Practical advises on recommendation systems in production

Most recommendations use the assumption that the user&#x27;s need is the same at every point in time. This is why it doesn&#x27;t work most of the time.<p>I think it&#x27;s strange recommendation systems don&#x27;t ask more about the context.<p>The answer to &#x27;How do you feel today&#x27; can result in a total different playlist.<p>Or a python programmer who also likes snakes might want to set the context before a search (in DDG you can).

&quot;Running a logistic regression might be too informed a baseline and can become a long drawn process.&quot;<p>What does this mean? To me it implies logistic regression is already so good that its hard to see the difference between it and most recommender systems. In which case, unless you&#x27;re operating at massive scale, you can just go with logistic regression

How about this-- give me a sticky toggle to switch to a dumb sorted list-- other channel content, alphabetical etc.<p>That at least lets the user run the experiment on themselves to check if the recommendation algo makes them sad.<p>The fact that the sentence above is based on a real thing Facebook did to its unwitting users speaks volumes about our shitty timeline.

Please correct the title. It&#x27;s &quot;advice&quot;, not &quot;advises&quot;.<p>Advise is a verb. It reads as if an organization called &quot;Practical&quot; is advising on something.<p>Edit: advice, no need for plural as in advices

Based on the advices and the feedbacks received about our datas and evidences, we added syntactic sugars to our codes, which is sure to make us more moneys. Thank you for your hard works.<p>This is indeed excellent new.