??a? ?

1
?T???? ??? ??????S?????? ??????S?????
?T??O? S???? T????O? ???S???O? ????? ?????F?????S
??? ??????????O??O? ???G????? ???????????O?
S????O? ????O?????? ??G?S??

• ??a? ?ß??d???? a??????µ?? ??a s?st?se?? ta?????
  se
• s?st?µata d?ad?ast???? t??e??as??.
• G?a??a??p????? ?a?a???ta - Spa?? ?????s?a

2
System architecture- Block diagram
Server
Subscriber Profiles/ Ratings
Media Content System
3
System architecture Server side

• Streaming Server ? open source Darwin Streaming
  Server t?? Apple pa???e? VoD ?a? Live Streaming
  ?p??es?e? 1.
• Media Content System T? s?st?µa st? ?p???
  ap????e???ta? ta a??e?a video st?? streaming
  server.
• Media Content Management System ???sd?de? ??a?
  µ??ad??? ??d??? se ???e ??a ta???a p?? e?s??eta?
  st? media content system, t? CRID (Content
  Reference IDentifier).
• Subscribers Profiles/ Ratings ??s? ded?µ????
  p?? pe????e? ta p??s?p??? st???e?a t?? ???st??
• ???µa, ep??et?, username, password, ?????a, f???,
  ep???e?µa ?a? ??a s????? ap? e?d? ta????? p??
  a??s??? st? ???st?.
• ?a? t?? ßa?µ?????e? t???
• ?? ßa?µ?????e? d????ta? e?te ?µµesa e?te ?µesa
  ap? t??? ???ste?.
• ?p?te?e?ta? ap? t??? p??a?e? users, profiles,
  ratings ?a? cbratings.

4
System architecture Server side

• Profiles Control System S?st?µa a?a?????s?? t??
  ta?t?t?ta? t?? ???st? (user authentication/sign
  in) ?a? e???af?? t?? (registration).
• Web Server Apache Tomcat web server 2
• Semantics Component ??µµ?t? t?? s?st?µat?? p??
  ep???????e? µe e??te????? p???? ?a? a?t?e?
  p????f???e? ??a t?? ta???e?. ? p??? p????f?????
  e??a? t? IMDB 3. ?a ded?µ??a p?? a?t?e? ??a
  ???e ta???a e??a?
• ? t?t???, ta e?d? t?? (genres), ? p????, ?
  p??ta????st??, t? ?t?? pa?a?????, ? s??????t??, ?
  pa?a?????, ? se?a??????f?? ?a? ? ?µ????µe??
  ???ssa.

5
System architecture Server side

• Data Set MovieLens dataset 4. ???s?µ?p??????e
  ??a ta pe???µata ?a? t?? a???????s? t??
  s?st?µat??. ?p?te?e?ta? ap? 100K ßa?µ?????e? p??
  d????a? ap? 943 ???ste? ??a 1682 ta???e?. ??
  dataset ap?te?e?ta? ap? 3 p??a?e?
• users (UserIDGenderAgeOccupationZipCode),
  
• movies (MovieIDTitleGenres) and
• ratings(UserIDMovieIDRatingTimestamp)
• Recommendation System S??d?asµ?? d?? a??????µ??
  se ??a? ?ß??d??? p?? p??te??e? ta???e? st???
  ???ste?. ? s??d?asµ?? a??????µ?? ??e? ap?de???e?
  ?t? d??e? ße?t??µ??a ap?te??sµata se s??s? µe
  a?t? p?? ?a ?d??e ???e a??????µ?? ?e????st?. ??
  a??????µ?? p?? s??d??st??a? e??a? ??
  collaborative filtering ?a? content based.
  ?p?p???? µe t?? s?st?se?? ta????? pa???eta? st?
  ???st? ?a? epe????s? ??a t? s?stas?.

6
System architecture Client side

• Proxy Server G?a ???? t?? EPG data ap? t? site
  t?? BBC e??atast????e ? Jetty 5 open source
  web server.
• Web Browser Mozilla Firefox 3.0.5 web browser
  6.
• Media Player G?a a?apa?a???? video e?s?µat????e
  st? web browser ? QuickTime Player Pro 7.5 7. ?
  client e????e? t?? media player µe t? QTJava API
  8.
• BBC ?a???e? t? p????aµµa t?? BBC TV ?a? Radio
  ??a 7 ?µ??e? 9 se TvAnytime format 10,11.

7
Content based

• ???p??????e ??a? bag-of-words naive Bayesian text
  classifier p??
• epe?t????e ?a ?e????eta? d?a??sµata ap? s?????
  ???e?? 12.
• ? ape?????s? t?? ta????? d?a?????eta? sta ped?a
  (slots) p??ta????st??, e?d?? ta???a?,
  s??????t??, pa?a?????, s????af?a?, ?????????a
  pa?a?????, ???ssa ta???a?.
• ??a ta???a µp??e? ?a ??e? pe??ss?te?e? ap? µ?a
  t?µ?? se ???sµ??a slots ? (bag-of-words).
  ??t?p??s?pe?eta? s??????? ap? ??a d????sµa ap?
  bag-of-words.
• ???ß??µa p??a??t???? ?at??????p???s?? 5 ???se?? ?
  ?p??????eta? ? p??a??t?ta µ?a? ta???a? ?a ??ße?
  ßa?µ?????a ap? 1 ??? 5.

8
Content based

• F?s? e?pa?de?s?? (??µ???s? p??f?? ???st??)
• ??µ??????a t?? bags-of-words ???? t?? slots (??e?
  ?? d??at?? ???e?? t?? slot).
• G?a ???e ???st?
• ???es? s?????? ta????? N p?? ??e? ßa?µ?????se?
  (training movies ).
• ???es? t?? p??a??t?ta? ???e ???s??
• ??a ?p??
• ???es? t?? ?p?-s?????? p??a??t?t?? ???e ?????
  t?? slot,
• ded?µ???? ?t? a???e? st?? ???s?
  ?a? st? slot
• 
  
• ?p??
  ,
• 
  p????? ???e?? s???? t?? m slot t?? ta???a?,
• 
  f???? eµf???s?? t?? ????? t?? ta???a?
  st? slot
• Smoothing t?? µ?de????? p??a??t?t?? ???e??.

9
Content based

• F?s? p??ß?e??? (???ß?e?? ßa?µ???????)
• G?a ??e? t?? ta???e? t?? ß?s??
• ???es? t?? posterior p??a??t?t?? ??a t?? ta???a ?
  ??a ??e? t?? ???se?? µe ß?s? t?? ?a???a t??
  Bayes.
• ????es? t?? ta???a? ? st?? ???s? c p??
  ?p?????st??e ? µe?a??te?? p??a??t?ta (rating
  c).
• ??µ??????a p??a?a user-ratings.

10
Content based - p??ß??µata

• ?e?????sµ??? a????s? pe??e??µ???? ?e????????ta?
  ap? ta ?a?a?t???st??? ?????sµata p?? e??a?
  s??deµ??a µe ta a?t??e?µe?a. ?? d?? d?af??et???
  a?t??e?µe?a a?t?p??s?pe???ta? ap? t? ?d?? s?????
  ?a?a?t???st???? ?????sµ?t??, e??a? ?µ??a.
• ?pe?e?d??e?s? ?p????? ?a s?st????? µ???
  a?t??e?µe?a p?? s?µe?????? ????? s??? se s??s? µe
  t? p??f?? t?? ???st? ? ? ???st?? pe??????eta? st?
  s?stas? a?t??e?µ???? pa??µ??a µe e?e??a p?? ??e?
  ?d? ßa?µ?????se?.
• Cold-Start Problem ? ???st?? p??pe? ?a
  ßa?µ?????se? ??a? ??a??p???t??? a???µ?
  a?t??e?µ???? ?ste ?a ?ata??????? ?? p??t?µ?se??
  t?? ???st?.

11
Collaborative filtering

• ?fa?µ?st??e ? Collaborative filtering with
  cluster based smoothing 13.
• ???a? ??a? memory based Pearson correlation
  coefficient
• a??????µ??. S????? a??????µ??
• Clustering ?µad?p???s? t?? ???st?? t?? dataset
  se ? ?µ?de? µe k-means 14 ?a? ???t???? t?
  s????t?s? s?s??t?s?? Pearson
• ? k-means te?µat??e? ?ta? e?a??st?p????e? ?
  a???µ?? t?? ???st?? p?? a??????? cluster a??µesa
  se d?? epa?a???e??.
• Data smoothing S?µp???????ta? ?? ßa?µ?????e?
  ???e ???st? ??a t?? ta???e? t?? dataset p?? de?
  ??e? ßa?µ?????se?. ?? ??e? ßa?µ?????e? ßas????ta?
  st?? ßa?µ?????e? p?? ????? d?se? ?? ?p????p??
  ???ste? t?? cluster t??.

12
Collaborative filtering

• ???ep????? t?? ?e?t???? ??ta?? ???e ???? ???st?
  se ??a ap? ta clusters. ?????? ?p??????eta? ?
  s?s??t?s? t?? µe ??a ta centroids. ?e????
  t?p??ete?ta? st? cluster µe t? ?p??? ??e? t?
  µe?a??te?? s?s??t?s?.
• ?p????? t?? ???t???te??? ?e?t???? ?? s????? t??
  ???t???te??? ?e?t???? e??a? ?? ? ???ste? t??
  cluster µe t??? ?p????? ? ???? ???st?? ??e? t?
  µe?a??te?? ?µ???t?ta.
• ???ß?e?? ßa?µ?????a? ???ß?e?? t?? ßa?µ?????a?
  t?? ???? ???st? ??a ??e? t?? ta???e? t?? dataset
  µe ß?s? t?? ßa?µ?????e? t?? ???t???te??? ?e?t????.

13
Collaborative filtering - p??ß??µata

• Cold-Start Problem ??µ?????e?ta? ap? t??? ?????
  ???ste? st? s?st?µa p?? de? ????? ßa?µ?????se?
  ?a???a a?t??e?µe?? ? t? s?st?µa ad??ate? ?a ß?e?
  pa??µ????? ???ste?-?e?t??e? ?a? de? µp??e? ?a
  ???e? p??ß???e??.
• First-Rater Problem ??µ?????e?ta? ap? ta ??a
  a?t??e?µe?a st? s?st?µa p?? de? ?????
  ßa?µ??????e? ap? ?a???a ???st?. ? CF e?a?t?ta?
  ap???e?st??? ap? t? ßa?µ?????a t?? ????? ???st??
  ? t? s?st?µa de? µp??e? ?a d?µ??????se? p??t?se??
  st??? ???ste? ??a a?t? ta a?t??e?µe?a µ???? ?a
  ??ß??? ??a? ??a??p???t??? a???µ? ßa?µ???????.
• Gray sheep Problem ? ???st?? s?µp?pte? µe ta
  s????a µeta?? t?? ?pa????t?? ???se?? t?? ???st??
  ? de? µp??e? ?a ta????µ??e? se ?aµ?a ???s?!
• Data Sparsity ?fe??eta? st? ?e????? ?t? ??
  ???ste? ßa?µ??????? µ??? ??a µ???? a???µ?
  a?t??e?µ????.

14
Hybrid approach Which are the problems to solve?

• ???ß??µa t?? content based e??a? ?t? p??te??e?
  µ??? pa??µ??e? ta???e? µe a?t?? p?? ? ???st??
  ??e? ?d? de?. ?p?s?? p?s?e? ap? t? cold start
  p??ß??µa.
• ?????? t?? collaborative filtering e??a? t? cold
  start, t? data sparsity ?a? t? gray sheep
  p??ß??µa.
• ??? ?a t??? s??d??s??µe ?ste ? ??a? ?a
  ?atast???e? ta p??ß??µata t?? ???????

15
Hybrid approach In which order ?

• S?µf??a µe t?? Burke 15 t? ap?d?t???te?? s??µa
  e??a? t? se???a?? (cascade hybrid).
• ? p??t?? a??????µ?? d??e? t?? p??ß???e?? t?? se
  µ?a se??? ap? d?ateta?µ??e? ???se?? ?0-?n 16.
  St?? ?d?a ???s? ta????µ???ta? ?? ta???e? p??
  e??a? t? ?d?? ?a??? p??t?se??.
• O de?te??? a??????µ?? ße?t???e? ta ap?te??sµata
  t?? p??t??.
• ?e p??a se??? ?a s??d??s??µe t??? d??
  a??????µ????

16
Hybrid approach In which order ?

• ?st? ?t? efa?µ????µe p??ta t?? collaborative
  filtering. Ta ????µe ?a a?t?µet?p?s??µe
• Data sparsity ?a?
• Cold start
• ??? a? efa?µ?s??µe p??ta t?? content based
• Cold start
• ??at? ? content based e?a?e?fe? t? data
  sparsity!!
• ?p?p???? t?p??et?saµe st? f??µa e???af?? t??
  ???st? µ?a ??sta ap? e?d? ta????? p?? µp??e? ?a
  ßa?µ?????se?. ?ts? a?t?µet?p?saµe ?a? t? cold
  start p??ß??µa.

17
Explanations Why?

• G?a ?a ??????e? ? ???st?? t? ???? p?? t??
  p??t????e ? ta???a.
• G?a ?a eµp?ste?te? t?? p??t?se?? t?? s?st?µat??.
• G?a ?a ?ata??ße? ?a??te?a t?? t??p? p??
  ?e?t????e? t? s?st?µa.
• G?a ?a µ? ???e? t? ????? t?? µe ta???e? p?? de?
  ?a t?? a??s???.

18
Explanations Content based

• ???a? ta features t?? ta???a? p?? t?? ?a??st???
  pa??µ??a µe t?? ???e? ta???e? p?? ? ???st?? ??e?
  a???????se? ?et???.
• G?a t?? ta???a p?? p??te??e ? a??????µ??
  s????????ta? ??e? ?? ?p?-s?????? p??a??t?te? t??
  ???e?? t?? ta???a?.
• ?? ???e slot t?? ta???a? µp??e? ?a ??e?
  pe??ss?te?e? ap? µ?a t?µ?? (bags-of-words).
• ?? feature t?? slot µe t?µ? t? µe?a??te??
  ?p?-s?????? p??a??t?ta e??a? t? ?s???? feature
  t?? ta???a? (µe t? µe?a??te?? d?a????st???
  ??a??t?ta) ?a? ap?te?e? t? ???? ??a t?? ?p???
  p??t????e ? s???e???µ??? ta???a.

19
Explanations Collaborative Filtering

• ???a? t? p?s?st? t?? ???t???te??? ?e?t???? p??
  ????? a???????se? ?et??? t?? ta???a
  Explanations Ratio (ER)
• ??????µe ?at?f?? t?? ?at?te?? ßa?µ?????a p??
  ?e??e?ta? ?et??? a???????s?.
• St? p?s?st? t?? ???t???te??? ?e?t????
  p??sµet????ta? ?s?? ????? ßa?µ?????se? t?? ta???a
  p??? ap? t? ?at?f??. G?a 1ltiltK, a?
  PositiveCounter PositiveCounter1

20
Experiments Content based

• ?? a????? dataset p?? ???s?µ?p??????e pe????e?
  943 ???ste? ?a? 1682 ta???e?.
• G?a t?? e?pa?de?s? t?? ??a ???e ???st?
  ???s?µ?p??????e t? 50 t?? ta????? p?? ??e?
  ßa?µ?????se? ? ???st??.
• St???? e??a? ? e??es? t?? dataset p?? ?a d??e?
  ?a??te?a ap?te??sµata st?? Collaborative
  Filtering a??????µ? ?a? ?at ep??tas? st??
  ?ß??d???.
• G?a ???e ???st? t? ??? µet?????e ??
• µe t? µ??e??? t?? test set.

21
Experiments Content based

• ?e??aµa 1 ?p?d?s? CB a?????a µe t? s????? t??
  ßa?µ??????? a?? ???st?.

22
Experiments Content based

• ?e??aµa 2 ?p?d?s? CB a?????a µe ta features p??
  ???s?µ?p?????ta? ??a ?a pe????????? µ?a ta???a.

23
Experiments Content based

• ?e??aµa 3 ?p?d?s? CB a?????a µe t? p????? t??
  ßa?µ??????? a?? ta???a (µe user_ratings / user
  gt 40).
• ??att??eta? p???
• t? p????? t??
• ratings! -gt ?
• CF de? µp??e?
• ?a e???e?
• ap?te??sµata!

24
Experiments Content based

• ?e??aµa 4 ?p?d?s? CB ??t??ta? ???t???a sta
  features t?? ta????? (µe user_ratings / user gt
  40).
• ??att????ta? p???
• t? p????? ta?????
• ?a? ???st?? -gt ?
• CF de? µp??e?
• ?a e???e?
• ap?te??sµata!

25
Experiments Collaborative filtering

• G?a ?a µp??e? ?a d?se? p??ß???e??, sta pe???µata
  s?µµete??a? ?? ???ste? µe pe??ss?te?e? ap? 40
  ßa?µ?????e? 622 ap? t??? 943.
• ?? 200 p??t?? ???ste? ???s?µ?p?????ta? p??ta ??a
  training ?a? ??a p?s?st? ap? t??? te?e?ta???? ??a
  evaluation.
• ?p? t??? evaluation users ??f???e ?p ??? ??a
  µ???? t?? ßa?µ??????? t??? (Evaluation Ratings
  Per User - ERPU).

26
Experiments Collaborative filtering

• 1? set pe??aµ?t?? ??a d?af??et???? t?µ?? t??
  pa?aµ?t??? ? (0lt?lt1).
• ? 0 ? collaborative filtering ???s?µ?p??e? t??
  ßa?µ?????e? t?? ???t???te??? ?e?t???? ??a
  p??ß???e??.
• ? 1 ???s?µ?p??e? t?? µ?se? t?µ?? t??
  ßa?µ??????? t?? ???t???te??? ?e?t????.

27
Experiments Collaborative filtering

• 2? set pe??aµ?t?? ??a d?af??et???? t?µ?? t?? ERPU
  (5, 7, 10, 12, 15, 17 ?a? 20 ).

28
Experiments - Hybrid

• St? 1? set pe??aµ?t?? ?µ??a µe t? 1? set
  pe??aµ?t?? collaborative filtering a????aµe t??
  t?µ?? t?? pa?aµ?t??? ?.
• ?? 2? set pe??aµ?t?? e??a? ?d?? µe t? 2? set t??
  collaborative filtering. ?????aµe t?? t?µ?? t??
  ERPU.
• ?? 3? set pe??aµ?t?? e??a? ?d?? µe t? 3? set
  pe??aµ?t?? t?? content based.
• ??? ?a ep??easte? ? ?ß??d???? ap? t?? a??a?????

29
Experiments - Hybrid

• ? ?ß??d???? de? ep??e??eta? ap? t?? a??a??? st??
  collaborative filtering!!

30
Experiments - Hybrid

• ?a??te?? ep?d?s? ??a ERPU5.

31
Experiments - Hybrid

• ? ?ß??d???? de? ep??e??eta? ap? t?? a??a??? st??
  content based!!

32
Hybrid vs Collaborative filtering

• ? ?ß??d???? ??e? ?a??te?? ap?d?s? ?a? p??
  sta?e?? p??e?a.
• ??s? MAE(colfilt)1,12 ??s? ???(hybrid)0,89
• ??p??? ap????s?(colfilt)0,014 ??p???
  ap????s?(hybrid)0,067

33
References

• http//developer.apple.com/opensource/server/strea
  ming/index.html
• http//tomcat.apache.org/
• http//www.imdb.com
• http//www.grouplens.org/node/73attachments
• http//www.mortbay.org/jetty/
• http//www.mozilla.com/en-US/firefox/firefox.html
• http//www.apple.com/quicktime/
• http//developer.apple.com/quicktime/qtjava/
• http//backstage.bbc.co.uk/data/7DayListingData?v
  16wk
• http//www.tv-anytime.org/
• http//www.bbc.co.uk/opensource/projects/tv_anytim
  e_api/
• Mooney, R. J., P. N. Bennett, and L. Roy. Book
  recommending using text categorization with
  extracted information. In Recommender Systems.
  Papers from 1998 Workshop. Technical Report
  WS-98-08. AAAI Press, 1998.
• Gui-Rong Xue, Chenxi Lin, Qiang Yang, WenSi Xi,
  Hua-Jun Zeng , Yong Yu and Zheng Chen, Scalable
  Collaborative filtering Using Cluster-based
  smoothing . In Proceedings of the 2005 ACM SIGIR
  Conference, Salvador, Brazil, 2005, pp. 114-121
• http//www.clustan.com/k-means_critique.html
• Robin Burke. Hybrid Recommender Systems Survey
  and Experiments. California State University,
  Fullerton 2002.
• Robin Burke. Integrating Knowledge-based and
  Collaborative-filtering Recommender Systems. In
  Workshop on AI and Electronic Commerce, AAAI
  1999.