Finally, we reach the contextual phase, where some rules are applied, such as diversifying the content to ensure that it is more enjoyable and does not have the same type of publication continuously.
Integrity process
Once all these posts have been filtered, the filtering process begins: this is where it is determined which posts comply with Facebook's policies, eliminating all those that do not comply with these conditions. These posts would include those that spread false information, those that present sensitive content and so-called clickbaits .
At this point we already have a smaller number of posts, around 500. However, there is still work to be done, since although all of them may be similar to the user's taste, not all of them are equally relevant. Therefore, before reaching the contextual step, which would be the result that the user sees, we have the following point.
Scoring pass
Next is the scoring pass where most of the personalization happens. Here, a score for each story is calculated independently, and then the 500 posts are sorted by score. For some, the score may be higher for likes than for comments , as some people like to express themselves more through likes than comments . Any action that a person rarely engages in (e.g. a prediction of liking that is very close to zero) automatically gets a minimal role in the ranking, as the predicted value is very low.
At this point we need to discern which posts are more relevant than others. To do this we need to focus for a moment on the user. Let's take the example given by the social network in its press release.
Since Juan logged in yesterday, his friend Wei posted a photo of his cocker spaniel. Another friend, Saanvi, posted a video of her morning run. His favorite page posted an interesting article about the best way to see india business email database the Milky Way at night, while his favorite cooking group posted four new sourdough recipes.
In this case, the user, Juan, has 4 posts that could be in first place . What will determine which of these two will have a better rating will be the level of interaction that he maintains with the user's posts and the type of post .
“We can use features of a post, such as who is tagged in a photo and when it was posted, to predict whether John might like it. For example, if John tends to engage with Saanvi’s posts (e.g., share or comment) frequently and her running video is very recent, there is a high likelihood that John will like her post. If John has engaged with more video content than photos in the past, the prediction for Wei’s photo of her cocker spaniel might be quite low. In this case, our ranking algorithm would rank Saanvi’s running video higher than the photo of Wei’s dog because it predicts a higher likelihood that John will like it.”
It is also necessary to evaluate the long-term impact that publications may have: Facebook achieves this through surveys carried out on its users, from which they obtain an estimated value with which to discern this impact for all its users.
Contextual Pass
Finally, we reach the contextual phase, where some rules are applied, such as diversifying the content to ensure that it is more enjoyable and does not have the same type of publication continuously.