3 ms·
Technologically the task can be automated by scraping each submission when it’s submitted, calculate the sentence embeddings of the article, and compare the new
by supriyo-biswas 3y ago
Technologically the task can be automated by scraping each submission when it’s submitted, calculate the sentence embeddings of the article, and compare the new submissions with other embeddings. However, that’s how you become a Facebook style social network with “more you might like” and whatnot.
- woolion 3y agoI agree, hn clearly stays minimal on purpose, which is why I put the last disclaimer line. >with “more you might like” In itself that's a very good feature, it becomes a problem with a host of other patterns to transform discoverability into addiction. Most often these 'bad recommendations' are high engagement content (controversial, or low-value but high attractiveness clickbait) that are purposefully known by the platform to only be tangentially related to the content, if at all.
- dang 3y ago> scraping each submission when it’s submitted Even that is surprisingly hard! Hard enough that we had to stop working on it when we realized that it's a startup, or at least a major undertaking, in its own right.