Questions & explanations
1. What is cross-lingual moderation?
Cross-lingual moderation is the process of checking user content in many different languages to enforce platform rules. The main challenge is that no single moderator can understand all languages, so platforms use translation tools or hire teams of speakers. Another challenge is that slang, jokes, or hate speech may not translate directly. Methods include using machine translation to flag suspicious posts, then sending them to human moderators who know the language. Some platforms also build language-specific AI models trained on local data. The goal is to catch harmful content like hate speech or misinformation across all languages equally.
2. Compare graph-based rumor detection with text-based detection.
Text-based detection analyzes the words in a post to spot false claims, like looking for sensational language or contradictions. Graph-based detection ignores content and focuses on how the post spreads. Text methods can be fooled by sophisticated wording, while graph methods are harder to evade because spread patterns are less controllable. However, graph methods need enough sharing data to work, so they may not catch early rumors. Combining both gives better accuracy: text flags suspicious content, graph confirms via spread pattern. For example, a post with fake-looking text that spreads like a rumor is highly likely false.
3. Compare a content-based recommendation with a collaborative filtering one. How do they differ?
Content-based recommendation suggests items similar to what the user already liked, based on features like genre or creator. For example, if you liked a 'Fortnite' video, it recommends more 'Fortnite' videos. Collaborative filtering looks at what other users with similar tastes liked. For instance, if users who liked 'Fortnite' also liked 'Apex Legends,' it recommends 'Apex Legends' to you. Content-based is more focused on the item itself, while collaborative filtering uses the wisdom of the crowd. Both have strengths: content-based is good for niche interests, collaborative filtering can surprise you with new things.
4. What is one main environmental problem with blockchain?
Blockchain, especially proof-of-work like Bitcoin, uses a lot of electricity because many computers solve hard math problems to confirm transactions. This high energy use often comes from burning fossil fuels, which releases carbon dioxide and contributes to climate change. For example, one Bitcoin transaction can use as much electricity as a US household in a month. However, not all blockchains have this problem; some use proof-of-stake, which is much more energy efficient. NFTs, or non-fungible tokens, are digital ownership records on a blockchain, so their environmental impact depends on which blockchain they use.
5. What future trend besides deepfakes could cause AI-generated content crises?
Another trend is AI-generated text that can write fake news articles, reviews, or social media posts automatically. For example, bots could flood a platform with fake negative reviews about a product, causing a reputation crisis. AI can also generate realistic fake images of events that never happened. These can be used to spread misinformation at scale. Platforms will need to detect not just individual fakes but coordinated bot campaigns. The challenge is that AI-generated text is harder to spot than deepfakes because it looks like normal writing. Future crises may involve massive, automated disinformation attacks.
6. Compare using engagement metrics (likes, comments) with conversion metrics (sales, sign-ups) to measure social media success. Which is more important for a business?
Engagement metrics show how much people interact with content, but they don't always lead to business goals. Conversion metrics directly measure actions that bring value, like purchases or newsletter sign-ups. For a business, conversion metrics are more important because they show return on investment. However, engagement can lead to conversions over time. A good strategy balances both: high engagement builds brand awareness, while conversions drive revenue. For example, a post with many likes but no sales may need a clearer call-to-action. So, both are useful, but conversions are the ultimate measure of success.
7. Compare bot detection and troll detection in terms of features used.
Bot detection relies heavily on behavioral features like posting frequency, timing, and account age. Troll detection focuses more on content features like language toxicity, sentiment, and argumentative style. Bots often have uniform behavior (e.g., same interval between posts), while trolls have human-like patterns but toxic content. Both may use network features: bots often connect to other bots, trolls may target specific users. However, bots can be detected with less data (e.g., metadata), while trolls require analyzing the text. In practice, combining both approaches catches more malicious accounts.
8. Compare the content moderation challenges of global versus regional platforms.
Global platforms face challenges moderating content across many countries with different laws and cultures. For example, what is acceptable in one country may be illegal in another. They must balance free speech with local regulations. Regional platforms have fewer cultural differences to manage but may face pressure from their own government. For instance, a regional platform in China must follow strict censorship rules. Global platforms often have more resources for moderation but can be criticized for inconsistent enforcement. Both types struggle to remove harmful content while respecting user rights.
9. What does 'omnichannel social strategy' mean?
Omnichannel social strategy means a brand uses multiple social platforms like Instagram, Facebook, and Twitter together to give customers a smooth experience. The brand's message, look, and offers are the same across all platforms. For example, a customer might see a product on Instagram, then get a reminder on Facebook, and finally buy through a link in a tweet. This coordination makes the brand feel consistent and reliable. It also allows customers to interact on their preferred platform without missing information. The goal is to create a unified brand experience no matter where the customer engages.
10. What is dynamic ad insertion (DAI) technology in podcasting?
Dynamic ad insertion (DAI) is a technology that places different ads into the same podcast episode for different listeners. It works by having an ad server decide which ad to play based on data like the listener's location or interests. When you download or stream a podcast, the ad server inserts the chosen ad into the audio file in real time. This allows advertisers to target specific audiences without changing the main content. For example, a listener in India might hear an ad for a local brand, while someone in the US hears a different ad. DAI makes podcast advertising more relevant and effective.
11. Compare filter bubbles and echo chambers. How do they relate to algorithmic amplification?
Filter bubbles and echo chambers are both results of algorithmic amplification, but they are slightly different. A filter bubble is when an algorithm limits what information you see, based on your past behavior, so you miss opposing views. An echo chamber is a social environment where your beliefs are repeated and reinforced by others, often because you only interact with like-minded people. Algorithmic amplification can create filter bubbles by personalizing content, and these bubbles can lead to echo chambers when users only connect with similar others. Both reduce exposure to diverse perspectives.
12. How can a platform prepare for crisis management in an authoritarian regime?
A platform can hire local legal experts who understand the country's laws and censorship rules. It should have a clear policy for when to comply with government requests, such as removing content that is illegal locally. The platform can also create a crisis plan that includes steps for negotiating with authorities. Transparency reports can show how many requests were received, but in authoritarian regimes, even that may be risky. Building relationships with local officials can help. The platform must also train its global team to respect local laws while protecting user rights as much as possible.