Artificial Intelligence with

Web Applications

Dell Zhang

Birkbeck, University of London

2010/11

CI Chapter 2

Making Recommendations

Recommendation Systems

State of the Art

An Ensemble The Neighbourhood Approach

User-based Collaborative Filtering Item-based Collaborative Filtering

The Latent-Factor Approach ……

The Neighbourhood Approach

The nearest-neighbour method works on the principle that a person tends to give similar ratings to similar movies. Joe likes the three movies on the left, so to make

a prediction for him, find users who also liked those movies and see what other movies they liked.

Here the three other viewers all liked Saving Private Ryan, so that is the top recommendation. Two of them liked Dune, so that’s ranked second, and so on.

The Latent-Factor Approach

A second, complementary method scores both a given movie and viewer according to latent factors, themselves inferred from the ratings given to all the movies by all the viewers.

The factors define a space that at once measures the characteristics of movies and the viewer’s interest in those characteristics. Here we would expect the fellow in the southeast

corner of the graph to love Norbit, to hate Dreamgirls, and, perhaps, to rate Braveheart about average.

Collaborative Filtering (user-based)

Collecting Preferences

Possible mappings of user actions to numerical scores

Collecting Preferences

A set of movie critics and their ratings of movies 'Lisa Rose': {'Lady in the Water': 2.5, 'Snakes on a Plane': 3.5, 'Just My

Luck': 3.0, 'Superman Returns': 3.5, 'You, Me and Dupree': 2.5, 'The Night Listener': 3.0}

'Gene Seymour': {'Lady in the Water': 3.0, 'Snakes on a Plane': 3.5, 'Just My Luck': 1.5, 'Superman Returns': 5.0, 'The Night Listener': 3.0, 'You, Me and Dupree': 3.5}

'Michael Phillips': {'Lady in the Water': 2.5, 'Snakes on a Plane': 3.0, 'Superman Returns': 3.5, 'The Night Listener': 4.0}

'Claudia Puig': {'Snakes on a Plane': 3.5, 'Just My Luck': 3.0, 'The Night Listener': 4.5, 'Superman Returns': 4.0, 'You, Me and Dupree': 2.5}

'Mick LaSalle': {'Lady in the Water': 3.0, 'Snakes on a Plane': 4.0, 'Just My Luck': 2.0, 'Superman Returns': 3.0, 'The Night Listener': 3.0, 'You, Me and Dupree': 2.0}

'Jack Matthews': {'Lady in the Water': 3.0, 'Snakes on a Plane': 4.0, 'The Night Listener': 3.0, 'Superman Returns': 5.0, 'You, Me and Dupree': 3.5}

'Toby': {'Snakes on a Plane':4.5,'You, Me and Dupree':1.0,'Superman Returns':4.0}

Finding Similar Users

Euclidean Distance Score Pearson Correlation Score

Euclidean Distance Score

The straight-line distance between two points in a multidimensional space, which is the kind of distance you measure with a ruler.

Given the Euclidean distance d between two points, their similarity score can be defined as 1/(d+1).

Euclidean Distance Score

The movie critics in (Euclidean) preference space

Pearson Correlation Score

A measure of how correlated two variables are (how well their values fit on a straight line) It is a value between 1 and –1

1: the variables are perfectly correlated 0: the variables are not correlated –1: the variables are perfectly inversely correlated

It tends to work better in situations where the dataset isn’t well normalized

Pearson Correlation Score

Comparing two movie critics on a scatter plot

lowpositive correlation

Pearson Correlation Score

Comparing two movie critics on a scatter plot

highpositive correlation

Pearson Correlation Score

User-based Collaborative Filtering

Step 1: Look for users who share the same rating patterns with the active user (the user whom the prediction is for) e.g., using the k-Nearest Neighbours algorithm

Step 2: Use the ratings from those like-minded users to calculate a prediction for the active user.

User-based Collaborative Filtering

The user-based recommendations for Toby

Weighted Mean

A type of average that has a weight for every observation being averaged

Finding Similar Items

Just switch user and items in the previous slides!

Finding Similar Items

Comparing two movies on a scatter plot

negative correlation

Finding Similar Items

Item-based Collaborative Filtering

Step1: Build an item-item matrix determining relationships between pairs of items. The similarities between items will not change as

often as the similarities between users. Step 2: Infer the taste of the active user, using

his top-rated items and the above item-item matrix.

Item-based Collaborative Filtering

The item-based recommendations for Toby

User-based vs Item-based

Efficiency The latter is significantly faster than the former on

large datasets as it allows many calculations to be performed in advance, but it has the additional overhead of maintaining the item similarity table.

Effectiveness The latter usually outperforms the former on

sparse datasets, while they perform about equally on dense datasets.

Simplicity The former is simpler to implement as it doesn’t

have the extra steps.