Intelligent Recommendor System Based on Textual Content

Objective: Provide a great user experience for students on Xloosv app (private social communication platform) by predicting most interesting posts!

Focus: choose the best feature engineering techniques or implement a new one that could improve the accuracy.

plan:

1. Predict the best posts (Part 1):

• predict the number of upvotes a post will receive based on its textual content (NLP) and store it in a "NLP score" variable

• use Elasticsearch Decay Function to predict the best 10 posts to show based on:

  1. creation Date
  2. number of likes
  3. number of comments
  4. NLP score

2. predict the hashtag/topic/subreddit of a post based on its textual content (Part 2)

Other possible tasks:

• Merge the models to the Xloosv app using Firebase ML then make it continue learning and improving from future data
• repeat the previous tasks, but based on image content (for posts that contain images)

Dataset:

A large collection of Reddit posts:

• https://github.com/umbrae/reddit-top-2.5-million/
• https://www.kaggle.com/unanimad/dataisbeautiful