Download presentation
Presentation is loading. Please wait.
Published byAllyson Quinn Modified over 9 years ago
1
Sean Blong Presents: 1
2
What are they…? “[…] specific type of information filtering (IF) technique that attempts to present information items (movies, music, books, news, images, web pages, etc.) that are likely of interest to the user.” More simply, enhances companies profits as well as the user’s shopping experience. (win-win) 2
3
Who uses them…? 3
4
So why should you care? Netflix Prize Open competition for the best collaborative filtering algorithm to predict user ratings for films, based on previous ratings Prize: $1,000,000 Biggest competitive advantage 4
5
3 Types of Recommendation Personalized recommendation Social recommendation Item recommendation 5
6
Personalized Recommendation Recommend items based on the individual's past behavior. Examples: Pandora Netflix Google 6
7
Netflix (Personalized Recommendation) 7
8
Social Recommendations Recommend items based on the past behavior of similar users Examples: Facebook friend recommendations Amazon Netflix 8
9
Amazon (Social Recommendation) 9
10
Item Recommendation Recommend things based on the item itself Examples: Amazon Most clothing companies Pandora 10
11
Urban Outfitters (Item Recommendation) 11
12
12
13
Google customizes your search results based on your location and/or recent search activity. When signed in to a Google Account, you will see even more relevant results based on your web history. 13
14
Google's search algorithm PageRank is dependent on social recommendations (who links to a webpage) Google also does item recommendations with its "Did you mean" feature. Try typing recursion in the search bar. 14
15
Pandora relies on a Music Genome that consists of 400 musical attributes covering the qualities of melody, harmony, rhythm, form, composition and lyrics. Item based recommendations based on these musical attributes. Not a social recommendation system!!! 15
16
Combines all 3 techniques: All recommendations are based on individual behavior, the item itself, and the behavior of other people on Amazon. 16
17
17 Item/Social Recommendation Personal/Item Recommendation
18
How it applies to Advanced Clothing Solutions… 18
19
Goals: Store user data: What they’ve bought/own, what they’ve tried on, what they like/don’t like. Make recommendations: Utilizing the Item, Social, and Personal Recommendation systems. Utilize data to create personalized sales, deals, and coupons. i.e. Increase profits and shopping experience! 19
20
Challenges: THE ALGORTHIM How to assign similarity through tags? ○ How to assign tags? (see ER diagram) How to assign individual weights of the three recommendation facets (personal, social, item). How to accurately portray user’s tastes using a binary ranking system (think Pandora) 20
21
A Look at the Database… 21
22
More simply… User: username, userid, name, address, phone Article: articleid, type, gender, color, size, description, company name Likes: userid, articleid, rating So what’s the issue…? 22
23
Problems with Social Recommendation vs. Personal Recommendation 23 Social: User: Social: User:
24
The technical side of Recommendation Systems… 24 Latent Factor (matrix factorization) vs. Nearest Neighbor Latent Factor: become popular in recent years by combining good scalability with predictive accuracy. In addition, they offer much flexibility for modeling various real- life situations.
25
Nearest Neighbor 25
26
Latent Factor 26
27
Matrix Factorization (cont.) Other items to consider: Adding biases Additional input sources Temporal dynamics Inputs with varying confidence levels 27
28
28
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.