1 of 16

Modern Recommendation Systems in Real Applications

Phạm Hoàng Anh - AI Engineer - R&D LAB - SUN ASTERISK VIETNAM

1

2 of 16

The Problems

Youtube

- 300 hours of video are uploaded to YouTube every minute!

- Almost 5 billion videos are watched on Youtube every single day.

- The total number of people who use YouTube 1,300,000,000 users

Amazon

- Sells more than 119,928,851 products

- In the U.S. alone, Amazon has over 150 million monthly unique visitors.

Viblo

- Over 26,000 verified users

- Over 21,000 published posts

- 1.5M Pageviews per month

2

3 of 16

Recommendation System using Big Data

Store & Collect Data

Gathering Raw Data about Users' Activities

- Any events like history/return history, Cart events, Pageviews, Clicks and search log...

Analysis Data

Designing the analysis phase base on the application's requirements

- Time response, Data get and return stream,...

Filter and Recommend Items*

A core component of building a recommendation engine

- Using some algorithms to make good recommendations for users.

3

4 of 16

Impact

Statistic from real world company

- 75% of what consumers watch on Netflix comes from the company’s recommender system

- Amazon credits recommender systems contribute 35% of their revenue

- Recommendations are responsible for 70% of the time people spend watching videos on YouTube.

4

5 of 16

Core of Recommender Engine

5

6 of 16

Content Based

Clustering

Collaborative

Session Based

Combination

6

7 of 16

Content Based

- Most popular in RS

- Based on a description of the item to suggest similar items.

- Creating recommendations for items that have attributes but few user ratings

7

8 of 16

Content Based

8

Vectorizer/Embedding

- Bag of word

- TFIDF

- Item2vec

...

9 of 16

Clustering User

- Cluster Users

- Suggest items that popular with their group

- Not work right with new user and new item

9

10 of 16

Clustering User

10

11 of 16

Collaborative Filtering

11

12 of 16

Collaborative Filtering

12

13 of 16

Session Based

13

14 of 16

Session Based

- Used by many famous E-commerce system (Alibaba, Youtube, Spotify...)

- Many challenge and research each year held

- Work with anonymous users

- Not work right with new item

14

15 of 16

Viblo RS

Developing

max growth

15

16 of 16

Thank you for your listening!

16