I am responsible for data management at LinkedIn, working on SEO and customer experience. Last month, I introduced to SMX London how to apply data science to SEO. The session covered topics such as metrics, A / B testing, SEO vs. SEM cannibalization tests, and machine learning for content quality. Here are some questions from the session participants with my answers:
Do you use specific tools / processes for A / B testing?
We have an internal infrastructure that supports usability A / B tests the configuration and performs an automatic statistical analysis for key indicators. If you're interested, you can read this article [pdf] about how we do experiments on LinkedIn. If you do not have the internal tools available, you can randomize your target set of URLs and compare your metrics from two URL groups using open source statistical test solutions such as R, the Scipy Python package, etc.
How do you sample the A / B referencing tests?
At the LinkedIn level, where we often have more than hundreds of thousands of URLs in each experiment, we simply randomized the URLs into two groups and compared their metric impact. However, when we begin the experiment, we are gradually rolling out the experimental functionality, which goes from a small percentage to 50%, to minimize the risks associated with the experiment.
In terms of duration, it depends on the type of experience and the type of experience. the type of product we experience. But in general, we try to run it at least a month in order to allow enough time for the search engines to analyze the new changes and reflect them in the search rankings.
What were the results of the SEO vs. SEM cannibalization test?
For the keywords we chose for this specific test, we found no impact on organic search traffic. So we were able to move to the SEM campaign. However, this learning would not apply in all cases because it depends on the keywords and pages you target. I recommend you to do an experiment before your SEM campaign if cannibalization concerns you.
How did you use quality of content for SEO recommendations? Do you oblige users to add photos or other product recommendations or SEO actions?
Rather than encourage users to improve the content of their profile or other pages, we plan to use this score in the cases of SEO use, such as the directory, cross links, etc. We would like to present better quality pages to search engines and searchers using the score.
How do you enter this field (data science and SEO)?
When I was at university with a major in Industrial Engineering, I attended a machine learning course and I immediately hooked on it. The idea of finding models and ideas from the data has fascinated me. In terms of application to SEO, it is only at LinkedIn that I started to work in the field of SEO and learn more. Even after working in SEO for three to five years, I think there is still a lot of exciting work in the field of data science in this area. It has been a great pleasure for me to share some of my work at SMX London this time and I look forward to seeing more to come!
The opinions expressed in this article are those of the invited author and not necessarily those of the search engine. Earth. Associated authors are listed here.
About the Author
Eun-Ji Noh is an experienced scientist working in the field of search engine optimization and product data science customer experience at LinkedIn. She informs product decisions to help people discover LinkedIn through analysis and understanding of the data.