Loading presentation...

Present Remotely

Send the link below via email or IM


Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.


Recommender systems (S523)

No description

Jingjing Zhang

on 20 April 2014

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Recommender systems (S523)

An Introduction Recommender Systems: Questions? Issues and Concerns Our World is an over-crowded place Examples Pandora Amazon Netflix Recommender System Recommender systems have become
extremely common in recent years.
A few examples of such systems: Pandora Radio takes an initial input of a song or musician and plays music with similar characteristics (based on a series of keywords attributed to the given artist or music). The stations created by Pandora can be refined through user feedback (emphasizing or deemphasizing certain characteristics). When viewing a product on Amazon.com, the store will recommend additional items based on what other shoppers bought along with the currently selected item.
35% sales at Amazon are from recommendations Netflix offers predictions of movies that a user might like to watch based on the user's previous ratings and watching habits (as compared to the behavior of other users), also taking into account the characteristics (such as the genre) of the film.
75% of videos watched by Netflix users is suggested by recommender system Approaches of RecSys Collaborative
Filtering Hybrid Approach Content-Based
Filtering Types of
Collaborative systems User-Based Method Item-Based Method Matrix Factorization Matrix Factorization (Netflix)
You many have watched thousands of movies, but perhaps I can describe these movies using 6 attributes
So 6 numbers are enough to describe your taste
Likewise, "Stargate" has been watched by millions people, but perhaps 6 numbers are enough to describe its features
Magic: the hidden aspects (both for item and user) can be discovered automatically by Matrix Factorization Pros It does not require large user groups to achieve reasonable recommendation accuracy.
New items can be immediately recommended once item attributes are available. Examples Pandora Radio is a popular example of a content-based recommender system that plays music with similar characteristics to that of a song provided by the user as an initial seed.
There are also a large number of content-based recommender systems aimed at providing movie recommendations, a few such examples include Rotten Tomatoes, Internet Movie Database, Jinni, Rovi Corporation and See This Next. Pros Example Cold Start Other Issues Privacy Problems
–User cold start : new users
–Item cold start : new items How to recommend items to new users?
Non-personalization recommendation
- Most popular items
- Highly Rated items
Using user register profile (Age, Gender, …) User Cold Start Item Cold Start Solution: 1) Registration of profiles
2) Detection and Analysis Privacy is considered as a very important aspect in e-commerce as there is legislation on the distribution of private information of users to third parties.
To assure privacy of customers and the fact that the company is not liable for any lawsuits, steps need to be taken to assure that privacy related material cannot be subtracted from statistics and other used material in the recommender system. Sparisty:
The total number of items is extremely large, however, even the most active users will only have consumed a small subset of the overall database.
E.g., 1% of the ratings are known in Netflixt dataset
In real world systems, there are millions of users and products. Thus, a large amount of computation power is often necessary to calculate recommendations Presented by
Jingjing Zhang, PhD
jjzhang@indiana.edu Evaluation Agenda Why Recommender System?
Objective of Recommender Systems
Types of Recommender System
Evaluation of Recommendation
Issues and Concerns We are overloaded But we really need and consume only a few of them! Can Google Help? Can Facebook Help? Can Experts Help? Yes, but only when we really know what we are looking for

What if I just want some interesting music tracks?
- Btw, what does it mean by “interesting”? Yes, I tend to find my friends’ stuffs interesting

What if I had only few friends, and what they like do not always attract me? Yes, but it won’t scale well
- Everyone receives exactly the same advice!

It is what they like, not me!
- Like movies, what get expert approval does not guarantee attention of the mass We need Help! Objective Applications Technologies Evaluation Definition Applications What do recommender
systems do, exactly? Definition(s) RS are software agents that elicit the interests and preferences of individual consumers[…] and make recommendations accordingly. They have the potential to support and improve the quality of the decisions consumers make while searching for and selecting products online. (Xiao & Benbasat, 2007)

Use the opinions of a community of users to help individuals in that community to identify more effectively content of interest from a potentially overwhelming set of choices. (Resnick & Varian, 1997)

Any system that produces personalized recommendations as outputs or has the effect of guiding the user in a persoanlzied way to interesting or useful objects in a large space of possible options. (Burke 2002) Find good items: presenting a ranked list of recommendations
Final all good items: identify all items that might be interesting
Recommend sequence of items: sequence of related items is recommended
Annotation in context: predict usefulness of an item that user is currently viewing
and many more... and Many (!) More applications Advertising messages
Investment choices
Music tracks
TV programs
Supermarket goods Tags
News articles
Online mates (Dating services)
Future friends (Social network sites)
Courses in e-learning
Drug components
Research paper
Code modules
Programmers Types of Recommender Systems User-based method (Movielens.org)
Many people liked "Music and Lyrics"
Can you tell how much I like it?
The idea is to find my "friends" who share similar taste with me, then how much I like depend on how much THEY liked.
User similarity: Pearson correlation, Cosine coefficient, etc.
Basic idea: you may like it because your “friends” liked it Item-based method (Amazon.com)
Based on so many (good & bad) movies that I watched in the past, would you recommend me watching “Taken”?
The idea is to find my previously watched movies that share similar audience with “Taken”, then how much I will like depend on how much I liked similar movies
Item similarity: Pearson correlation, Cosine, etc.
Basic idea: I tend to like this movie because I have liked those similar movies... or
because people who have watched those movies also liked this movie (Amazon implementation) User-based Collaborative Method
Item-based Collaborative Method
Matrix Factorization Hybrid approaches can be implemented in several ways: Hybrid Approach By making content-based and collaborative-based predictions separately and then combining them

By adding content-based capabilities to a collaborative-based approach(vice versa)

Or by unifying the approaches into one model Hybrid methods can provide more accurate recommendations than pure approaches. These methods can also be used to overcome some of the common problems in recommender systems such as cold start and the sparsity problem.

Netflix and See This Next are good examples of hybrid systems. They make recommendations by comparing the watching and searching habits of similar users (i.e. collaborative filtering) as well as by offering movies that share characteristics with films that a user has rated highly (content-based filtering). Result: It will predict the rating of “Music and Lyrics” for Alice as 9, because that’s how Christine and David (most similar users to Alice) liked this movie. Task: predict the rating of “Music and Lyrics” for Alice using the ratings of most similar users other examples? Have you ever used
Grouplens? Amazon recommendation? Netflix Cinematch? Google news? Strands? Tivo? Frindory? Pandora? Web page: words, hyperlinks, images, tags, comments, titles, URL, topic

Music: genre, rhythm, melody, harmony, lyrics, meta data, artists, bands, press releases, expert reviews, loudness, energy, time, spectrum, duration, frequency, pitch, key, mode, mood, style, tempo

User: age, sex, job, location, time, income, education, language, family status, hobbies, general interests, Web usage, computer usage, fan club membership, opinion, comments, tags, mobile usage Item Descriptions & User Profiles Content Representation What is Content? Can we acquire content pieces automatically?
Fairly easy for text
Difficult for music and video, except for digital signals, e.g. music genre classification 60-80% accuracy
A lot of noise, e.g. misplaced tags What can we do with these contents?
Compute similarity between items or users
Query items that are similar to a given item
Match item’s content and user’s profile How do we know the recommendation is good?
Evaluation criteria (Herlocker et al., 2004)
Mostly measured by Accuracy
Closeness between predicted rating and actual user rating
e.g., MAE, RMSE, Precision, Recall
Predictive accuracy isn't enough
Other important metrics:
Coverage, diversity, novelty, robustness, profitability, etc
Consistency/Stability (Zhang, 2012) Recommender Systems in Academia Early days: 3 papers by HCI researchers (1995)
Annual ACM RecSys Conference
in Hong Kong this year
Over 1000 papers
Fields: CS/IS, marketing, DM/statistics, MS/OR
Netflix $1M Prize CompetitionData:
18K movies, 500K customers, 100M ratings
$1M Prize: improve Netflix RMSE rates by 10%
> 40K contestants from 179 countries
Final winner: BellKor’s Pragmatic Chaos at 10.06% To recommend to us something we may like
- It may not be popular
- Based on our history of using services
- Based on other people like us OK, here is the idea called Recommender Systems ! Your Taste Item Feature How to recommend new items to user? Do not recommend: Using content information: Nuke attack is inserted to decrease the number of times an item is recommended to different users.
Push attacks increase the number of times an item is recommended A profile injection attack (or shilling attack): the goal is to create a bias in the recommender system by inserting fake user ratings. Malicious users Content-based Approach
Full transcript