How Recommendation System work
My friend and I were going to exam center we were listening to music on his EarPods the song was good, weather was good, Everything perfect right ??
After reaching there he asked for my Spotify playlist
I was silent for a moment he asked again I replied
“Bro I don’t even use Spotify Youtube music works the best for me ”
He said while taking out the key from ignition
“Yeah but Spotify recommendation is the best even I have used Youtube music less adds but its recommendation system sucks Spotify is best at this”
Thats when I the thought hit me Hey Maybe he is right it is true to some extent Spotify works at this best thats when the curiosity hit and I started asking ais to give me relevant blogs and after reading several blogs and watching The wall street Journal videos on Spotify recommendation AI driven recommendation here is the blog for knowing recommendations system a little more
𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗦𝘆𝘀𝘁𝗲𝗺
Mostly Recommendation System is divided into 3 categories
Before we dive into categories lets know that these recommendation relies on feedback (likes,shares etc)
Feedbacks are further Divided into two types
1 . Explicit Feedback : When a user directly tells the system what they like via they dislike or click on don’t show this type of post but since most users rarely upload a feedback so this is hard to track
2. Implicit Feedback : The system takes user data based on their actions, based on what the buy or what they watch, this data is abundant and easy to collect
After collecting our Feedback data is being transformed into a
Rating Matrix
where row shows data and columns shows rating given my each user
There are three main types of recommendation systems
- Content Based
- Collaborative Filtering
- Context filtering
1. Content Based recommendation
It working model is pretty simple it suggest items based on what User liked in past and quietly ignores what users have disliked uses a negative feedback
for ex : if a User liked Mugen train in the past they might like Infinity castle or different movie with same voice actors or genres
It track the tags or metadata to what users liked in the past, it might work good but its actually poor at showing unique or surprising items
Get Aarish Mansur’s stories in your inbox
Join Medium for free to get updates from this writer.
It recommends using Cosine Similarity
In simple terms cosine similarity formula find how similar two things things instead of checking values it checks directions. If two movies,song have similar pattern the angle between them is smaller meaning they are more alike
2. Collaborative Filtering
Now this is somehow pretty interesting and popular technique
It predicts a user’s preferences by using the collective knowledge and behaviors of large numbers of users . Unlike content based filtering, which relies on the properties of items this recommendation focuses on user item interactions, utilizing the feedback or activity history of all users to make predictions.
Some of Famous techniques that collaborative Filtering uses are
Matrix Factorization : Convert users and items into hidden features and use them to predict ratings it tries to fill the ? columns
K nearest neighbor : Find the “K” most similar neighbors and use them to make decisions
lets understand with an example
if I search for sad heart broken song after a breakup then this is how it will search based on tags and in those tag the most nearest one
3 . Context Filtering
We have seen Content Based recommendation now in context filtering recommendations are made based on the user’s current situation or context now instead of only checking what you like the system also checks when, where, mood, time, activity etc
you have often seen people complaining about their Feed in night it becomes sad and filled with couples well you can now blame context filtering for it
Thats it fellas This was all the things which I could understand and tried to convey through this blog
Have a nice day ദ്ദി(˵ •̀ ᴗ — ˵ ) ✧
