LinkedIn is optimizing its “People You May Know” feature to eradicate biases keeping members with smaller networks behind.
As part of its commitment to creating economic opportunity for its members through equitable results, LinkedIn has previously helped with closing the network gap and sharing data-backed recommendations for boosting one’s network.
Related | LinkedIn Looks To Expand Its Video Creator Tools With Jumprope Acquisition
Now, it’s looking to optimize member experiences by creating more equity in connection opportunities through its People You May Know (PYMK) recommendation system.
PYMK – a part of the My Network tab – has been a long-standing part of the platform. Powered by machine learning, it serves to help members connect to others who may be relevant to their own professional networks.
LinkedIn describes it as using “data like the Economic Graph and platform interactions to mine features and use ML algorithms to come up with relevant recommendations.” Essentially, it estimates “the propensity to connect between two members” recommending a list of potential new connections.
However, as an AI system, it is prone to accuracy biases coming from external factors, like a member’s general visibility off-platform. As such, the system may reflect an existing bias towards some groups of people over others.
Over the last year, LinkedIn has been improving and updating the underlying PYMK algorithms in order to make the feature more equitable and more effective for members, regardless of their existing network strength or frequency of platform usage.
Results from the changes demonstrated a surge in engagement with PYMK – i.e. invitations sent. LinkedIn’s previous experiments had also led to similar results. where the platform changed the LinkedIn Feed to optimize for creators and not just viewers. “In that instance, too, moving away from strictly ranking feed updates based on, essentially, the potential for virality led to positive engagement wins,” explains Qiannan Yin, Tech Lead of Growth Data Science at LinkedIn.
LinkedIn wants to ensure that both members benefit when connecting to each other. So its engineering teams looked into improving the poor experience that PYMK might provide to extremely popular LinkedIn members – like industry influencers, high-profile senior executives, or recruiters from big companies – who receive a large number of connection requests.
Essentially, members who are inundated with connection invites may end up with a crowded feed and too many notifications to handle – mostly random ones – thus leading to them missing relevant networking opportunities.
So, LinkedIn started de-ranking members with an excess number of invitation requests. The result? Popular members who are flooded with invitations show up less in PYMK results.
The changes reduced the number of invitations received by overloaded recipients, providing them with a better user experience overall.
However, to ensure that PYMK fairly represents infrequent members as well, LinkedIn de-ranked connection suggestions to fairly represent infrequent and frequent members by giving each equal representation in recommendations. As a result, invitations sent to infrequent members increased by 5.44% and they also established a further 4.8% connections.
Despite expecting to see fewer connections for people who are suggested less in PYNK, this wasn’t the case. With the change, metrics for frequent members remained neutral.
This was a clear indication that recommendation quality was improved.