On 1.3.2018 I gave a lecture on the topic of data science and SEO at the SEO Campixx, and since there were some discussions afterwards :-), I will describe the contents in more detail here, in several parts. This part is first of all about what data science is and what already exists on the topic.
What exactly is data science?
“The sexiest job of the 21st century” is rather dull on closer inspection, because most of the time is spent acquiring and cleaning data and using it to build models. It’s coding, it’s math, it’s statistics, and with larger amounts of data, it’s also a lot of knowledge about where to wire which instances with each other, such as on Amazon Web Services or Google Cloud Platform. A global galactic definition of data science does not exist to my knowledge, but I would consider data science to be the intersection of
- Data Mining
- Statistics and
- Machine Learning
define. These are not new topics, but what is new is that we have much more data, much faster processors, cheap cloud processing and many development libraries. For the statistics language and development environment R used here, libraries exist for almost every purpose; somewhere there was someone who was faced with the same problem and then built a solution for it. What is also new is that more and more companies feel that you can do something with data, after all, Spotify uses data to know what music you might still like, and Google knows when you should set off if you want to get to work on time.
Unfortunately, the data hype (which will be followed by a healthy understanding of what is possible after a plateau of disappointment) is countered by relatively few people who feel at home in all three disciplines (plus cloud computing). Which in turn leads to the fact that these data scientist unicorns are sometimes offered unreasonable sums and 1000s of courses are offered on Udemy & Co that are supposed to provide you with the necessary knowledge.
A real problem with data science, however, is that not only knowledge in several areas is necessary, but also the understanding that data is used to solve a problem. I can deal with algorithms and data all day long, for me it’s like a kind of meditation and relaxation. In fact, sometimes I feel like playing with Lego, but at the end of the day, it’s all about solving problems. Not only collecting data, but also extracting the right information from it and then taking the right action (the holy trinity of data). And here is the challenge that often enough people just say, here is data, make something out of it. Therefore, it is an art for the data scientist to understand exactly what the problem actually is and to translate this into code.
In addition, many people have bad memories of math. Accordingly, the audience’s willingness to consume slides with lots of numbers and formulas tends to be on the lower end of the scale. That’s why I also worked with smaller examples in the lecture, which everyone should be able to understand well.
What kind of topics am I working on? Very different. Classification. Clustering. Personalization. Chatbots. But also analyses of slightly larger data volumes of 25 million rows of Analytics data and more, which have to be processed in a few minutes. All kinds of things.
What is already there?
On the search engine side, there is already a lot. When I was still at Ask, we had already worked with Support Vector Machines, for example, to create the ranking for the queries where the pages had almost no backlinks. Even then, there was a dynamic ranking. The topic recognition of most search engines is based on machine learning. RankBrain will be based on machine learning. So it’s not a new topic for search engines.
On the other hand, that of SEOs, the topic still seems to be relatively fresh. Search Engine Land says thatevery search marketer can think of themselves as a data scientist. I’m not sure I would subscribe to that, because most search marketers I know don’t build their own models. As a rule, they use tools that do this for them. On SEMRush you can find a collection of ideas, but more for SEA. Remi Bacha is still exciting, although I haven’t seen any data from him yet. Keyword Hero have come up with something pretty cool by using deep learning to identify the organic keywords that are no longer included since the switch to https. Otherwise, I haven’t seen much on the subject. So we see that we are at the very beginning.
What would we like to have?
Back to the question of what problem I actually want to solve with my work. In an ideal world, SEO naturally wants to be able to re-engineer the Google algorithm. However, this is unlikely, because of the more than 200 ranking signals, only a few are available to us. What we can do, however, is try to build models with the signals we have, and possibly create smaller tools. And that’s exactly what the next part is about