Downstream effects

LinkedIn Engineering said in April 2023 that shares, comments and reactions can expose posts to more members and create a cascade that makes a post spread through the network8. In May 2020, LinkedIn Engineering gave the example of a reshare carrying an article downstream because the member's connections would see it in their feed2. LinkedIn Engineering also said in May 2020 that a comment gives the creator feedback that may encourage them to post more often2. A Microsoft patent from November 2022 describes downstream effects as effects related to an immediate action that happen later, with sharing as a major example, because the people who see a share may interact with the item and share it again, and so on1. A Microsoft patent from October 2021 adds that not all viral actions are equal: a share by the president of a multinational company has more effect than a share by a factory worker, and a share may have more effect than a like3.

Estimating downstream engagement means turning that spread into numbers, and LinkedIn Engineering said in May 2020 that it trains machine learning models to predict several quantities for each possible click and viral action (click, react, comment, share), including two described here2. One, P(action), is the probability that the member takes that action on the post2. Another, E [downstream clicks/virals | action], is the number of downstream clicks and viral actions expected if the member takes that action2. The 2022 Microsoft patent says it is technically complex to design an algorithm that captures how likely an action is to cause a downstream event, let alone what that event is worth1. A Microsoft patent from October 2021 describes a score called expected engagement, written E(m, u), which is a predicted score for showing post u to member m and which includes expected downstream engagement3. In that formula, one part is the chance the member takes any click-based action, and a second part adds up, for viral actions such as like, share and comment, the chance of each action multiplied by the extra downstream engagement it would bring on the post3. This formula comes from a patent and is not confirmed live LinkedIn behavior3.