Tag: algorithm

Categories
Mobile Syrup

Netflix refreshes UI to spotlight personalized categories and genres

Netflix is rolling out a new ‘Category Hub’ globally to offer users more personalized content recommendations.

This updated hub will be available starting April 21st in the left-hand menu on both adult and kids’ profiles. Here, you’ll see ‘Top 3’ categories based on what you normally watch. Curated collections centred around occasions like Earth Day or Internation Women’s Day will also be shown here.

However, Netflix didn’t mention how specific these categories will get. In the sample image provided in the news release, the streamer simply shows ‘Dramas,’ ‘Comedies’ and ‘Action’ in the Top 3, which are all fairly broad. Meanwhile, Netflix already has more defined categories elsewhere on the platform, including those for ‘Blockbuster Sci-Fi Movies,’ ‘Oscar-winning films,’ ‘Movies Based on Real Life’ and, as of this month, ‘Short-Ass Movies.’

In related news, Netflix confirmed earlier this week that it had its first quarterly loss of subscribers — 200,000 — in Q1 2022. The company attributed this to the suspension of its business in Russia amid the country’s invasion of Ukraine, although it estimates further losses of two million subscribers in the second quarter.

To help improve matters, the company has confirmed it’s looking into a paywall on password sharing and lower-cost, ad-supported tier, although it’s unclear when either may launch.

Source: Netflix

Categories
Mobile Syrup

Twitter’s algorithm favours right-leaning content in Canada, other countries

In a blog post, Twitter revealed that its algorithm promotes right-leaning content more often than left-leaning content. However, the company isn’t sure why that’s happening.

The findings discussed in the blog post draw from an internal study that looked at how Twitter’s algorithm amplifies political content. In the study, Twitter looked at millions of tweets posted between April 1st and August 15th, 2020 from news outlets and elected officials in Canada, France, Germany, Japan, Spain, the U.K. and the U.S.

In all those countries, except Germany, Twitter found that right-leaning accounts received more algorithmic amplification than the political left. Similarly, right-leaning content from news outlets benefitted from the same bias.

Twitter says it doesn’t know why the data suggests its algorithm favours right-leaning content. The company claims it’s a “significantly more difficult question to answer” because it’s a result of “interactions between people and the platform.”

However, The Verge cites Ph. D. candidate Steve Rathje, who published research explaining that divisive content about political outgroups is more likely to go viral. Rathje told The Verge that negative posts about political outgroups tend to receive more engagement on social platforms like Facebook and Twitter. For example, if a left-leaning politician posts something negative about a right-leaning politician (or vice versa), that negative post will likely receive more engagement.

With that in mind, it’s possible right-leaning posts on Twitter spark more engagement, leading to more algorithmic amplification. Further, it’s worth noting that Germany — the only country where the algorithm didn’t favour right-wing content — has an agreement with Facebook, Twitter and Google to remove hate speech within 24 hours. While the factors may not be related, if the algorithm favours divisive, negative posts, and right-leaning users post that kind of content, it could be why Twitter’s seeing the algorithm favour right-leaning content everywhere but Germany, where there’s an active effort to remove that content.

It’s not a problem isolated to Twitter either. Frances Haugen, the Facebook whistleblower who leaked internal documents from the company, claimed Facebook’s algorithm also favours divisive content and hate speech.

Source: Twitter, (2) Via: The Verge

Categories
Mobile Syrup

Study finds algorithm can use smartphone sensor data to detect cannabis use

Researchers from Rutgers University in New Jersey say they can use smartphone data and machine learning to detect cannabis intoxication.

The project started as a proof-of-concept way to passively detect cannabis use rather than rely on existing testing measures like blood, urine or saliva tests. The researchers published their findings in the Drug and Alcohol Dependence journal (via CTV News).

The study involved an experiment with 57 young adults who reported using cannabis at least twice a week. Researchers asked participants to complete three surveys a day over 30 days. The survey asked about how high participants felt at a given time, when they had last used cannabis and the quantity consumed. Participants reported a total of 451 episodes of cannabis use.

Additionally, researchers asked participants to download a smartphone app that analyzed GPS data, phone logs, accelerometer data and other smartphone sensor data and usage statistics.

The researchers found that when looking at the time of day, a machine learning algorithm could detect an episode of cannabis use with 60 percent accuracy. With just the smartphone data, the algorithm had an accuracy of 67 percent.

But with both time-of-day data and sensor data combined, the algorithm accurately predicted cannabis use with 90 percent accuracy.

The researchers said that GPS and accelerometer sensor data were the most important in detecting cannabis use — the study found that participants didn’t travel as far while high, while the accelerometer could be used to measure body movements.

While certainly interesting results, there could be potential concerns with applying the algorithm in real-world scenarios. For example, bias in the algorithm (unintentional or otherwise) could skew results. Another problem could be the accuracy — 90 percent is impressive, but if you fall in the 10 percent where the algorithm gets it wrong, that could cause problems.

The researchers say that this is the first study to examine how smartphone sensors could help detect cannabis intoxication. However, some of the researchers were involved in a similar 2018 study that investigated if smartphone data could be used to detect heavy drinking episodes.

Source: Drug and Alcohol Dependence Via: CTV News