3 moderation outsourcing options you need to know
If you’ve almost done a fesbuk, and thousands of users began to actively exchange photos, posts and comments, then over time you’ll find a picture of a long, strong male genital organ in the tape of little girl Mila. And you will understand that the rules for using the service are good, but you will still have to follow their implementation.
In this article we will talk about what solutions for content moderation exist today. During our market research, we found a considerable number of companies offering their services. We tried to classify them, to describe their advantages and disadvantages.
The whole market can be divided into 3 conditional groups:
Let's look at a little more detail all 3 groups.
Most of the market is made up of just such companies. Some of them specialize only in content moderation, but most often it includes call-center services, technical support, and much more.
')
Here are just some of them:
We believe that the number of such companies is due to the availability of cheap labor in the same Thailand, the Philippines, India, and so on. Hence, the low level of qualification of employees, and the overpriced price for moderating content in languages other than English.
And to use the services of such companies, you will need several things at once:
The second category of companies advanced further in studying the problem of content moderation and provides its own technical solutions, including:
Perhaps the brightest representative of this class is Crisp Thinking (http://crispthinking.com).
This category seems to be much more attractive: the staff is already specially trained and is familiar with the general problem of moderation; they work with their own software, and therefore are well acquainted with it; The API and SDK are documented and it is assumed that the client must cope with the integration itself.
However, not so good. Initial integration involves the creation of moderation rules. And here the problems begin.
Of course, after all the difficulties and the completion of the integration, the project completely forgets about moderation - the content automatically goes into the moderation system, the results also automatically come and are processed by the client application ... it remains only to replenish the balance with a tidy sum every month ...
In the course of market research, we found companies that fully automate the process of content moderation, that is, without the participation of people. One of the typical representatives is Keepcon (http://keepcon.com/).
Automatic moderation is certainly cheaper than manual, but has a large number of problems. Moderation without “manual” control - without checks by real people - is ineffective.
Firstly, it is extremely difficult to algorithmically compute a text that does not contain obscene words, but is of an unacceptable nature. For example, “A charming fairy will meet a kind gentleman in her apartment on the Shchukinskaya metro station” or hundreds of sophisticated phrases about selling drugs, such as “Hedgehog. Raccoon. Speed. Salt. SMS 8 901 *** ***** ”, for which you can quickly get acquainted with representatives of the rule of law. And now add to this support at least five languages. And ten!? ..
The situation with images is even more pitiable: the algorithms of face recognition and photo recognition of an erotic nature are imperfect today. Even in “laboratory” conditions, they give an unstable result, and in “field” conditions they have a very high percentage of errors and errors.
Secondly, in most cases we need to take into account the business logic of a specific project: for example, somewhere there should be only personal photographs of a specific user, with a clearly distinguishable face. The automatic moderation mode is failing here - the machines have not learned to take the subjective factor into account, only real people are capable of it.
Both our experience and the experience of other projects show that automatic moderation is only suitable as a preliminary filter that can weed out obviously low-quality content.
We have finished the technical description of the existing solutions. Now let's get to the fun part: compare prices!
We really wanted to provide you with not just dry numbers, but to show prices using a specific project as an example. As a result, we contacted most of the leading companies that are represented on the market, and all were asked the same task: you need to check 200,000 content units (photos, statuses, names) in 4 languages in a month.
Let's see what happened:
At the same time, all companies have a very clear approach to adding another language: “pay for each separately”. Thus, if instead of 4 languages you need 8, then the price tag for the month will increase by 2 times. As a result, 32 languages at ICUC will cost $ 0.9 million per month. Awesome!
During the study, we tried not only to classify existing companies. We tried to proceed precisely from the role of the model project generating UGC, the social mobile application. And our task was to find an easy, inexpensive way to moderate content, which would not take much time to integrate with existing code.
So, you have a project that is gaining an audience, do you want a ready-made solution that can be connected in a few days, and forget about the problem with checking the content, staff, developing the admin panel? Of all of the above, only Crisp Thinking seems to suit you, but prepare $ 25,000 per month.
Naturally, this article was written for a reason, and to draw attention to the issue of moderation and our content moderation service 2Hive.org .
Once we had our own need for content moderation, and 2Hive.org emerged as a way to solve it. First of all, we wanted to make the service very easy to use, and secondly, very budget. And we fulfilled both requirements. And how, read our future articles.
In this article we will talk about what solutions for content moderation exist today. During our market research, we found a considerable number of companies offering their services. We tried to classify them, to describe their advantages and disadvantages.
The whole market can be divided into 3 conditional groups:
- Staffing;
- Integrated into your SaaS project
- Moderation using algorithms (robots)
Let's look at a little more detail all 3 groups.
Type 1: Staff Provision
Most of the market is made up of just such companies. Some of them specialize only in content moderation, but most often it includes call-center services, technical support, and much more.
')
Here are just some of them:
- ICUC (http://www.icucmoderation.com)
- MicroSourcing (http://microsourcing.com)
- Moravia (http://moravia.com)
We believe that the number of such companies is due to the availability of cheap labor in the same Thailand, the Philippines, India, and so on. Hence, the low level of qualification of employees, and the overpriced price for moderating content in languages other than English.
And to use the services of such companies, you will need several things at once:
- Having your own software. In order for the staff hired by you to perform direct moderation, you will need to write an environment / program / web interface, where, apparently, there will be a list of unchecked content, and the moderators will press some buttons.
- Training. Yes, exactly: the companies themselves declare that before starting work you should hold a seminar and distribute professional instructions. And all this at your expense!
- Most companies of this kind require that you yourself draw up a schedule for the work of moderators, especially taking into account time zones and corresponding languages.
In simple terms, such companies are a large hangar on the outskirts of an Indian city, full of old computers and poor Indians, whose managers are ready to put staff anywhere, just to pay money. About the quality does not have to say.
Type2: Integrated into your SaaS project
The second category of companies advanced further in studying the problem of content moderation and provides its own technical solutions, including:
- API and SDK for integration
- Own moderation panel in which their own staff works
- Automatic content checkers
- Personal account for customers, where you can see reports, charts, analytics, etc.
Perhaps the brightest representative of this class is Crisp Thinking (http://crispthinking.com).
This category seems to be much more attractive: the staff is already specially trained and is familiar with the general problem of moderation; they work with their own software, and therefore are well acquainted with it; The API and SDK are documented and it is assumed that the client must cope with the integration itself.
However, not so good. Initial integration involves the creation of moderation rules. And here the problems begin.
For most companies, the API integration process is so complex that it can take several weeks!
Of course, after all the difficulties and the completion of the integration, the project completely forgets about moderation - the content automatically goes into the moderation system, the results also automatically come and are processed by the client application ... it remains only to replenish the balance with a tidy sum every month ...
Type3: Robots
In the course of market research, we found companies that fully automate the process of content moderation, that is, without the participation of people. One of the typical representatives is Keepcon (http://keepcon.com/).
Automatic moderation is certainly cheaper than manual, but has a large number of problems. Moderation without “manual” control - without checks by real people - is ineffective.
Firstly, it is extremely difficult to algorithmically compute a text that does not contain obscene words, but is of an unacceptable nature. For example, “A charming fairy will meet a kind gentleman in her apartment on the Shchukinskaya metro station” or hundreds of sophisticated phrases about selling drugs, such as “Hedgehog. Raccoon. Speed. Salt. SMS 8 901 *** ***** ”, for which you can quickly get acquainted with representatives of the rule of law. And now add to this support at least five languages. And ten!? ..
The situation with images is even more pitiable: the algorithms of face recognition and photo recognition of an erotic nature are imperfect today. Even in “laboratory” conditions, they give an unstable result, and in “field” conditions they have a very high percentage of errors and errors.
Secondly, in most cases we need to take into account the business logic of a specific project: for example, somewhere there should be only personal photographs of a specific user, with a clearly distinguishable face. The automatic moderation mode is failing here - the machines have not learned to take the subjective factor into account, only real people are capable of it.
Both our experience and the experience of other projects show that automatic moderation is only suitable as a preliminary filter that can weed out obviously low-quality content.
Prices
We have finished the technical description of the existing solutions. Now let's get to the fun part: compare prices!
Prices shocked. For the crowd of Indians who need to be trained at their own expense, they asked for up to $ 111,600 per month!
We really wanted to provide you with not just dry numbers, but to show prices using a specific project as an example. As a result, we contacted most of the leading companies that are represented on the market, and all were asked the same task: you need to check 200,000 content units (photos, statuses, names) in 4 languages in a month.
Let's see what happened:
- HTC
- Price: $ 10,000 per month
- ICUC
- Price: $ 111,600 per month
- One of Setup fee: $ 7,500
- Source fit
- $ 9600 ~ $ 15,600 per month depending on the qualifications of the moderators
- Microsourcing
- $ 15,000 ~ $ 20,000 per month depending on the qualifications of the moderators
- Moravia
- Price: $ 54,000 per month
- Crisp thinking
- Price: $ 25,000 per month
- One of Setup fee: $ 10,000
At the same time, all companies have a very clear approach to adding another language: “pay for each separately”. Thus, if instead of 4 languages you need 8, then the price tag for the month will increase by 2 times. As a result, 32 languages at ICUC will cost $ 0.9 million per month. Awesome!
Conclusion
During the study, we tried not only to classify existing companies. We tried to proceed precisely from the role of the model project generating UGC, the social mobile application. And our task was to find an easy, inexpensive way to moderate content, which would not take much time to integrate with existing code.
So, you have a project that is gaining an audience, do you want a ready-made solution that can be connected in a few days, and forget about the problem with checking the content, staff, developing the admin panel? Of all of the above, only Crisp Thinking seems to suit you, but prepare $ 25,000 per month.
Naturally, this article was written for a reason, and to draw attention to the issue of moderation and our content moderation service 2Hive.org .
2Hive.org can deal with a voiced problem
for $ 4000 per month, in 32 languages.
Once we had our own need for content moderation, and 2Hive.org emerged as a way to solve it. First of all, we wanted to make the service very easy to use, and secondly, very budget. And we fulfilled both requirements. And how, read our future articles.
Source: https://habr.com/ru/post/241768/
All Articles