As the new season of Premier League football gets under way, a few things are certain. There will be goals, drama and excitement, and unfortunately, players will be subjected to vile abuse on social media.
My colleagues and I at the Alan Turing Institute have published a report, commissioned by Ofcom, in which we found that seven out of ten Premier League footballers face abuse on Twitter. One in 14 is raped every day.
These are grim statistics, with huge implications for player welfare. Other analyzes have revealed a high rate of online abuse, particularly racist abuse, of footballers that has gone largely unchallenged by football’s governing bodies. Mental health is increasingly a concern in football and there is a lot of evidence that online abuse can lead to a range of mental health problems, from depression to suicidal thoughts.
Our report is one of the first to use artificial intelligence (AI) to systematically detect and track online abuse against footballers at scale. This is nearly impossible to do manually due to the sheer size and complexity of social media.
We focused our analysis on Twitter because it is widely used by soccer players and fans and makes its data freely available to researchers. In total, we collected 2.3 million tweets that mentioned or directly replied to the posts of 618 Premier League footballers during the first half of the 2021-22 season.
At the heart of our analysis is a new machine learning model developed by the Turing Internet Security team as part of our Online Harms Observatory. This model is able to automatically assess whether a tweet is abusive or not by analyzing its language.
To provide a benchmark for our AI model and a more in-depth overview of tweet content, we also manually labeled 3,000 tweets, categorizing them as positive, neutral, critical, or abusive. Critical tweets were those that criticized a player’s actions on or off the field, but not in a way that could be considered abusive.
We recognize that categorizing tweets in this way is somewhat subjective, but we tried to reduce human bias as much as possible by consistently applying the same definitions and guidelines to all tweets.
What did we find?
Of the 3,000 tweets we tagged, the majority (57%) were positive. Tweets routinely expressed admiration, praise and support for players, often using emojis, exclamation marks and other indicators of strong positive emotion. A smaller proportion of tweets were labeled as critical (12.5%), neutral (27%) or abusive (3.5%).
Our machine learning model, applied to all 2.3 million tweets, found that 2.6% contained abuse. This may seem like a low percentage, but it represents almost 60,000 abusive tweets over just five months.
Abuse is widespread: 68% of players received at least one abusive tweet during this period. But players have very different experiences online: just 12 players received half of all abuse. Cristiano Ronaldo, Harry Maguire and Marcus Rashford received the most abusive tweets.
Abuse also varied greatly throughout the season, with major peaks after major events. For example, the number of abusive tweets increased on 27 August 2021, when Manchester United re-signed Cristiano Ronaldo, and on 7 November 2021, when Harry Maguire tweeted an apology after Manchester United lost to Manchester City.
We found that about 8.5% of abusive tweets (0.2% of all tweets) attacked players’ identities by referring to a protected characteristic such as religion, race, gender or sexuality. This is a surprisingly low percentage given concerns about racial abuse of footballers online. But we only looked at identity attacks using keywords (while we had a full AI solution for identifying abuse) and didn’t specifically look at the experiences of non-white players.
Read more: Euro 2020: can Twitter stop racist abuse before it happens?
Being a good fan online
Addressing abuse online is no easy task – finding and categorizing abuse is technically difficult and raises fundamental questions about free speech and privacy. But we cannot let abuse go unchallenged.
Some social media platforms, including Twitter, are already taking steps to improve their trust and security processes, but more can be done. This may include amplifying and promoting content that is not abusive; providing additional practical support and advice to players (and others) who are receiving large amounts of abuse; and making more use of properly tuned machine learning tools to automatically detect and take action against abuse. Ultimately, platforms should bear most of the responsibility for cleaning up their services.
Prostock-studio / Shutterstock
In general, abusive content is underreported. Ofcom surveyed the public about their experiences of players being targeted online, finding that more than a quarter of teenagers and adults who go online (27%) saw abuse directed at a footballer last season. Among those who encountered abuse, more than half (51%) said they found the content extremely offensive, and about 30% took no action in response.
There is absolutely nothing wrong with being emotional about football and expressing how you feel online, but we all need to be careful not to cross the line into being abusive and intimidating. And if you see someone else abusing, be proactive. Report it and make it clear that this content has no place in football (or anywhere else). Football is a beautiful game and we can all help keep it that way.
