aiMOOC

E9 M - Recommendation systems and choices

QR-Code

E9 M - Recommendation systems and choices

Mini-description: Learn how platforms can recommend personalised content and reflect on how recommendations may influence choices. You work mainly in English at a B1-oriented level and use German only as a short support.

Look first. What do you notice? What may happen between a large content pool and your personal feed?

Deutsch-Hilfe: Eine Plattform kann aus sehr vielen Inhalten eine kleinere, personalisierte Auswahl anzeigen. Du sollst verstehen, wie diese Auswahl entstehen kann und wie sie Entscheidungen beeinflussen kann.

Learning path

English first → understand → respond → ask → hint → improve → use again

Understand: notice the main idea in a picture, video or short text.
Respond: give a short reaction in English.
Ask: ask when a word, claim or recommendation is unclear.
Hint: use sentence starters, context and a short German help box.
Improve: revise your answer after feedback.
Use again: transfer the language to a new example.

I-can goals

By the end, you can say:

I can explain in simple English what a recommendation system does.
I can name signals that may be used for recommendations.
I can compare two personalised feeds.
I can discuss benefits and risks without assuming that every recommendation has the same effect.
I can explain how a recommendation may influence a choice.
I can ask critical questions about sources, data and AI-generated feedback.
I can write a coherent short opinion with reasons and an example.
I can mediate key information from German into English without translating every sentence.

What you already know

Think for one minute.

Where do you see “For you”, “Recommended”, “You may also like” or similar labels?
What makes you click on one item but not another?
Can two people open the same platform and see different content?

Rule for this course: Use invented examples only. Do not share private social-media histories, real personal profiles, passwords, private messages or other sensitive data.

Media entry: one pool, different feeds

A recommendation system selects and ranks items from a much larger pool. The exact method can differ between platforms.

Quick response: Complete the sentence: “The system does not show everything. It first …”

Hint: Try: “It first selects possible items and then ranks them.”

Core idea: what is a recommendation system?

A recommendation system is software that suggests items which may be relevant to a user. These items can be videos, songs, films, products, articles or posts.

It does not “know you” like a human friend. It works with data and patterns. A system may use information about items, past interactions or patterns from many users.

Two simple ideas:

Method	Simple explanation	Example
content-based	It looks for items with similar features.	You liked a science video, so another science video may appear.
collaborative	It looks for patterns across users or items.	People with similar viewing patterns also watched another video.

You do not need the mathematics behind these methods in this course.

Deutsch-Hilfe: content-based = ähnliche Inhalte; collaborative = Muster aus ähnlichem Nutzungsverhalten.

Signals: what can a platform learn from?

A platform may use signals. For example, YouTube publicly describes signals such as clicks, watch time, likes, dislikes, sharing and survey responses. Different platforms may use different signals and weight them differently.

Important: A signal is not the same as a true personal preference. You may click because you are curious, surprised or doing homework.

Think-pair-share: Which is stronger evidence of interest: one click, a full watch, a repeated search, or a direct rating? Explain your choice.

Feedback loop

A recommendation can change what you see. What you see can change what you click. Your new interaction can then become another signal.

This creates a feedback loop:

recommendation → interaction → new data → new recommendation

A loop can be useful because suggestions can become more relevant. But repeated similar suggestions may also reduce variety.

Choices: influence is not control

Recommendations can make some options more visible than others. Visibility can influence attention and decisions, but it does not mean that an algorithm fully controls a person’s choice.

Ask three questions:

What is shown?
What is missing?
What can I still choose or search for myself?

Useful language:

Function	Language
possibility	The recommendation may influence my choice.
comparison	Feed A shows more … than Feed B.
reason	I would choose this because …
uncertainty	We cannot know from one example whether …
alternative	Another option would be …

Filter bubbles and variety

A filter bubble is a concept used to describe a situation in which personalised filtering may narrow the information a person sees. Researchers do not agree that every personalised system always creates a strong filter bubble or changes opinions in the same way. Context, platform design and user behaviour matter.

Critical sentence frame: “This feed may be narrow because …, but we would need more evidence to know …”

Deutsch-Hilfe: Nicht vorschnell behaupten: „Der Algorithmus macht alle gleich.“ Besser: Möglichkeit, Beobachtung und Beleg unterscheiden.

Listening and viewing 1: two experts

This WIRED interview features two members of YouTube’s Search and Discovery team. The speech is natural and moderately fast.

Before watching: Predict three words you may hear: recommend, signal, watch.

While watching: Listen for answers to these questions:

What kinds of user actions are mentioned?
Why can different people receive different recommendations?
How do the speakers describe search and recommendation?

After watching: Give a 30-second summary with this frame: “The speakers explain that … One important signal is … I still want to know …”

Listening strategy: Do not stop at every unknown word. First listen for repeated key words and examples.

Listening and viewing 2: the filter-bubble idea

This TED talk introduced the filter-bubble idea to a wide audience. It is an older talk, so use it as one perspective, not as proof that all current platforms work in exactly the same way.

Viewing task: Find one claim, one example and one question you would ask today.

Source check: Separate what the speaker argues from what you can verify with newer sources.

Reading: A fictional recommendation test

Profile A – Jordan often watches short basketball clips, school science videos and comedy.

Profile B – Sam often watches cooking tutorials, travel clips and school science videos.

A fictional platform recommends:

To Jordan	To Sam
“Amazing last-second basketball shots”	“Five easy pasta sauces”
“Why volcanoes erupt”	“Why volcanoes erupt”
“Best comedy moments”	“Train travel through Italy”

Reading questions: What is personalised? What is shared? Which recommendation could come from content similarity? Which could come from a broader user pattern?

Important: These profiles are invented for learning. Do not replace them with real classmates’ private data.

Context vocabulary

Word	Meaning in this topic	Example
recommendation	a suggested item	This recommendation looks relevant.
signal	information used by a system	A click can be a signal.
rank	put items in an order	The system ranks possible videos.
feed	a stream of selected content	My fictional feed shows sports and science.
personalised	adapted to one user or group	The homepage may be personalised.
pattern	something that repeats	The model looks for patterns.
bias	a systematic tendency	A system can reproduce a bias in data.
variety	different types of content	More variety can show new topics.
choice	a decision between options	A visible option can affect a choice.
evidence	information supporting a claim	We need evidence before we generalise.

Pronunciation and fluency

Practise these chunks as whole phrases:

a recommendation system – stress the key content words.

personalised content – keep the phrase smooth, not word-by-word.

It may influence my choice. – stress may, influence and choice.

I need more evidence. – use falling intonation for a clear statement.

Fluency challenge: Speak for 40 seconds. Use at least three chunks without reading a full script. Clarity is more important than accent.

Recording rule: Recording is voluntary. Keep recordings private unless you actively choose to share them. Do not upload them to an external AI service without permission.

Speaking: compare, react, negotiate

Work with the two fictional feeds.

Round 1 – Compare: “Both feeds show …, but Feed A … whereas Feed B …”

Round 2 – React spontaneously: Your partner says, “Personalisation is always helpful.” Reply with one agreement point and one limitation.

Round 3 – Negotiate: Design a better feed together. It should contain familiar content, one surprising item and one reliable information source.

Round 4 – Present: Give a one-minute explanation of your choices.

Hint ladder:

Start: “We kept … because …”
Add contrast: “However, we also added …”
Add caution: “This may help, but …”

Functional grammar

Use may, might, can to avoid claims that are too strong.

Too strong	Better for careful reasoning
The algorithm changes your opinion.	The recommendation may influence what you notice.
One click means you like the topic.	One click can be a signal, but it may have other reasons.
Personalisation creates a filter bubble.	Personalisation might reduce variety in some situations.

Use an if-clause for cause and effect:

“If I repeatedly watch similar videos, the system may recommend more similar content.”

Writing: short opinion

Write 120–160 words:

“Personalised recommendations are useful, but users should stay active decision-makers.” Discuss.

Use this structure:

opening opinion
one benefit
one risk or limitation
one example
one practical strategy
short conclusion

Keep your own text. If you use AI feedback, compare it with your original version. Accept only changes you understand. Check factual claims and sources yourself.

Writing help: Useful connectors: firstly, however, for example, because, therefore, in my view, on the other hand.

Mediation

Situation: An English-speaking exchange student asks what your school means by this German notice. Explain the key message in English. Do not translate sentence by sentence.

German source: „Empfehlungen auf Plattformen können personalisiert sein. Prüfe deshalb, warum dir etwas angezeigt wird, suche bei wichtigen Themen auch selbst nach weiteren Quellen und teile im Unterricht keine privaten Nutzungsdaten.“

Your task: Give a 40–60 word English explanation for the exchange student.

Hint: Start with: “The notice says that personalised recommendations can shape what you see. It advises students to …”

Ask better questions

Instead of asking only “Is this recommendation good?”, ask:

What information might the system be using?
What goal might the ranking have?
What other content could have been shown?
Is the source reliable?
Can I change settings, search directly or ask for another view?
What evidence supports the claim I am seeing?

Feedback and error analysis

After a speaking or writing task, mark one sentence in each category:

Check	Question
meaning	Did I answer the task clearly?
evidence	Did I separate observation from assumption?
language	Did I use may, might or can where needed?
cohesion	Did I connect ideas with because, however or for example?
source check	Did I verify important AI or platform claims?

Improve one thing, not everything. Then use the improved sentence again in a new example.

Safe and responsible use

Use fictional profiles and invented recommendation data in class. Do not collect private social-media histories from classmates. Do not upload another person’s work, image, voice or data to an external AI system without permission. AI-generated explanations, facts, sources and feedback can be wrong, so check them. Keep your own original text so that you can see what you changed and why.

Can-do check

I can …	Yes	Not yet
explain a recommendation system simply	☐	☐
name and question recommendation signals	☐	☐
compare two feeds	☐	☐
discuss influence without overclaiming	☐	☐
give a short spoken response	☐	☐
write a coherent opinion	☐	☐
mediate a short German message into English	☐	☐

Interaktive Aufgaben

Quiz: Teste Dein Wissen

What does a recommendation system mainly do? (It suggests items that may be relevant to a user) (!It shows every available item) (!It removes all user choice) (!It only works with written text)

Which item can be a recommendation signal? (Watch time) (!Screen colour) (!Battery shape) (!Keyboard size)

What is a careful way to describe influence? (A recommendation may influence what I notice) (!A recommendation always controls my decision) (!A recommendation proves my opinion) (!A recommendation makes all users identical)

What does content-based recommendation focus on? (Similar features of items) (!Only the user’s age) (!Random numbers only) (!The device brand only)

What does collaborative filtering use in a simple explanation? (Patterns across users or items) (!Only one fixed list for everyone) (!Only spelling rules) (!Only the newest item)

Why can one click be difficult to interpret? (It can have different reasons) (!It always means strong interest) (!It cannot be recorded) (!It always means dislike)

What is a feedback loop in recommendation? (Interactions can affect later recommendations) (!Recommendations never change) (!Users see no content) (!Only teachers can create signals)

What is a useful source-checking question? (What evidence supports this claim) (!How colourful is the page) (!How fast can I share it) (!How many emojis are visible)

Which modal verb helps express uncertainty? (may) (!must always) (!proves) (!guarantees)

What is good classroom practice for this topic? (Use fictional profiles instead of private social-media data) (!Share classmates’ private histories) (!Upload recordings without asking) (!Copy AI feedback without checking it)

Memory

recommendation	suggested item
signal	information used by a system
ranking	ordering possible items
variety	different kinds of content
evidence	support for a claim
feedback loop	interaction affects later suggestions

Drag and Drop

Ordne die richtigen Begriffe zu.	Thema
Signal	A click or watch time
Ranking	Putting items in an order
Personalisation	Adapting suggestions to a user
Variety	Showing different types of content
Evidence	Information that supports a claim

Kreuzworträtsel

signal	What word means information that can be used by a recommendation system?
ranking	What word means putting possible items into an order?
profile	What word can describe a set of information linked to a user?
choice	What word means a decision between options?
bubble	What word completes the term filter ...?
feedback	What word completes the term ... loop?

LearningApps

Lückentext

Offene Aufgaben

Leicht

Recommendation spotting: Find three recommendation labels on fictional screenshots or teacher-provided examples and explain what each label suggests.
Signal cards: Create six cards with possible signals and sort them into “strong clue”, “weak clue” or “depends”.
Feed comparison: Compare the two fictional profiles from the reading and write five sentences using both, but and whereas.
Question maker: Write four critical questions you could ask before following a recommendation.

Standard

Mini debate: Discuss “Personalisation saves time” with one benefit, one limitation and one example.
Design a feed: Build a fictional six-item feed for a made-up learner and explain your ranking in English.
Listening response: Use one of the embedded videos and record a private 60-second summary or present it live; recording is optional.
Mediation challenge: Write an English message that explains the German school notice to an exchange student without translating line by line.

Schwer

Recommendation experiment: With teacher-provided fictional data, change one signal and predict how the ranking might change. Explain why your result is only a hypothesis.
Source triangle: Compare a platform source, a Wikipedia article and one independent educational source. Note where they agree and where they use different language.
Fair feed proposal: Design rules for a feed that balances relevance, variety and user control, then defend the trade-offs.
One-minute panel: In groups, take roles as user, platform designer and media educator. Negotiate one change that could make recommendations more transparent.

Lernkontrolle

Transfer to shopping: A fictional shop recommends only products similar to the last item clicked. Explain one benefit, one possible problem and one way the user could broaden the search.
Evidence check: A student says, “My feed changed after one click, so the platform knows exactly what I want.” Explain why this conclusion is too strong and rewrite it more carefully.
Feed design: Choose five items from a teacher-provided content pool for two different fictional users. Explain your ranking criteria and what information you deliberately did not use.
Decision pathway: Describe how a recommendation could influence a choice from first view to final decision. Include at least one point where the user can actively change direction.
Platform comparison: Two fictional platforms use different signals. Explain how this could produce different feeds even for the same person.

Lernnachweis

For a successful learning record, show that you can:

explain the basic purpose of a recommendation system in clear English.
use key vocabulary such as signal, ranking, feed, personalised, choice and evidence.
compare recommendations and explain possible reasons for differences.
distinguish a fact, an observation, an assumption and an opinion.
use careful language such as may, might and can.
participate in a short discussion and react to another view.
write a structured 120–160 word opinion.
mediate key information from German into English.
explain safe handling of personal data, recordings and AI feedback.
verify important claims with suitable sources.

Sources and fact check

Wikipedia: Empfehlungsdienst – overview of recommendation systems.
Wikipedia: Recommender system – English background reading.
YouTube: On YouTube’s recommendation system – platform description of recommendation signals and goals.
Wikimedia Commons: Recommender systems – freely licensed learning media.
Wikimedia Commons: Filter bubbles – freely licensed diagrams for critical discussion.

OERs zum Thema

Follow-up course

Continue with E9 M - Algorithms, feeds and responsibility to analyse ranking goals, transparency and responsible platform design in more depth.

Verknüpfte Lernbereiche

E9 M - Recommendation systems and choices

aiMOOC-Projekte

Inhalte werden geladen ...

Mediathek wird aus dem Wiki geladen ...

Frag mich zum aktuellen Lerninhalt oder wähle eine kurze Hilfe.

Textmodus bereit

Sprache