Meet "Meet the Team"
June 17, 2016
Meet the Team: A New Addition to the Job Detail Page
As one of the most viewed pages on the LinkedIn website, the job detail page is very important for job seekers—it has information from the job poster and company, and it's also the landing page when a member clicks through the “Jobs You May Be Interested In” email. Our team’s responsibility is to help this page provide the best insights for job seekers so that they can better achieve their career goals. For this reason, we built a new feature on the job detail page called Meet the Team, which allows members to see basic profile information about current employees at companies they’re researching.
New "Meet the Team" Detail View
To get an idea of how Meet the Team works, check out the flow diagram below.
How we build offline flow to replace online flow
Finding the “Team”
For any job posted on the website, we want to find people who already have this position at the same company. However, there are a lot of different ways to describe a job title, so it's easier said than done. For example, a member who is a “software engineer” could write “software eng,” “software engineer I,” or “software engineer, talent solution” in his profile. To solve this problem, we used a title standardization library to standardize both the job description and people’s titles. Based on the standardized result, we randomly select and display up to four members with the same title who are also based in the same geographic location as the job posting to the Meet the Team page.
Scaling Meet the Team
Initially, we were using LinkedIn’s search function to find people to match for Meet the Team, but there are two downsides of using search. First, search is very expensive and it does not fit well with our aggressive performance goal because it is not scalable and may take longer than expected. Second, our high number of pageviews on the job details page create a lot of extra traffic on downstream search service. Given these two limitations, we moved to an offline workflow, where we created a Hadoop job to scan all members and store the calculated result in a Voldemort store with the key as (company, title, location). This offline job runs every night and pushes the results to Voldemort, our distributed key-value storage system. When the job detail page is loaded, it gets the key from the job: (company, title, location) and uses it to find corresponding members.
Member security and confidentiality is one of our most important considerations when we roll out new features at LinkedIn. After consulting with the LinkedIn Security team, we identified several key areas where we needed to ensure that member privacy would be maintained. Because the Meet the Team feature involves both members and companies, we needed to make sure members could choose to not appear in the Meet the Team module. If a member’s profile has a private section, it should remain private on the Meet the Team page. In addition, we allow company administrators to turn off this feature for all jobs for their company.
When we first launched the Meet the Team feature on LinkedIn, the experience was not as good as it is today. At times we could not see the member’s profile picture, past experience, or skills because we did not have the permission to view them. We also sometimes saw duplicate members in the module. Since then, we have improved the data quality by doing deduplication, fetching more members at a single company from the data storage, and selecting members with a strong profile for display. As a result, we are able to showcase higher quality member profiles and give job seekers better insight.
Future of Meet the Team
We believe the newly created Meet the Team module will open a new perspective for our job seekers. We would love to further enhance this module by including aggregate skills of members in similar positions, supporting this feature in other pages (for example the job search page), and providing it to guest users. We also learned how a data-driven new feature can be built from end-to-end from all of the production and engineering challenges we faced. This feature would not have been possible without help and support from the careers team, product team, search team, and data team. A big thanks to them!