eeghor / cat-feeder

Simple and Innovative Recommender

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

--------------- A smart feed guided by semantic AI. You won't miss what matters ---------------

How does it work? In short, Cat Feeder runs an efficient proprietary algorithm on a vast number of data points to figure out similar users and similar posts.

To determine to what extent any two users are similar, our algorithm makes use of the following key data points:

  • age group (as per user profile)
  • gender (inferred by AI)
  • interests (self-reported by user)

To establish similarity between any two posts our system focuses on

  • post text, i.e. what has been written
  • hashtags added by users (if any)
  • explicitly created relationships between posts, e.g. some are replies to others

This similarity knowledge is then used to understand what a user who just logged in may be into. One that's done, Cat Feeder reviews millions of posts to pick a few that deserve attention and shows them in the feed.

Quick Start

Using Cat Feeder is really simple. Here's one of the most popular scenarios to integrate it in your app and greet your user whose ID is f84-7a5-45-a9-91ef with an AI-optimised feed:

import datetime
from cat_feeder import CatFeeder

cf = CatFeeder().feed(uid='f84-7a5-45-a9-91ef', 
                      current_time=datetime.datetime.utcnow())

A list of posts IDs to show to the user is now available in cf.posts_to_show . Note that the list shows the more recommended posts first:

cf.posts_to_show

['5aa337f6-e9e2-489b-8a22-f4e87d2c6ef5',
 'c7341bf2-1c19-41ec-8b0d-3bb0040824b3',
 '5858955c-68c4-4674-9fff-acd968eae9a4']

Selected Features

  • Full spam control - how many posts to recommend is up to you
  • Smart prioritisation by engagement
  • If nothing stands out as worth seeing, let user explore new ideas by like-minded people
  • Time flies, things change, priorities shift. Our recommendations adjust

About

Simple and Innovative Recommender

License:MIT License


Languages

Language:Python 100.0%