clustering-based collaborative filtering for web page recommendation csce 561 project proposal...
TRANSCRIPT
Clustering-based Clustering-based Collaborative filtering for web Collaborative filtering for web
page recommendation page recommendation
CSCE 561 project ProposalCSCE 561 project Proposal
Mohammad Amir SharifMohammad Amir Sharif
[email protected]@louisiana.edu
Presentation Presentation OutlineOutline
Introduction to Recommendation SystemsIntroduction to Recommendation Systems Clustering based recommendation Clustering based recommendation
algorithmalgorithm Implementing Clustering based webpage Implementing Clustering based webpage
recommendation using mahoutrecommendation using mahout Experimental set-upExperimental set-up Evaluation of the developed SystemEvaluation of the developed System ReferencesReferences
Information OverloadInformation OverloadNews items,
Books, Journals,
Research
papers
TV programs,
Music CDs,
Movie titles
Consumer
products, e-
commerce
items,
Web pages,
Usenet articles,e-mails
IntroductionIntroduction What is recommendation system? What is recommendation system? – – Recommend related items Recommend related items – – Personalized experiences Personalized experiences
Introduction(cont)Introduction(cont)
Components of a recommender Components of a recommender systemsystem
– – Set of users, set of items Set of users, set of items (products)(products) – – Implicit/explicit user rating on Implicit/explicit user rating on itemsitems – – Additional informationAdditional information: : trust, trust, collaboration, etc.collaboration, etc. – – Algorithms for generating Algorithms for generating recommendationsrecommendations
Introduction Introduction (cont)(cont)
Recommendation techniquesRecommendation techniques-Collaborative Filtering (CF)Collaborative Filtering (CF)
-Memory-based algorithms: user-Memory-based algorithms: user-based, item-basedbased, item-based-Model-based algorithms: Bayesian Model-based algorithms: Bayesian network ; network ; Clustering Clustering ; Rule-based ; ; Rule-based ; Machine learning on graphsMachine learning on graphs;; PLSA; PLSA; Matrix factorization Matrix factorization
-Content-based recommendationContent-based recommendation-Hybrid approachesHybrid approaches
CF AlgorithmCF Algorithm
Problems: large-scale data; sparse Problems: large-scale data; sparse rating matrix rating matrix
Clustering based Clustering based Collaborative FilteringCollaborative Filtering
x xx x
x x xx x x
x x xx x x
x xx x
x x xx x x
x x xx x x
x xx x
x x xx x x
x x xx x x
Cluster 2Cluster 1
item-based CF
User clustering
item-based CF
Find the most similar cluster for an active userFind the most similar cluster for an active user Apply Similarity measure among current and other Apply Similarity measure among current and other usersusers Users’ similarities are used to predict the Users’ similarities are used to predict the recommendation value of an item for active userrecommendation value of an item for active user
Experimental set Experimental set upup
13745 preprocessed user session data on 683 13745 preprocessed user session data on 683 pages are available for this experiment.pages are available for this experiment.
User-Item pageview Matrix of size 13745×683 User-Item pageview Matrix of size 13745×683 where each cell represents the page view where each cell represents the page view time of a user for a page in a particular time of a user for a page in a particular session.session.
Apache Mahout which works on top of Hadoop Apache Mahout which works on top of Hadoop will be used to make the clustering of user will be used to make the clustering of user sessionssessions
The Apache Hadoop and mahout are open The Apache Hadoop and mahout are open source software library for large scale source software library for large scale distributed computing and machine learning distributed computing and machine learning respectively respectively
Experimental set up(cont)Experimental set up(cont) Sample row of the data setsSample row of the data sets
0 0 3 0 0 5 0 4 6 2 0 0 0 0 0 70 0 3 0 0 5 0 4 6 2 0 0 0 0 0 7 Vector similarity Vector similarity
– similarity among active session and similarity among active session and
cluster centercluster centerx x
x xx x x
x x xx x xx x x
Experimental set Experimental set up(cont)up(cont)
Similarity of Similarity of
user u and v, user u and v, WWu,vu,v
Predicted Recommendation of user Predicted Recommendation of user aa to item to item i, i, PPa,ia,i
Here, Here, rru,iu,i is the user is the user uu’s pageview time ’s pageview time for page for page ii
I I is set of pages, is set of pages, rruu and and rrvv are average are average pageview time of userpageview time of user u u and and vv..
x xx x
x x xx x x
x x xx x x
Evaluation Evaluation MetriceMetrice
Where, Where, rrmaxmax and and rrminmin are the upper and lower bounds of pageview are the upper and lower bounds of pageview time time ppi,ji,j is the prediction for user is the prediction for user ii to item to item j, rj, ri,j i,j is the is the pageview time of user pageview time of user ii to page to page jj
Mean Absolute Error and Normalized Mean Absolute Error:Mean Absolute Error and Normalized Mean Absolute Error:
ReferencesReferences
http://hadoop.apache.org/http://hadoop.apache.org/ http://mahout.apache.org/http://mahout.apache.org/ Manh, C., P., Yiwei C., Ralf K., Matthias J., A Clustering Manh, C., P., Yiwei C., Ralf K., Matthias J., A Clustering
Approach for Collaborative Filtering Recommendation Approach for Collaborative Filtering Recommendation Using Social Network Analysis, Using Social Network Analysis, Journal of Universal Journal of Universal Computer Science, vol. 17, no. 4 (2011), 583-604Computer Science, vol. 17, no. 4 (2011), 583-604
An Efficient Information An Efficient Information Retrieval SystemRetrieval System
Objectives:Objectives: Efficient Retrieval incorporating Efficient Retrieval incorporating
keyword’s position; and occurrences keyword’s position; and occurrences of keywords in heading or titles in the of keywords in heading or titles in the inverted index. inverted index.
Retrieve relevant documents Retrieve relevant documents considering proximity (considering proximity (Example: “dogs” Example: “dogs” and “race” within 4 words)and “race” within 4 words) of query termsof query terms
Evaluation of the systemEvaluation of the system
Extension to Assignment # 3 Extension to Assignment # 3
Inverted IndexInverted Index
system
computer
database
science D2, 4
D5, 2
D1, 3
D7, 4
Index terms df
3
2
4
1
Dj, tfj
Index file Postings lists
cats dogs fish goats sheep
whales
(1,1): 1
(1,2): 2,3
(4,1): 1
(3,1): 2
(2,1): 3
(3,1): 1
(2,1): 2
(2,1): 1 (3,1): 2
(4,2): 1
Inverted Index (cont.)Inverted Index (cont.)
HashMaptokenHash
Stringtoken
TokenInfo
doubleidf
ArrayListoccList
TokenOccurenceDocumentReferencedocRef
intcount
Filefile
doublelength
TokenOccurenceDocumentReferencedocRef
intcount
Filefile
doublelength
…
Inverted Index (cont.)Inverted Index (cont.)
HashMaptokenHash
Stringtoken
TokenInfo
doubleidf
ArrayListoccList
TokenOccurenceDocumentReferencedocRef
Weight
Positi-ons
Filefile
doublelength
TokenOccurenceDocumentReferencedocRef
Weight
Positi-ons
Filefile
doublelength
…
Based on frequency, heading etc
Stores the positions of occurrences
Creating an Inverted IndexCreating an Inverted Index
Create an empty HashMap, H;Create an empty HashMap, H;For each document, D, (i.e. file in an input directory):For each document, D, (i.e. file in an input directory): Create a HashMapVector,V, for D;Create a HashMapVector,V, for D;
For each (non-zero) token, T, in V:For each (non-zero) token, T, in V: If T is not already in H, create an empty If T is not already in H, create an empty
TokenInfo for T and insert it into H;TokenInfo for T and insert it into H; Create a TokenOccurence for T in D and Create a TokenOccurence for T in D and add it to the occList in the TokenInfo for T;add it to the occList in the TokenInfo for T;Compute IDF for all tokens in H;Compute IDF for all tokens in H;Compute vector lengths for all documents in H;Compute vector lengths for all documents in H;
Inverted-Index Retrieval Inverted-Index Retrieval AlgorithmAlgorithm
Create a HashMapVector, Q, for the query.Create a HashMapVector, Q, for the query.Create empty HashMap, R, to store retrieved documents Create empty HashMap, R, to store retrieved documents
with scores.with scores.For each token, T, in Q:For each token, T, in Q: Let I be the IDF of T, and K be the count of T in Q;Let I be the IDF of T, and K be the count of T in Q; Set the weight of T in Q: W = K * I;Set the weight of T in Q: W = K * I; Let L be the list of TokenOccurences of T from H;Let L be the list of TokenOccurences of T from H; For each TokenOccurence, O, in L:For each TokenOccurence, O, in L: Let D be the document of O, and C be the count of Let D be the document of O, and C be the count of
O O (tf of T in D);(tf of T in D);
If D is not already in R If D is not already in R (D was not previously retrieved) (D was not previously retrieved)
Then add D to R and initialize score to 0.0;Then add D to R and initialize score to 0.0; Increment D’s score by W * I * C; Increment D’s score by W * I * C; (product of T-weight (product of T-weight
in Q and D)in Q and D)
Precision and Precision and RecallRecall
documents relevant of number Total
retrieved documents relevant of Number recall
retrieved documents of number Total
retrieved documents relevant of Number precision
Relevant documents
Retrieved documents
Entire document collection
retrieved & relevant
not retrieved but relevant
retrieved & irrelevant
Not retrieved & irrelevant
retrieved not retrieved
rele
vant
irre
leva
nt
Computing Computing Recall/PrecisionRecall/Precision
R=3/6=0.5; P=3/5=0.6
R=1/6=0.167; P=1/1=1
R=2/6=0.333; P=2/3=0.667
R=6/6=1.0; p=6/14=0.429
R=4/6=0.667; P=4/8=0.5
R=5/6=0.833; P=5/9=0.556
n doc # relevant
1 588 x2 5763 589 x4 3425 590 x6 7177 9848 772 x9 321 x10 49811 11312 62813 77214 592 x
EvaluationEvaluation
0
0.2
0.4
0.6
0.8
1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Recall
Pre
cisi
on
No Position Position
oConsidering position information the system should give better performanceoThe curve closest to the upper right-hand corner of the graph indicates the best performance
