Title: Recommender Systems 1: Contentbased filtering
1Recommender Systems 1 Content-based filtering
2Why filtering?
- Information overload
- Too many
- movies, books, webpages, songs, plumbers, etc
- Searching is difficult
3Recommender Systems
- Systems that help find the good stuff
- Systems that make personalized recommendations of
goods, services, and people (Kautz)
4Recommender Systems How
- User identifies one or more objects as being of
interest - The recommender system suggests other objects
that are similar
5What does similar mean?
- Similar in content
- Similar in appreciation by other users gt
social filtering (next lecture)
6How do we know users opinion?
- Explicit users rate items
- Implicit looking at clicks etc.
7Example XLibris
- User reads text and annotates
- System generates links and further reading list
8Example MovieCentral
To test out a Movie Recommender go to
http//movielens.umn.edu/
- User rates movies
- The system suggests best bets
- Users keep rating movies while checking best bets
9Exercise
- Suppose our user has rated ten movies
- Rating from A to F
- - Jurassic Park - Harry Potter
- - ET - Lord of the Rings
- - Alien - Terminator
- - 101 Dalmatians - Titanic
- - Sleepless in Seattle - Mr Bean
- Which movie do we recommend?
10Observations
- We use our knowledge
- about the items rated
- about other items
- In particular, attributes like type of movie.
- Multiple attributes are likely to be important.
11Needs something like
- Description of items in terms of attributesFor
example Type, Director, Actors, ... - Description via keywords
- Possibility to look at content itself, like the
text
12What has been used
- Important words in content
- 100 words with highest TF-IDF weights (words
occurring more frequently than on average) - 128 most informative words (words more associated
with one class of documents than another) - For example for restaurant descriptions words
like noodle, shrimp, basil, exotic,
salmon
13Example
- noodle shrimp basil exotic
salmon Opinion - Kitima Y Y Y Y
Y - - MarcoPolo Y Y
- Spiga Y Y
- Thai Touch Y Y Y
- - Dolce Y Y
Y ?
14Possible algorithm (1)
- Tend to be classifiers
- Learn weights (wi) for words so that ? wi for
words occurring gt threshold - Initially, weights are 1
- For each rated example determine sum
- If sum above threshold, and user did not like
example, then divide all weights by 2
15Possible algorithm (2)
- If sum below threshold, and user did like
example, then multiply all weights by 2 - Recommend items with highest sum
16Example (Step 1)
- noodle shrimp basil exotic
salmon - Weights 1 1 1
1 1 - Threshold 2
noodle shrimp basil exotic
salmon Opinion Kitima Y Y
Y Y Y - Sum 5 gt 2,
and opinion negative, so, divide weights by 2
17Example (Step 2)
- noodle shrimp basil exotic
salmon - Weights 1/2 1/2 1/2 1/2
1/2 - Threshold 2
noodle shrimp basil exotic
salmon Opinion MarcoPolo Y
Y Sum 1
lt 2, and opinion positive so, multiply by 2
18Example (Step 3)
- noodle shrimp basil exotic
salmon - Weights 1/2 1 1 1/2
1/2 - Threshold 2
noodle shrimp basil exotic
salmon Opinion Spiga Y
Y Sum
1 1/2 lt 2, and opinion positive so, multiply by
2
19Example (Step 4)
- noodle shrimp basil exotic
salmon - Weights 1 1 2
1/2 1/2 - Threshold 2
noodle shrimp basil exotic
salmon Opinion ThaiTouch Y Y
Y - Sum 2
1/2 gt 2, and opinion negative so, divide by 2
20Example (Step 5)
- noodle shrimp basil exotic
salmon - Weights 1/2 1/2 2 1/4
1/2 - Threshold 2
noodle shrimp basil exotic
salmon Opinion ThaiTouch Y Y
Y - Sum 2
1/2 gt 2, and opinion negative so, divide by 2
ETC ETC, repeat till all examples ok, or do all
10 times
21Multimedia Information Retrieval
- Images Photo collections, Face recognition
- Video Movie recommendation, Electronic Program
Guides - Spoken documents
- Music
- Other sounds
22Concept-Based Image Retrieval
- Key Concept-based indexing of images
- Based on attributes extracted manually
- Based on logical, high level features
- Systems for image indexing
- ICONCLASS, AAT,
- What?
- Time, location, content
23Content-based Image Retrieval
- Key Automatic indexing of images based on
low-level features - Color
- Texture
- Shape
- Spatial orientation and layout
- Sketch
24Examples of Content-Based Image Retrieval
- QBIC - IBMs Query By Image Content
http//wwwqbic.almaden.ibm.com - MIT PhotoBook (Source of following examples)
http//vismod.media.mit.edu/vismod/demos/photoboo
k/ - VisualSeek http//www.ctr.columbia.edu/VisualSEEk
25Image input to search
26Image input to search
27(No Transcript)
28(No Transcript)
29(No Transcript)