by Bernard Marr
What Were The Results?
Spotify's Discover Weekly playlists mean that it is able to recommend new music that its users will love, and in return they are likely to remain as subscribers to the service.
Its success at predicting new music that users will love has been cited as a driving factor behind its success, with its subscriber base growing by 8 million, and share price rising by 25% in the three months following its listing on the New York Stock Exchange in April 2018.8
Key Challenges, Learning Points And Takeaways
Big streaming services like Spotify have access to so much data that they can make highly accurate predictions, even about very personal and human issues such as our taste in music.
Individual elements in a music track, such as the tempo, beat and content of lyrics, are good indicators that can be used to match it to listeners who will enjoy it.
Amalgamating the results of analysis of several different datasets – user behavior, song data and external text data – allows Spotify's deep learning systems to produce increasingly accurate predictions.
Presenting predictions to users in a way they will understand and feel comfortable with is often as important a factor as the predictions themselves – Spotify chose the Discover Weekly format because music is increasingly consumed through playlists.
Notes
1Spotify, Spotify Technology S.A. Announces Financial Results for Second Quarter 2018: https://investors.spotify.com/financials/press-release- details/2018/Spotify-Technology-SA-Announces-Financial-Results-for- Second-Quarter-2018/default.aspx
2YouTube, Vidhya Murali and Ching-wei Chen on predicting music: https://www.youtube.com/watch?time_continue=166&v=n5gCQWLX Jcw
3HPAC, Music Recommendation System Spotify: http://hpac.rwth- aachen.de/teaching/sem-mus-17/Reports/Madathil.pdf
4Music:Ally, Spotify talks playlists, skip rates and NF's Nordic-fuelled success (#SlushMusic): https://musically.com/2017/11/29/spotify-playlists-skip-rates-nf/
5Music Business Journal, Spotify's Secret Weapon: http://www.thembj. org/2014/10/spotifys-secret-weapon/
6Quartz, The Magic That Makes Spotify's Discover Weekly Playlists So Damn Good: https://qz.com/571007/the-magic-that-makes-spotifys-discover-weekly-playlists-so-damn-good/
7Computer World, How Spotify migrated everything from on-premise to Google Cloud Platform: https://www.computerworlduk.com/cloud-computing/how-spotify-migrated-everything-from-on-premise-google-cloud-platform-3681529/
8Financial Times, Spotify gains 8m paid subscribers aided by Latin America growth: https://www.ft.com/content/16c0c91c-90cd-11e8-bb8f-a6a2 f7bca546
27
Telefonica: Using Artificial Intelligence To Connect The Unconnected
Telefonica is a Spanish multinational telecom company, which is one of the largest telephone operators, broadband providers and mobile network providers in the world. In the United Kingdom it is known as O2 since the parent company acquired the brand, a spin-off from British Telecom, in 2006.
In 2018, it announced ambitious plans to connect up to 100 million inhabitants of some of the most remote regions of South America. It will do this by using artificial intelligence (AI) to locate communities underserved by telecoms infrastructure, and to allocate resources to enable them to be brought online.
What Problem Is Artificial Intelligence Trying To Solve?
Advanced technologies can do a lot to change people's lives for the better. Enhanced connectivity broadens opportunities and horizons for business and education, and allows vital services such as utilities and transport infrastructure to be planned and administered efficiently.
However, with more than half of the world's population still unable to access the internet,1 vast swathes of the world are still unable to take advantage of these opportunities.
The problem is caused by the fact that while the cost of distributing online and connectivity infrastructure has fallen dramatically in urban areas, where there are large numbers of customers willing to pay for the services concentrated in relatively small geographic areas, in remote rural regions the story is very different.
Without high concentrations of people, the cost of rolling out connectivity infrastructure can be prohibitive. The problem is exacerbated by the fact that populations in remote regions are often difficult to track, and data on their movements and locations is limited, even today.
How Is Artificial Intelligence Used In Practice?
Telefonica's Internet Para Todos (Internet For All) project involves using AI to tackle the problem of rolling out online connectivity solutions to 100 million people living in remote regions of Latin America.
Latin America was chosen because 20% of the region's population still lack access to mobile broadband services, which can often play an essential part in social and economic development.2
By first using computer vision technology to study satellite images and understand where people are living, it was able to draw up plans to overcome the logistical difficulties inherent in connecting these isolated populations.
It was then able to analyze transport networks in the regions and use the data to optimize logistics of rolling out network coverage to reach as many people as possible. As remote rural areas are often poorly served by transport links such as roads and railways, deploying the necessary equipment to bring people online is generally the most expensive part of an operation to bring an area online.3
By comparing this information with data from its own network coverage, it was able to see which parts of the region were most in need of coverage, and where infrastructure could be rolled out with the greatest efficiency.
What Technology, Tools And Data Were Used?
Telefonica has partnered with Facebook for the Internet Para Todo project, which initially used machine learning analysis of high-definition satellite imagery to create maps showing where people are living.
It also uses satellite data to understand transport infrastructure links, as well as the Telefonica network's own data on the location of its transmitters and towers, and local census data.
The project incorporates plans for predictive maintenance – essential when it could take engineers days to reach locations where equipment has developed faults.4
Cooperation with local infrastructure operators, community organizers and entrepreneurs is an essential part of the scheme. Machine learning is used to process all of the data on the resources that are available and suggest solutions that have the potential to connect the largest number of people.
Together with Facebook, Telefonica is assessing technologies such as microwave and radio access network solutions, including Facebook's OpenCellular wireless connection platform, specifically engineered for bringing connectivity to isolated rural communities.
OpenCellular uses radio waves to transmit mobile broadband signals and is designed to take advantage of existing infrastructure such as towers, which may already be in place, dramatically reducing the cost of deployment.5
What Were The Results?
The machine learning and computer vision component of the program was able to map 95% of the population of the remote areas that were analyzed, with a less than 3% rate of false positives.6
A pilot scheme in Peru has already seen 10,000 residents of the Amazon basin connected to the internet.
Eventually, it is planned that up to 100 million could benefit. Connecting them to the internet will improve the economic outlook of entire communities, as well as give them access to modern healthcare and education resources for the first time.
This should lead to huge improvements in the quality of life of people who until now have not benefited from the technological advances that have reshaped much of the developed world.
Key Challenges, Learning Points And Takeaways
Isolated populations have not benefited from the breakthroughs in communication technology enabled by the internet across the developed world.
AI makes it possible to map population densities using satelli
te images to give more accurate data on where people are living.
Analysis of transport infrastructure enables technology to be deployed in the most cost-effective way, making it viable to connect previously unreachable populations to the internet.
Predictive maintenance makes it possible to understand when and how things are likely to break or require servicing, meaning repairs can be scheduled efficiently – this is essential when maintaining networks spread over large, sparsely inhabited regions.
Notes
1ITU, ICT Facts and Figures 2016: https://www.itu.int/en/mediacentre/ Pages/2016-PR30.aspx
2Computer Weekly, MWC 2018: Telefónica aims to connect 100 million in Latin America: https://www.computerweekly.com/news/252435708/ MWC-2018-Telefonica-aims-to-connect-100-million-in-Latin-America
3LUCA, Ready For A Wild World: https://www.slideshare.net/wap13/big- data-for-social-good-106562070
4Fierce Telecom, Telefónica's “Internet para Todos” project uses modern tools to find and connect Latin Americans: https://www.fiercetelecom. com/telecom/telefonica-s-internet-for-all-project-uses-modern-tools-to -find-and-connect-latin-americans
5TechCrunch, Facebook's OpenCellular is a new open-source wireless access platform for remote areas: https://techcrunch.com/2016/07/06/ facebooks-opencellular-is-a-new-open-source-wireless-access-platform- for-remote-areas/
6Telefonica, How Telefónica uses artificial intelligence and machine learning to connect the unconnected: https://www.telefonica.com/en/ web/public-policy/blog/article/-/blogs/how-telefonica-uses-artificial- intelligence-and-machine-learning-to-connect-the-unconnected
28
Twitter: Using Artificial Intelligence To Fight Fake News And Spambots
There are over 330 million Twitter users using the social media platform to send hundreds of millions of tweets every day.1
People all around the world love the service for the ease it brings to staying in touch with friends, their favorite celebrities and keeping up to date with the news.
Unfortunately, due to the huge number of people using the service and its essentially anonymous nature, sometimes the news being pumped out 24/7 is fake. And sometimes the other people using the service may not have your best interests at heart.
One of the ways the social media giant uses artificial intelligence (AI) is to try to stay on top of the enormous challenge of keeping its users safe from those using it to spread harmful information.
What Problem Is Artificial Intelligence Helping To Solve?
The age of social media has given everyone a voice. And, as always, some people choose to use theirs to spread lies and misinformation.
Whether it's for political reasons or greed, since its birth social media has been a magnet for all types of scammers and propagandists, up to and including accusations of state-level electoral interference.2
While trolls targeting foreign elections may have made the headlines this year, more personal scamming is also rampant and often goes under the radar. A Gizmodo investigation found that fraudsters routinely steal photographs from innocent third parties to create fake accounts.3
Part of it is the inherent ease with which anybody can present themselves as anything. Just pick a name and an avatar and you have a safe shroud of anonymity from which you can spread anything from ponzi schemes to conspiracy theories and terrorist propaganda. An ongoing study by the Knight Foundation illustrates the problem as it identifies millions of fake news tweets being spread on Twitter.4 Like most social networks, Twitter is keen to tackle this problem.
How Is Artificial Intelligence Used In Practice?
Since the rise in public awareness of the seriousness of fake news, Twitter has begun to take a more proactive stance towards identifying and removing offending accounts from its service.
Part of its strategy is to develop machine learning tools that can identify the networks of spambot accounts that fake news peddlers and scammers use to give their voice the illusion of legitimacy.5
This allows it to identify and shut down close to 10 million accounts every week, without having to wait until the accounts are reported by users.
It works by identifying patterns in an account's behavior – for example, linking to known fake news sites – and matching them with patterns displayed by known fake or bot accounts that were identified in the past.
Once an account has been highlighted as a possible offender, it is put into a read-only state, so its owner won't be able to use it to post.
Then Twitter asks the owner of the account to verify themselves as an actual human – with a phone number or legitimate email address.
Because fake news, conspiracy and scam networks operate by using hundreds or thousands of fake accounts to amplify their message, it often isn't feasible for the human operating the network to do this.
What Technology, Tools And Data Are Used?
Twitter has said that it does not want to publicly discuss the signs it uses to detect if an account is fake to hinder those who would try to develop workarounds.6
However, it is most likely that Twitter looks for accounts displaying patterns of activity that correlate with fake accounts identified in the past.
This is likely to include posting frequency, networking behavior (who the account follows and unfollows), large numbers of accounts appearing to originate from a limited number of IP addresses and the use of technology such as VPN to obscure identity and geographical location.
When accounts following particular patterns around those activities also seem to be consistently sharing content from websites identified as untrustworthy or dishonest, there's a higher probability that they could be fake accounts.
Twitter's involvement with AI technology is certainly not limited to dealing with fake accounts. The platform also uses deep learning to decide how interesting particular tweets will be to individual users, and the prominence they should receive in their timeline.7
It does this by analyzing every individual tweet from the accounts that a user follows, and assessing it based on how popular it is, the user's prior interactions with the author and how well it matches features of other tweets you have interacted with in the past.
Many of Twitter's initiatives are directed by Cortex, its in-house specialist AI team.
What Were The Results?
In two months, Twitter used automated detection tools to take down more than 70 million “fake and suspicious” accounts.8
Year on year, 214% more accounts have been removed for violating spam policies. At the same time, reports from users that they have encountered spam dropped from 25,000 per day in March 2018 to 17,000 per day in May 2018. Twitter points to this as evidence that its proactive policy is removing spam and fake accounts from its platform before users even see them.9
As well as social good, the initiative by Twitter to drop fake users from its system serves a business purpose. Advertisers want to know the ads they pay to show on Twitter are being seen by real people, not bots.
Key Challenges, Learning Points And Takeaways
Scammers and those with malicious intent can be identified by their behavior online using AI, employing very similar techniques to those used by marketers to decide who to target ads to.
Sometimes anonymity is important. Because Twitter understands that occasionally asking for accounts to identify themselves may restrict free speech in ways that could be dangerous, it operates a trust and safety council.10
Twitter realizes that the freedom of speech its platform offers is important, but has put its belief that the safety of users is paramount at the top of its agenda.
Notes
1Twitter, How policy changes work: https://blog.twitter.com/official/ en_us/topics/company/2017/HowPolicyChangesWork.html
2Financial Times, Senate panel backs finding of Russian meddling in US election: https://www.ft.com/content/04385510-7f13-11e8-8e67-1e1a0846c475
3Gizmodo, The Bizarre Scheme Using Viral Abuse Stories and Stolen Pics to Sell Diet Pills on Twit
ter: https://gizmodo.com/the-bizarre-scheme-using-viral-abuse-stories-and-stolen-1829173964
4Knight Foundation, Disinformation, “Fake News” and Influence Campaigns on Twitter: https://www.knightfoundation.org/reports/ disinformation-fake-news-and-influence-campaigns-on-twitter
5Twitter, How Twitter is Fighting Spam and Malicious Automation: https://blog.twitter.com/official/en_us/topics/company/2018/how- twitter-is-fighting-spam-and-malicious-automation.html
6Twitter, Our approach to bots and misinformation: https://blog.twitter. com/official/en_us/topics/company/2017/Our-Approach-Bots- Misinformation.html
7Twitter, Using Deep Learning at Scale in Twitter's Timelines: https:// blog.twitter.com/engineering/en_us/topics/insights/2017/using-deep- learning-at-scale-in-twitters-timelines.html
8Washington Post, Twitter is sweeping out fake accounts like never before, putting user growth at risk: https://www.washingtonpost.com/ technology/2018/07/06/twitter-is-sweeping-out-fake-accounts-like- never-before-putting-user-growth-risk
9Twitter, How Twitter is fighting spam and malicious automation: https:// blog.twitter.com/official/en_us/topics/company/2018/how-twitter-is- fighting-spam-and-malicious-automation.html
10Twitter, Announcing the Twitter Trust & Safety Council: https://blog. twitter.com/official/en_us/a/2016/announcing-the-twitter-trust-safety- council.html