More transparency and openness is something we have written on the LOVOO flag and communicated previously. On these grounds, we want to publish a quarterly ‘Fake and Spam Transparency Report’ to provide some insight into our internal numbers related to our daily fight against spam. The first report mainly deals with basic terms and processes at LOVOO.
How does a user become a spammer at LOVOO?
This is one of the first questions we have and want to answer. Users can turn into spammers in many different ways:
- Other users report someone. When a user is frequently reported, his profile is checked automatically and the user will be blocked (i.e. marked as a spammer). He/she can no longer use the app and will no longer be visible to other users.
- A second possibility is a block by the anti-spam system. This system automatically detects that a user is a spammer and blocks the account instantly.
- Furthermore, our support works hard to react to reports as quickly as possible. This also ensures that spammers get blocked fairly early on – even before the critical amount of reports is reached.
How many of our users are actually spammers?
How many accounts have been identified as spammers, either by user reports, support or the anti-spam system? The following graph shows the number of spammers in comparison to daily active users (DAU) in the period between July and October 2016:
The green area shows all “real” users, the red area all those that have been marked as spammers on the corresponding day. The graph therefore illustrates that spammers only make up a very small percentage of our user base. In numbers, it is on average a percentage of only 0.3% of users active on a daily basis.
In the detailed view of a single month (for example, August 2016), it becomes clear that organized spammers often attack in waves. For instance, there has been higher spam activity in the second week of August. It might also be due to the release of a new version of our Anti-Spam system. It could be, for example, that a software update enables us to identify new types of spam and use this knowledge to label users active in the past.
Whoever is familiar with the LOVOO app, might wonder whether the interactions of spammers are more significant than their overall impact. Even when the number of spammers is relatively small in comparison to the total number of users, their disproportionately high amount of actions can have a significantly negative impact on other users.
For this reason, we also looked at the relation between votes caused by spammers in comparison to the overall daily votes:
The graph clearly shows that spammers are not always active right away. Their activity is at the beginning of the third quarter, for example, much higher than at the end. What is more, the impact spammers have on daily votes is higher than their share of daily active users. That was, however, to be expected. Spammers want to reach as many users as possible and the most obvious method for this is using votes in the Match. Their share of likes is on average at 2.89%. Our goal is of course set at 0% – and we are on the right track.
Who detects more spam: users or the anti-spam system?
Spammers can be reported by users or identified automatically by the anti-spam system. Our goal is to block as many spammers as possible automatically, before the critical number of reports is reached or the support staff have to intervene manually. In the third quarter of 2016 we succeeded in blocking 75.62% of all spammers automatically using the anti-spam system. The remaining 24.38% reported by users was also helpful.
These reports help us to continuously improve our system. If there is a new type of spam or a completely new behavior among spammers, we take this new information and include it in our system. This helps us to recognize these type of spammers in future and also retroactively blocks all profiles which have used this behavior in the past.
Does report mean block?
A question often asked is why a user is not blocked immediately after the first report. The answer is very simple: users report other users for a variety of reasons. Not all of these reasons justify blocking the reported user.
The huge number of reports alone is significant: in the third quarter of 2016, there was a total of 1,051,560 reports on 819,710 different user profiles. An astounding 82.19% of these reported profiles, however, were not spammers. The sensation of spam or fake is subjective and often not based on facts. A learning that is taken by the product team, to improve the verification process.
How fast is the anti-spam system?
We have to be very careful when blocking users automatically. On the one hand, we want to catch as many spammers as possible, but we must not be too quick to judge. Otherwise, we risk blocking real users who only briefly exhibit spam-like behavior. This can happen, for example, if Match is played very fast. The anti-spam system therefore waits until a user has exhibited negative behavior on several occasions. On average it takes 2.2 hours from the time of the first incident, e.g. a vote in Match, for the block to be executed by the anti-spam system. We are continuously working on shortening this time as much as possible.
- Some only give out likes very slowly over a long period of time.
- Some give out likes in small waves, e.g. 100 per day within a few minutes.
- Others give out a large number of likes continuously.
A consistently high amount of likes is identified very quickly and the corresponding user is blocked within a few seconds. The first two types take a bit longer to be recognized, sometimes several days, because the users are only active for very short amounts of time. These profiles need to exhibit the negative behavior in several instances before the anti-spam system can justify blocking them.
Are spammers more active on male or female accounts?
Spammers mainly use female profiles, in 83.90% of all cases to be exact. This approach seems to have proven to be more promising for them. The average age of spammer profiles is surprising: male profiles and female profiles have an average age of 29 years and this is relatively high. A reason for this might be that profile information is not really important for spammers. They themselves do not want to get likes, but only create events for other users.
The top 10 countries of spammer origin*
At the end of our analysis, we uncovered which countries the spammers, who register on the platform, are mainly from:
These figures are, however, only of partial of importance: the attackers often use co-called ‘bot networks’. Using this method, the spammers use a large number of hacked computers all over the world to orchestrate their attacks. This means that the attackers seemingly come from ‘everywhere’ and cannot be accurately located.
A glance at the backend
When users at LOVOO interact with each other, they communicate via our backend servers. When a user sends another user a ‘Like’ in Match, the backend servers make sure that the liked user also receives that ‘Like’. In addition, the back end sends this information as anonymized data to the anti-spam system as well as to our database, where it is stored. The anti-spam system evaluates all user interactions and blocks users, if they abuse the Match for spamming. All information from our database is used in further analysis to ensure that we can stop spam as efficiently as possible.
What is the process for combatting spam at LOVOO?
The fight against spammers is basically a never-ending story: every time you think you have blocked their latest method, spammers find a new way to create havoc. This often occurs with the imitation of normal user behavior, which makes them even harder to detect.
Success (as regards detected spammers) and failure (as regards user complaints concerning profiles that have not been blocked) often go hand in hand. We take on the challenge every day and are really efficient at detecting new types of spam as a faster rate.
This report shall constitute a public display of our progress, as well as our setbacks. We take spam seriously at LOVOO and tackle it head-on instead of simply accepting it. You can read more on this in our next ‘Fake and Spam Transparency Report’.
* based on the IP address when installing the app