters: The system being measured changes over time partly in response to the behavior it has already influenced. - Competition and extreme content: Explain the section’s account of how competition for engagement may make recommended content more extreme over time.
Why it matters: This illustrates a possible way engagement-driven competition can change the character of a feed.
Key terms: Value, engagement, behavioural signals, popularity, self-reinforcing feedback loop, societal impact, population level, amplification, preferences, interactions
1.3.4 Algorithmic amplification as a measure of societal impact
This section introduces algorithmic amplification as a way to examine recommender systems’ influence, its limitations, and the importance of how it is defined.
- Algorithmic amplification and exposure: Explain algorithmic amplification in terms of content exposure, and describe why measuring it is not straightforward.
Why it matters: This idea provides a starting point for examining what recommender systems promote. - Amplification reports and limits: Describe the nutrition-label analogy for amplification reports and its limits.
Why it matters: A useful measure can inform choices without describing everything about a platform. - Power to define measures: Explain why deciding how amplification is measured is an issue of power.
Why it matters: The definition of a measure shapes which effects people pay attention to. - Selection and ranking amplification: Explain why amplification should be considered in both content selection and ranking.
Why it matters: Looking only at ranking misses the influence of choosing what gets considered in the first place. - Nonhuman algorithmic decisions: Distinguish convenient language about algorithms deciding from what the section says algorithms actually do.
Why it matters: This distinction prevents ordinary shorthand from making automated processes sound like people.
Key terms: algorithmic amplification, exposure, amplification reports, nutrition labels, concrete measures, scientifically grounded understanding, selection, ranking, out-of-network, anthropomorphize
1.4 Summary
Where this is used later
- Feedback loops in recommender systems and social media (chapter 4) · Provides the basic user–recommendation cycle for discussing feedback loops in social media.
- Feedback loops and extreme content (chapter 8) · Provides the cycle underlying the discussion of feedback loops and extreme content.
- Rabbit holing (chapter 8) · Further discusses paths through recommended content.
- Heavy ranking (chapter 2) · Introduces the purpose of ordering items before the discussion of heavy ranking.
- Platforms’ design and amplification (chapter 2) · Introduces the connection between platform choices and amplification.
- Attention economy (chapter 6) · The platform objective of time spent connects to the discussion of the attention economy.
- Homogenization (chapter 7) · The possibility of homogenized culture and content connects to the later discussion of homogenization.
- Parts of a RecSys (chapter 2) · The components of a recommender system build on the idea of instructions that perform tasks.
Concepts explained in this chapter
- recommender systemA computational system that selects, ranks, or presents items to users based on inferred preferences or other criteria.
- rank orderAn ordering of items according to their relative position or assessed importance.
- feedback loopA cyclic process in which a system's outputs influence subsequent inputs, causing later outputs to be shaped by earlier ones.
- social mediaAn online service that supports user-generated content, communication, and social interaction.
- user engagementThe degree to which users interact with or remain attentive to content or a platform.
- algorithmic feedA feed whose contents or ordering are selected by algorithms rather than solely by chronological or follow-based rules.
- amplificationAn increase in the visibility, reach, or influence of information or viewpoints through a communication or distribution process.
- societal impactThe effects that a technology, policy, or activity has on society, including its effects on behavior, culture, institutions, and public life.
- recommendation algorithmAn algorithm that selects or ranks items for recommendation based on user, item, or other information.
- algorithmic amplificationAn increase in the visibility, reach, or influence of content caused by automated algorithmic selection and distribution.
- EdgeRankFacebook's early feed-ranking algorithm that weighted post recency and user interaction relationships to estimate relevance.
- algorithmic modelA social-media distribution model in which algorithms select and rank content from both followed and not-explicitly-followed users or topics.
- content reshareA social-media capability that lets a user distribute content created by another user to the user's own network.
- in-network contentSocial-media content originating from users whom a reader explicitly follows, including content they reshare.
- indirect observationInference of a user's preferences from observed behavior rather than from explicitly stated preferences.
- network modelA social-media distribution model in which users can reshare followed users' content to extend its reach through the social graph.
- out-of-network contentContent produced by users or sources that a reader does not directly follow.
- personalized relevanceThe estimated usefulness or interest of an item for a particular user.
- subscription modelA social-media distribution model in which a user's feed consists of content from creators the user explicitly follows.
- behavioral dataData recording users’ actions and interactions, including their activity while browsing websites.
- mutual influenceA reciprocal causal relationship in which users affect algorithms and algorithms affect users.
- popularityThe degree to which content attracts attention or use across a population of users.
- valueThe subjective benefit or worth that a user attributes to recommended content.
- barrier to entryA cost, constraint, or disadvantage that makes it difficult for new competitors to enter a market.
- monopolyA market condition in which a single seller or small controlling entity dominates the supply of a product or service.
- network effectA market effect in which the value of a product or service increases as more participants join its network.
- online platformAn internet-based service that provides digital functionality and connects users, content, or transactions.
- regulationThe use of legally enforceable rules and oversight to govern conduct in a market or sector.
- machine learningA computational approach in which models use data and interactions to learn patterns for generating predictions or recommendations.
- recommendationA content or item suggestion generated by a recommender system for a particular user.
- surveillance capitalismAn economic system in which behavioral data gathered through monitoring people is treated as a source of prediction, influence, and profit.
- extreme contentContent expressing unusually intense, radical, or socially extreme viewpoints or language.
- content moderationThe review, filtering, labeling, or removal of user-generated or platform-hosted content according to defined rules.
- free speechThe civil liberty to communicate opinions, ideas, and information without unjustified restriction.
- unsafe contentContent that can cause or materially contribute to physical, psychological, social, or other significant harm.
- autoplayAutomatic playback of a media item without requiring the user to initiate playback for that item.
- cultureThe shared beliefs, norms, practices, and artifacts through which a community interprets the world and relates to others.
- endless scrollA user-interface pattern that loads additional content continuously as a user scrolls, without requiring navigation to another page.
- personalizationThe adaptation of item selection, ordering, or presentation to an individual user's inferred characteristics or preferences.
- shared understandingA commonly held understanding of events, facts, or social conditions among members of a community.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.
