using game reviews to recommend games michael meidl, steven lytinen depaul university school of...

43
Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle WA

Upload: sandra-farmer

Post on 08-Jan-2018

216 views

Category:

Documents


0 download

DESCRIPTION

Recommender Systems are Everywhere

TRANSCRIPT

Page 1: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Using Game Reviews to Recommend Games

Michael Meidl, Steven LytinenDePaul University School of Computing,

Chicago IL

Kevin RaisonChatsubo Labs, Seattle WA

Page 2: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle
Page 3: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Recommender Systems are Everywhere

Page 4: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle
Page 5: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle
Page 6: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle
Page 7: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Our Task• Provide a game player with recommendations

of games s/he has not played (and will like)• Recommendations are based on two sources

of information: • Corpus of game reviews (free-form text)• Knowledge about which games a user

already likes (user’s numerical rankings)

Page 8: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Review of Assassin’s Creed Unity

“its complex Abstergo storyline has long since jumped the shark… the story is much darker in tone than anything else in the series…hard to get bored… the attention to detail … is nothing short of astonishing” 6 out of 10

- Mark Walton

What else will Mark like?

Page 9: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Recommender System Techniques

1. Collaborative-based– System compares you to other users, and

recommends what they’ve liked or bought– May know nothing else about the products or

other items that they recommendAmazon, Barnes&Noble, CDW, …

Page 10: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

??

• Collaborative Example (Candeliier et al., 2007)

Page 11: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Recommender System Techniques

2. Content-based– System uses information about the items it

recommends (e.g., recommend books by the same author, or same genre)

– Might not use information about other customers/users

Page 12: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Content-based ExampleTom Hanks

Daisy Ridley

Drama SciFi Comedy Did I like it?

Movie 1 x x x

Movie 2 x x

Movie 3 x x x x x

Movie 4 X X ???

Page 13: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Recommender System Techniques

3. Hybrid: some combination of collaborative-based and content-based

Page 14: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Our game recommender

1. Content-based: A game “representation” is based on the (free-form text) reviews written by a community of users

2. User profile is based on a small sample of items liked by the user

Page 15: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Corpus

• Reviews from www.gamespot.com• 400,000 reviews of 8279 different

games• Mixture of professional reviews

and user reviews

Page 16: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Representing games

• Representation of each game is constructed from a corpus of free-form text reviews of games

• Games represented as vectors• Vector features are based on co-

occurrence of word pairs: adjectives and “context words”

Page 17: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Vector space model

• Originated in information retrieval• Task: judge “similarity” of documents (e.g.,

game reviews)• Document representation: bag of words

Page 18: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Vector space model

1. Build a vocabulary– terms which are “important” in the

collection of documents2. Build the document representations– What terms from the vocabulary appear in the

document, and how frequently relative to other documents?

3. Starting with a document, what others are similar?

Page 19: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Vocabulary: [story, plot, animation, interest, bore, astonish, series, complex,….]

“its complex Abstergo storyline has long since jumped the shark… the complex story is much darker in tone … hard to get bored… the attention to detail … is nothing short of astonishing”

Vector:[1, 0, 0, 0, 1, 1, 1, 2, …]

Page 20: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Vector space, cont.

• Vector values are typically “normalized” to account for a document’s length, the frequency of each term across documents,…

• Documents are similar if their vectors are similar

[1, 0, 0, 0, 1, 1, 1, 2][1, 1, 0, 1, 2, 1, 2, 2] similar[0, 1, 2, 1, 0, 3, 0, 0] dissimilar

Page 21: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Feature space

• 700 adjectives were chosen as most relevant to the description of games (Zagal and Tomuro 2010)• Bootstrapping approach, began with adjectives

modifying “gameplay”• “context words”: words that appear in a

window of +- 2 words from an adjective• Over 3,500,000 adjective-context word pairs• Unworkable feature space size

Page 22: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Feature space

• 700 adjectives were chosen as most relevant to the description of games (Zagal and Tomuro 2010)• Bootstrapping approach, began with adjectives

modifying “gameplay”• “context words”: words that appear in a

window of +- 2 words from an adjective• Over 3,500,000 adjective-context word pairs• Unworkable feature space size

Page 23: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Feature space

• 700 adjectives were chosen as most relevant to the description of games (Zagal and Tomuro 2010)• Bootstrapping approach, began with adjectives

modifying “gameplay”• “context words”: words that appear in a

window of +- 2 words from an adjective• Over 3,500,000 adjective-context word pairs• Unworkable feature space size

Page 24: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Reduction of Feature Space Using Co-clustering

• Simultaneously cluster two sets of related items while minimizing loss of mutual information (Dhillon, Mellela and Mohdha 2003)

• In our case, a set of adjectives X and a set of “context words” Y

• Input: X,Y Output: X’ = {X1 X2, …, Xm}, a partition of X Y’ = {Y1, Y2, … Yn} a partition of Y

Page 25: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Reduction of Feature Space Using Co-clustering

• Simultaneously cluster two sets of related items while minimizing loss of mutual information (Dhillon, Mellela and Mohdha 2003)

• In our case, a set of adjectives X and a set of “context words” Y

• Input: X,Y Output: X’ = {X1 X2, …, Xm}, a partition of X Y’ = {Y1, Y2, … Yn} a partition of Y

Page 26: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Reduction of Feature Space Using Co-clustering

• Simultaneously cluster two sets of related items while minimizing loss of mutual information (Dhillon, Mellela and Mohdha 2003)

• In our case, a set of adjectives X and a set of “context words” Y

• Input: X,Y Output: X’ = {X1 X2, …, Xm}, a partition of X Y’ = {Y1, Y2, … Yn} a partition of Y

Page 27: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Representation of Games

• Collection of reviews for a game were treated as one “document”

• Games represented as vectors• Vector feature = pair of (adjective

cluster) and (context word cluster)• Frequency of co-occurrence of clusters

were counted, and weighted in various ways

Page 28: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Recommending games

G = games already liked by a userG’ = all games user has already played (including disliked ones)S = “seeds” – a small subset of GN = games that user does not knowR = games that our system recommends

Page 29: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Recommending games

R = the k games in N with minimum distance from any of the members of S

|R| = k

Page 30: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle
Page 31: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Evaluation

• “Live” testing was not available to us

• Instead, offline testing:• Recommend k games (|R| = k)

in G’ – S• Find overlap between R and G

Page 32: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Evaluation

• We conducted a n-fold cross-validation of our system’s performance

• Number of folds n = |G’| / |S|• Partition G’ into G’/|S| folds• Measure performance n times for

each S

Page 33: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Evaluation

• We conducted a n-fold cross-validation of our system’s performance

• Number of folds n = |G’| / |S|• Partition G’ into G’/|S| folds• Measure performance n times

for each S

Page 34: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Evaluation

• We measured performance in terms of precision

• precision = |R ∩ (G-S)| / |R|• Precision tends to be highest

for small k and decrease as k increases

Page 35: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Evaluation

• We measured performance in terms of precision

• precision = |R ∩ (G-S)| / |R|• Precision tends to be highest

for small k and decrease as k increases

Page 36: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Evaluation

• We also varied:• Weighting techniques for

features• Dimensionality of co-

clustering

Page 37: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Feature Weighting

• Most common: tf-idf• Document frequency = # of

documents in which a cluster pair appears

• Term frequency (cluster pairs) is multiplied by the inverse of the document frequency

Page 38: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Other Feature Weighting

• tf: “raw” co-occurrence counts• tf-normc: normalize frequency

across documents (“column-wise” normalization)

• boolean: feature value is 1 if cluster pair appears, 0 if not

Page 39: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Results: Feature Weighting

Page 40: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Results: Co-cluster dimensions

Page 41: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Results: Co-clustering vs. “Bag of words”

Page 42: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Conclusions

• Representation of games using approach based on adjective – context word pairs produces high quality recommendations

• Precision of first recommendation is 85-90%

Page 43: Using Game Reviews to Recommend Games Michael Meidl, Steven Lytinen DePaul University School of Computing, Chicago IL Kevin Raison Chatsubo Labs, Seattle

Conclusions

• Precision is approximately 80% even for 10 recommendations

• Co-clustering technique dramatically reduces feature space while maintining high precision

• Dimensionality reduced from 3,500,000 to 1,000 in 10 x 100 co-clustering