Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Stream reasoning: mastering the velocity and the variety dimensions of Big Data at once

1,432 views

Published on

More and more applications require real-time processing of heterogeneous data streams. In terms of the “Vs” of Big Data (volume, velocity, variety and veracity), they require addressing velocity and variety at the same time. Big Data solutions able to handle separately velocity and variety have been around for a while, but only Stream Reasoning approaches those two dimensions at once. Current results in the Stream Reasoning field are relevant for application areas that require to: handle massive datasets, process data streams on the fly, cope with heterogeneous incomplete and noisy data, provide reactive answers, support fine-grained information access, and integrate complex domain models. This talk starting from those requirements, frames the problem addressed by Stream Reasoning. It poses the research question and operationalise it with four simpler sub-questions. It describes how the database group of Politecnico di Milano positively answered those sub-questions in the last 7 years of research. It briefly surveys alternative approaches investigated by other research groups world wide and it elaborates on current limitations and open challenges.

Published in: Internet
  • Be the first to comment

Stream reasoning: mastering the velocity and the variety dimensions of Big Data at once

  1. 1. Stream  Reasoning:   mastering  the  velocity  and  the  variety     dimensions  of  Big  Data  at  once   Emanuele  Della  Valle   DEIB  -­‐  Politecnico  di  Milano   @manudellavalle   emanuele.dellavalle@polimi.it   hBp://emanueledellavalle.org     University  of  Olso,  Norway  -­‐    3.11.2015  
  2. 2. It's  a  streaming  world  …   •  Off-­‐shore  oil  operaQons   •  Smart  CiQes   •  Global  Contact  Center   •  Social  networks   •  Generate  data  streams!   E.  Della  Valle,  S.  Ceri,  F.  van  Harmelen,  D.  Fensel  It's  a  Streaming  World!  Reasoning  upon   Rapidly  Changing  Informa:on.  IEEE  Intelligent  Systems  24(6):  83-­‐89  (2009)   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   2
  3. 3. …  looking  for  reacQve  answers  …   •  What  is  the  expected  Qme  to  failure  when  that   turbine's  barring  starts  to  vibrate  as     detected  in  the  last  10  minutes?     •  Is  public  transportaQon   where  the  people  are?     •  Who  are  the  best  available  agents  to     route  all  these  unexpected  contacts     about  the  tariff  plan  launched  yesterday?     •  Who  is  driving  the  discussion     about  the  top  10  emerging  topics  ?     •  Require  conQnuous  processing     and  reacQve  answer   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   3
  4. 4. …with  conflicQng  requirements  1/8   A  system  able  to  answer  those  queries  must  be  able  to     •  handle  massive  datasets   –  A  typical  oil  producQon  plaeorm  is  equipped     with  about  400.000  sensors   –  Telecom  data  is  the  most  pervasive  data   source  in  urban  are,  in  Milano  there  are   1.8  million  mobile  users   –  A  global  contact  centre  of  a  Telecom     operator  counts  500  millions  of  clients     –  Facebook  alone  has  1.1  billion     of  acQve  users         UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   4
  5. 5. …with  conflicQng  requirements  2/8   A  system  able  to  answer  those  queries  must  be  able  to     •  process  data  streams  on  the  fly     –  The  sensors  on  typical  oil  producQon     plaeorm  generates  10,000  observaQons   per  minute  with  peaks  of  100,000  o/m   –  The  mobile  users  in  Milano  generates   20,000  call/sms/data  connecQons   per  minute  with  peaks  of  80,000  c/m   –  A  global  contact  centre  receives   10,000  contacts  per  minute  with   peaks  of  30,000  c/m   –  Facebook,  as  of  May  2013,  observes   3  millions  "I  like"  per  minute       UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   5
  6. 6. …with  conflicQng  requirements  3/8   A  system  able  to  answer  those  queries  must  be  able  to     •  cope  with  heterogeneous  dataset       –  The  sensors  on  typical  oil  producQon   have  been  deployed  over  10  years   by  10s  of  different  producers     –  Tens  of  data  sources  are  normally   needed  to  make  sense  of  an  urban   phenomena   –  A  global  contact  centre  consists  in  100s   of  offices  owned  by  different  subsidiary     companies  engaged  yearly   –  Each  social  network  has  its  own   data  model,  APIs,  …       UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   6
  7. 7. …with  conflicQng  requirements  4/8   A  system  able  to  answer  those  queries  must  be  able  to     •  cope  with  incomplete  data         –  10s  of  sensors  and  networking  links     broke  down  daily     –  Coverage  is  incomplete       –  Only  standard  cases  are  covered  by   fully  machine  processable  data  records   100s  of  contacts  per  minute  are     manage  ad-­‐hoc   –  Conversa:ons  happen  outside  the   social  networks,  too!   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   7
  8. 8. …with  conflicQng  requirements  5/8   A  system  able  to  answer  those  queries  must  be  able  to     •  cope  with  noisy  data           –  Sensor  out-­‐of-­‐opera:ng  range         –  Faulty  sensors       –  Agents  misunderstand,  get  :red,  …       –   Irony,  sarcasm,  …   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   8
  9. 9. …with  conflicQng  requirements  6/8   A  system  able  to  answer  those  queries  must  be  able  to     •  provide  reac:ve  answers             –  detecQon  of  dangerous  situaQons     must  occur  within  minutes       –  recommendaQons  to  ciQzens  must   be  performed  in  few  seconds     –  rouQng  a  contact  through  each  step  of     the  decision  tree  must  take  less  than  a   second   –  Search  autocompleQng  may  need   to  be  updated  every  few  minutes     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   9
  10. 10. …with  conflicQng  requirements  7/8   A  system  able  to  answer  those  queries  must  be  able  to     •  support  fine-­‐grained  informa:on  access               –  IdenQfy  a  turbine  among  thousands       –  Locate  a  bus  among  thousands       –  Contact  an  agent  among  thousands       –  IdenQfy  an  opinion  maker  among   thousands  of  influencers  for  a  topic   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   10
  11. 11. …with  conflicQng  requirements  8/8   A  system  able  to  answer  those  queries  must  be  able  to     •  integrate  complex  domain  models  of               –  opera:onal  and  control  process         –  various  city  aspects       –  contact  management,  contract  types,     agent  skills,  contactor  profiles,  …       –  topics,  user  profiles,  …   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   11
  12. 12. Challenges   A  system  able  to  answer  those  queries  must  be  able  to     •  handle  massive  datasets              x     •  process  data  streams  on  the  fly            x     •  cope  with  heterogeneous  datasets            x     •  cope  with  incomplete  data                  x  x           •  cope  with  noisy  data                        x           •  provide  reac:ve  answers                x             •  support  fine-­‐grained  access              x        x               •  integrate  complex  domain  models              x     Volume' Velocity' Variety' Veracity' In Big Data terms UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   12
  13. 13. Grand  challenge   •  Volume  +  Velocity  +  Variety  =  hard  deal   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   Volume months days hours min. sec. ms. velocity ZB EB PB TB GB MB KB Variety 13
  14. 14. A  good  reason  to  embrace  it!   •  ++  Variety  à  ++  value     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   Value ms. sec. min. hours days months years velocity Variety 14
  15. 15. From  challenges  to  opportuniQes   •  Formally  data  streams  are  :     –  unbounded  sequences  of  Qme-­‐varying  data  elements   •  Less  formally,  in  many  applicaQon  domains,  they  are:     –  a  “conQnuous”  flow  of  informaQon     –  where  recent  informa:on  is  more  relevant  as  it  describes  the   current  state  of  a  dynamic  system   •  OpportuniQes   –  Forget  old  enough  informa:on   –  Exploit  the  implicit  ordering  (by  recency)  in  the  data     time UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   15
  16. 16. State-­‐of-­‐the-­‐art:  DSMS  and  CEP     •  A  paradigma:c  change!   •  ConQnuous  queries  registered  over  streams  that   are  observed  trough  windows     window input streams streams of answerRegistered   ConQnuous   Query   Dynamic   System UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   16
  17. 17. DSMS  and  CEP  vs.  requirements   Requirement DSMS CEP massive datasets data streams heterogeneous dataset incomplete data noisy data reactive answers fine-grained information access complex domain models ✗ ✗ ✗ UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   17
  18. 18. State of the art: OBDA   •  Given  ontology  O  and  query  Q,  use  O  to  rewrite  Q   as  Q’  so  that,  for  any  set  of  ground  facts  A  contained  in  mulQple   databases:   –  answer(Q,O,A)  =  answer(Q’,!,A)   The  answer  of  the  query  Q  using  the  ontology  O  for  any  set  of  ground  facts  A   is  equal  to  answer  of  a  query  Q’  without  considering  the  ontology  O     •  Use  mapping  M  to  map  Q’  to  mulQple  SQL  queries  to  the  various   databases   Rewrite O Q Q’ Map SQL M answer A UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   18
  19. 19. DSMS/CEP,OBDA  vs.  requirements   Requirement DSMS CEP OBDA massive datasets data streams heterogeneous dataset incomplete data noisy data reactive answers fine-grained information access complex domain models ✗ ✗ ✗ ✗ ✗ ✗ UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   19
  20. 20. Stream  Reasoning   •  Research  quesQon   –  is  it  possible  to  make  sense  in  real  :me  of     mul:ple,  heterogeneous,  gigan:c  and  inevitably  noisy  and   incomplete  data  streams  in  order  to  support  the  decision   processes  of  extremely  large  numbers  of  concurrent  users?   •  Proposed  approach     Complexity   Raw  Stream  Processing   SemanQc  Streams   DL-­‐Lite   DL  AbstracQon   SelecQon   InterpretaQon   Reasoning   Querying   Re-­‐wriQng   Change  Frequency   PTIME   NEXPTIME   104  Hz   1  Hz     Complexity  vs.  Dynamics     AC0   H.  Stuckenschmidt,  S.  Ceri,  E.  Della  Valle,  F.  van  Harmelen:  Towards  Expressive  Stream  Reasoning.  Proceedings   of  the  Dagstuhl  Seminar  on  SemanQc  Aspects  of  Sensor  Networks,  2010.     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   20
  21. 21. Sub-­‐research  quesQons   1.  Is  it  possible  extend  the  Seman:c  Web  stack     in  order  to  represent  heterogeneous  data  streams,   conQnuous  queries,  and  conQnuous  reasoning  tasks?   2.  Does  the  ordered  nature  of  data  streams  and  the   possibility  to  forget  old  enough  informaQon  allow  to   op:mize  con:nuous  querying  and  con:nuous  reasoning   tasks  so  to  provide  reac:ve  answers  to  large  number  of   concurrent  users  without  forsaking  correctness  or   completeness?     3.  Can  SemanQc  Web  and  Machine  Learning  technologies  be   jointly  employed  to  cope  with  the  noisy  and  incomplete   nature  of  data  streams?   4.  Are  there  prac:cal  cases  where  processing  data  stream  at   semanQc  level  is  the  best  choice?     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   21
  22. 22. Sub-­‐research  quesQons   1.  Is  it  possible  extend  the  Seman:c  Web  stack     in  order  to  represent  heterogeneous  data  streams,   conQnuous  queries,  and  conQnuous  reasoning  tasks?   2.  Does  the  ordered  nature  of  data  streams  and  the   possibility  to  forget  old  enough  informaQon  allow  to   op:mize  con:nuous  querying  and  con:nuous  reasoning   tasks  so  to  provide  reac:ve  answers  to  large  number  of   concurrent  users  without  forsaking  correctness  or   completeness?     3.  Can  SemanQc  Web  and  Machine  Learning  technologies  be   jointly  employed  to  cope  with  the  noisy  and  incomplete   nature  of  data  streams?   4.  Are  there  prac:cal  cases  where  processing  data  stream  at   semanQc  level  is  the  best  choice?     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   22
  23. 23. State-­‐of-­‐the-­‐art:  RDF  model   •  RDF:  Resource  DescripQon  Framework   –  It  allows  to  make  statements  about  resources  in  the  form   of  subject-­‐predicate-­‐object  expressions   •  In  RDF  terminology  triples   •  E.g.                @BarakObama                posts                    "Four  more  years"       –  A  collecQon  of  RDF  statements  represents  a  labelled,   directed  graph   •  In  RDF  terminology  a  graph   •  E.g.,  the  tweet  above  by  Barak  Obama  is  connected  to   –  800,000+  twiBer  user  profiles  via  retweets   –  300,000+  twiBer  user  profiles  favorite   –  …   subject predicate object UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   23
  24. 24. ContribuQon:  RDF  stream  Models       •  RDF  Stream  (the  C-­‐SPARQL  way)   –  Unbound  sequence  of  :me-­‐varying  triples   –  each  represented  by  a  pair  made  of  an  RDF  triple  and  its   Qmestamp   –  Timestamp  are  non-­‐decreasing  (allowing  for  simultaneity)            …    @BarakObama                posts              "Four  more  years",                              8:16PM  6  Nov  2012    @Alice                                      posts              "RT:  Four  more  years",                  8:17PM  6  Nov  2012            …     D.F.  Barbieri,  D.  Braga,  S.  Ceri,  E.  Della  Valle,  M.  Grossniklaus:  Querying  RDF  streams  with     C-­‐SPARQL.  SIGMOD  Record  39(1):  20-­‐26  (2010)     subject predicate object timestamp UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   24
  25. 25. ContribuQon:  RDF  stream  Models     •  RDF  Stream  (the  Streaming  Linked  Data  way)   –  Unbound  sequence  of  :me-­‐varying  graphs   –  each  represented  by  a  pair  made  of  an  RDF  graph  and  its   Qmestamp     –  Timestamps  (if  present)  are  monotonically  increasing   –  Graphs  act  as  a  form  of  punctuaQon  (all  triples  in  a  graph  are   simultaneous)     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   D.F.  Barbieri,  E.  Della  Valle:  A  Proposal  for  Publishing  Data  Streams  as  Linked  Data  -­‐  A   Posi:on  Paper.  LDOW  (2010)     25
  26. 26. RDF  streams  Qme  semanQcs  1/3   •  A  RDF  stream  without  Qmestamp  is  an  ordered  sequence   of  data  items   •  The  order  can  be  exploited  to  perform  queries   –  Does  Alice  meet  Bob  before  Carl?   –  Who  does  Carl  meet  first?   S   e1   :alice  :isWith  :bob   e2   :alice  :isWith  :carl   e3   :bob  :isWith  :diana   e4   :diana  :isWith  :carl   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   26
  27. 27. RDF  streams  Qme  semanQcs  2/3   •  One  Qmestamp:  the  Qme  instant  on  which  the  data  item   occurs   •  We  can  start  to  compose  queries  taking  into  account  the   Qme   –  How  many  people  has  Alice  met  in  the  last  5m?   –  Does  Diana  meet  Bob  and  then  Carl  within  5m?   e1   e2   e3   e4  S   t  3   6   9  1   :alice  :isWith  :bob   :alice  :isWith  :carl   :bob  :isWith  :diana   :diana  :isWith  :carl   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   27
  28. 28. RDF  streams  Qme  semanQcs  3/3   •  Two  Qmestamps:  the  Qme  range  on  which  the  data  item   is  valid  (from,  to]   •  It  is  possible  to  write  even  more  complex  constraints:   –  Which  are  the  meeQngs  the  last  less  than  5m?   –  Which  are  the  meeQngs  with  conflicts?   .   S   t  3   6   9  1   :alice  :isWith  :bob   :alice  :isWith  :carl   :bob  :isWith  :diana   :diana  :isWith  :carl   e1 e2 e3 e4 D.  Anicic,  P.  Fodor,  S.  Rudolph,  &  N.  Stojanovic.  EP-­‐SPARQL:  a  unified  language  for  event   processing  and  stream  reasoning.  In  WWW  2011,  pages  635–644   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   28
  29. 29. Finding   •  The  Seman:c  Web  stack  can  be  extended  so  to   incorporate  streaming  data  as  a  first  class  ciQzen   –  RDF  stream  data  model(s)   –  Con:nuous  SPARQL  syntax  and  semanQcs   –  Con:nuous  deduc:ve  reasoning  semanQcs             UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   29
  30. 30. Work  in  progress   •  In  2013,  an  RDF  Stream  Processing  (RSP)   community  group  was  created  at  W3C   hBp://www.w3.org/community/rsp/     •  RSP  data  model  and  serializaQon   – hBps://github.com/streamreasoning/RSP-­‐QL/blob/ master/SerializaQon.md     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   30
  31. 31. State-­‐of-­‐the-­‐art:  SPARQL   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   31
  32. 32. ContribuQon:  ConQnuous-­‐SPARQL   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   32
  33. 33. ContribuQon:  ConQnuous-­‐SPARQL Who  are  the  opinion  makers?  i.e.,  the  users  who  are   likely  to  influence  the  behavior  their  followers   REGISTER STREAM OpinionMakers COMPUTED EVERY 5m AS CONSTRUCT { ?opinionMaker sd:about ?resource } FROM STREAM <http://…> [RANGE 30m STEP 5m] WHERE { ?opinionMaker ?opinion ?res . ?follower sioc:follows ?opinionMaker. ?follower ?opinion ?res. FILTER (cs:timestamp(?follower ?opinion ?res) > cs:timestamp(?opinionMaker ?opinion ?res) ) } HAVING ( COUNT(DISTINCT ?follower) > 3 ) SR  2015,  Austria  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   33
  34. 34. ContribuQon:  ConQnuous-­‐SPARQL Who  are  the  opinion  makers?  i.e.,  the  users  who  are   likely  to  influence  the  behavior  their  followers   REGISTER STREAM OpinionMakers COMPUTED EVERY 5m AS CONSTRUCT { ?opinionMaker sd:about ?resource } FROM STREAM <http://…> [RANGE 30m STEP 5m] WHERE { ?opinionMaker ?opinion ?res . ?follower sioc:follows ?opinionMaker. ?follower ?opinion ?res. FILTER (cs:timestamp(?follower ?opinion ?res) > cs:timestamp(?opinionMaker ?opinion ?res) ) } HAVING ( COUNT(DISTINCT ?follower) > 3 ) Query  registra:on   (for  con:nuous  execu:on)   FROM  STREAM  clause   WINDOW   RDF  Stream  added  as     new  ouput  format       Buil:n  to  access   :mestamps   SR  2015,  Austria  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   D.F.  Barbieri,  D.  Braga,  S.  Ceri,  E.  Della  Valle,  M.  Grossniklaus:  Querying  RDF  streams  with     C-­‐SPARQL.  SIGMOD  Record  39(1):  20-­‐26  (2010)     34
  35. 35. Finding   •  The  Seman:c  Web  stack  can  be  extended  so  to   incorporate  streaming  data  as  a  first  class  ciQzen   –  RDF  stream  data  model   –  Con:nuous  SPARQL  syntax  and  semanQcs   –  Con:nuous  deduc:ve  reasoning  semanQcs           UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   35
  36. 36. AlternaQves  to  C-­‐SPARQL   •  CQELS   –  What:  STREAM  clause,  focus  on  new  answer     –  Ref:  Le-­‐Phuoc,  D.,  Dao-­‐Tran,  M.,  Xavier  Parreira,  J.,  &  Hauswirth,  M.     A  naQve  and  adapQve  approach  for  unified  processing  of  linked  streams  and   linked  data.  In  ISWC  2011,  pages  370–388.     •  SPARQLStream   –  What:  window  in  the  past,  focus  on  RDF  to  Stream  operators   –  Ref:  Calbimonte,  J.-­‐P.,  Corcho,  O.,  &  Gray,  A.  J.  G.  Enabling  ontology-­‐based   access  to  streaming  data  sources.  In  ISWC,  2010,  pages  96–111.     •  EP-­‐SPARQL   –  What:  focus  on  event  specific  operators   –  Ref:  Anicic,  D.,  Fodor,  P.,  Rudolph,  S.,  &  Stojanovic,  N.  EP-­‐SPARQL:  a  unified   language  for  event  processing  and  stream  reasoning.  In  WWW  2011,  pages   635–644.     •  TEF-­‐SPARQL   –  What:  adds  "facts"  as  first  class  elements     –  Ref:  hBps://www.merlin.uzh.ch/publicaQon/show/8467      UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   36
  37. 37. AlternaQves  to  C-­‐SPARQL   •  Comparison  between  exisQng  approaches   System   S2R   R2R   Time-­‐aware   R2S   C-­‐SPARQL  Engine   Logical  and   triple-­‐based   SPARQL  1.1   query   Qmestamp  funcQon   Batch  only   Streaming  Linked   Data  Framework   Logical  and   graph-­‐based   SPARQL  1.1   no   Batch  only   SPARQLstream   Logical  and   triple-­‐based   SPARQL  1.1   query   no   Ins,  batch,  del   CQELS   Logical  and   triple-­‐based   SPARQL  1.1   query   no   Ins  only   TEF-­‐SPARQL   no   SPARQL-­‐like   Temporarily  Facts,   BEFORE  SINCE,  UNTIL,   DURING,     Batch  only   EP-­‐SPARQL   no   SPARQL  1.0   SEQ,  PAR,  AND,  OR,   DURING,  STARTS,   EQUALS,  NOT,  MEETS,   FINISHES   Ins  only   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   37
  38. 38. Work  in  progress  at  RSP@W3C   •  RSP-­‐QL   –  Syntax   •  hBps://github.com/streamreasoning/RSP-­‐QL/blob/master/RSP-­‐ QL%20Sample%20Queries.md     –  Proposed  semanQcs   •  D.Dell'Aglio,  E.Della  Valle,  J.-­‐P.Calbimonte,  Ó.  Corcho:  RSP-­‐QL   SemanQcs:  A  Unifying  Query  Model  to  Explain  Heterogeneity  of   RDF  Stream  Processing  Systems.  Int.  J.  SemanQc  Web  Inf.  Syst.   10(4):  17-­‐44  (2014)   –  SemanQcs  (work  in  progress)   •  hBps://github.com/streamreasoning/RSP-­‐QL/blob/master/ SemanQcs.md     –  Quick  ref.   •  D.  Dell'Aglio,  J.-­‐P.  Calbimonte,  E.  Della  Valle,  Ó.  Corcho:  Towards   a  Unified  Language  for  RDF  Stream  Query  Processing.  ESWC   (Satellite  Events)  2015:  353-­‐363   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   38
  39. 39. ContribuQon:     conQnuous  deducQve  reasoning     •  DL  Ontology  Stream  ST   – A  ontology  stream  with  respect  to  a  staQc  Tbox  T  is  a   sequence  of  Abox  axioms  ST(i)   •  A  Windowed  Ontology  Stream  ST(o,c]   – A  windowed  ontology  stream  with  respect  to  a  staQc   Tbox  T  is  the  union  of  the  Abox  axioms  ST(i)  where   o<i≤c   •  Reasoning  on  a  Windowed  Ontology  Stream   ST(o,c]  is  as  reasoning  on  a  staQc  DL  KB   SR  2015,  Austria  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   39 Emanuele  Della  Valle,  Stefano  Ceri,  Davide  Francesco  Barbieri,  Daniele  Braga,  Alessandro   Campi:  A  First  Step  Towards  Stream  Reasoning.  FIS  2008:  72-­‐81    
  40. 40. discusses   discusses   discusses   discusses   discusses   discusses   discusses   Example  of     conQnuous  deducQve  reasoning   What impact has been my micropost p1 creating in the last hour? Let’s count the number of microposts that discuss it … REGISTER STREAM ImpactMeter AS SELECT (count(?p) AS ?impact) FROM STREAM <http://…/fb> [RANGE 60m STEP 10m] WHERE { :Alice posts [ sr:discusses ?p ] } p1   p3   p5   p8   p2   p4   p7   p6   7! Transitive property Alice posts p1 . UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   40
  41. 41. Finding   •  The  Seman:c  Web  stack  can  be  extended  so  to   incorporate  streaming  data  as  a  first  class  ciQzen   –  RDF  stream  data  model   –  Con:nuous  SPARQL  syntax  and  semanQcs   –  Con:nuous  deduc:ve  reasoning  semanQcs           UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   41
  42. 42. AlternaQves  to  conQnuous  deducQve     (RDFS++)  reasoning     •  ETALIS   –  What:  RDFS  +  Allen  Algebra   –  Ref:  Anicic,  D.,  Rudolph,  S.,  Fodor,  P.,  &  Stojanovic,  N.  Stream  reasoning  and   complex  event  processing  in  ETALIS.  SemanQc  Web,  3(4),  2012,    397–407.     •  STARQL   –  What:     •  DL-­‐Lite  +  ConjuncQve  Query  +  Qme-­‐series   •  SHI  +  Grounded  ConjuncQve  Queries  +  Qme-­‐series   –  Ref:  ÖL  Özçep,  R  Möller.  Ontology  Based  Data  Access  on  Temporal  and   Streaming  Data.  Reasoning  Web,  2014   •  ASP-­‐based   –  What:  Qme-­‐decaying  ASP   –  Ref:  hBp://arxiv.org/abs/1301.1392   •  LARS   –  What:  high-­‐level  unified  formal  foundaQon  for  stream  reasoning     –  Ref:  H.  Beck,  M.  Dao-­‐Tran,  T.  Eiter,  M.  Fink:  LARS:  A  Logic-­‐Based  Framework   for  Analyzing  Reasoning  over  Streams.  AAAI  2015:  1431-­‐1438H.     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   42
  43. 43. Sub-­‐research  quesQons   1.  Is  it  possible  extend  the  Seman:c  Web  stack     in  order  to  represent  heterogeneous  data  streams,   conQnuous  queries,  and  conQnuous  reasoning  tasks?   2.  Does  the  ordered  nature  of  data  streams  and  the   possibility  to  forget  old  enough  informaQon  allow  to   op:mize  con:nuous  querying  and  con:nuous  reasoning   tasks  so  to  provide  reac:ve  answers  to  large  number  of   concurrent  users  without  forsaking  correctness  or   completeness?     3.  Can  SemanQc  Web  and  Machine  Learning  technologies  be   jointly  employed  to  cope  with  the  noisy  and  incomplete   nature  of  data  streams?   4.  Are  there  prac:cal  cases  where  processing  data  stream  at   semanQc  level  is  the  best  choice?     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   43
  44. 44. ContribuQon:  opQmize  querying     for  reacQve  answers   •  C-­‐SPARQL  engine  Qme  window-­‐based  selecQon  outperforms                            SPARQL  filter-­‐based  selecQon  (Jena-­‐ARQ)   D.  Barbieri,  D.  Braga,  S.  Ceri,  E.  Della  Valle,  Y.  Huang,  V.  Tresp,  A.Re•nger,  H.  Wermser:   Deduc:ve  and  Induc:ve  Stream  Reasoning  for  Seman:c  Social  Media  Analy:cs     IEEE  Intelligent  Systems,  30  Aug.  2010.   Our In-memory RDF stream processing engine UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   44
  45. 45. Finding   •  Stream  Reasoning  task  is  feasible  and  the  very  nature  of   streaming  data  offers  opportuniQes  to  op:mise   reasoning  tasks  where  data  is  ordered  by  recency  and   can  be  forgoBen  a€er  a  while   –  C-­‐SPARQL  Engine  prototype   –  IMaRS  conQnuous  incremental  reasoning  algorithm   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   45
  46. 46. Work  in  progress   •  When  volumes  also  maBers  …   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   46 Join   Data  Stream   SPARQL  endpoint   Window   Maintenance   Policy   Local   View   RSP  engine   Web   Soheila  Dehghanzadeh,  Daniele  Dell'Aglio,  Shen  Gao,  Emanuele  Della  Valle,  Alessandra   Mileo,  Abraham  Bernstein:  Approximate  Con:nuous  Query  Answering  over  Streams  and   Dynamic  Linked  Data  Sets.  ICWE  2015:  307-­‐325  
  47. 47. State-­‐of-­‐the-­‐art     deducQve  reasoning   •  Data-­‐driven  (a.k.a.  forward  reasoning)     •  Query-­‐driven  –  backward  reasoning   •  Query-­‐driven  –  query  rewriQng  (a.k.a.  ontology  based  data  access)   Reasoner   RDFd ata   SPARQL   Inferred   data   ontology   SPARQL   ontology   RewriBen   query   Reasoner   Reasoner   RDFd ata   SPARQL   ontology   data   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   47
  48. 48. Naïve  approaches  to  Stream  Reasoning   windowing  then  reasoning   •  Data-­‐driven  (a.k.a.  forward  reasoning)   •  Query-­‐driven  –  backward  reasoning   •  Query-­‐driven  –  query  rewriQng  (a.k.a.  ontology  based  data  access)   Reasoner   RDF   data   SPARQL   Inferred   data   ontology   ontology   RewriBen   query   Reasoner   Reasoner   RDF   data   ontology   Window   Window   Window   SPARQL   SPARQL  data   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   48
  49. 49. Not  so  naïve  approach  to   stream  reasoning   •  The  problem  is  that  materializaQon  (the  result  of  data-­‐driven   processing)  are  very  difficult  to  decrement  efficiently.   –  State-­‐of-­‐the-­‐art:  DRed  algorithm   •  Over  delete   •  Re-­‐derive   •  Insert   Reasoner   Inferred   data   ontology   window   inserQons   deleQons   Incremental  !!!   SPARQL   Y.  Ren,  J.  Z.  Pan.  OpQmising  ontology  stream  reasoning  with  truth  maintenance  system.   In  CIKM  (2011)   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   49
  50. 50. Is  DRed  needed?   •  DRed  works  with  random  inserQons  and  deleQons   •  In  a  streaming  sedng,  when  a  triple  enters  the  window,     given  the  size  of  the  window,  the  reasoner  knows  already     when  it  will  be  deleted!   •  E.g.,     –  if  the  window  is  40  minutes   long,  and,     –  it  is  10:00,  the  triple(s)     entering  now   –  will  exit  on  10:40.   •  Conclusion   –  dele:ons  are  predictable   Time Enter window Exit window Explicitly in window Infer win 10:00 A!B 10:10 B!C 10:20 A!E 10:30 E!C 10:40 A!B 10:50 B!C 11:00 A!E A B A B C A A B C E A A B C E A A C E A A B C E A C E UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   50
  51. 51. ContribuQon:  IMaRS  algorithm   •  Idea:     –  add  an  expira:on  :me  to  each  triple  and     –  use  an  hash  table  to  index  triples  by  their  expiraQon  Qme   •  The  algorithm   1.  deletes  expired  triples     2.  Adds  the  new  derivaQons  that  are  consequences  of   inserQons  annota:ng  each  inferred  triple  with  an   expira:on  :me  (the  min  of  those  of  the  triple  it  is   derived  from),  and   3.  when  mul:ple  deriva:ons  occur,  for  each  mulQple   derivaQon,  it  keeps  the  max  expiraQon  Qme.   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   51
  52. 52. ContribuQon:  IMaRS  algorithm   •  Incremental  Reasoning  on  RDF  streams  (IMaRS):  new  reasoning   algorithm  opQmized  for  reacQve  query  answering   D.F.  Barbieri,  D.  Braga,  S.Ceri,  E.  Della  Valle,  M.  Grossniklaus:  Incremental  Reasoning  on   Streams  and  Rich  Background  Knowledge.  ESWC  (1)  2010:  1-­‐15   D.  Dell'Aglio,  E.  Della  Valle:  Incremental  Reasoning  on  RDF  Streams.  In  A.Harth,  K.Hose,   R.Schenkel  (Eds.)  Linked  Data  Management,  CRC  Press  2014,  ISBN  9781466582408   !  Re-materialize after each window slide !  Use DRed !  IMaRS % of deletions w.r.t. the content of the window UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   52
  53. 53. ContribuQon:  IMaRS  algorithm   •  comparison  of  the  average  Qme  needed  to  answer   a  C-­‐SPARQL  query,  when  2%  of  the  content  exits  the  window  each   Qme  it  slides,  using     –  A  backward  reasoner  on  the  window  content   –  DRed  +  standard  SPARQL  on  the  materializaQon   –  IMaRS  +  standard  SPARQL  on  the  materializaQon   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   53
  54. 54. Finding   •  Stream  Reasoning  task  is  feasible  and  the  very  nature  of   streaming  data  offers  opportuniQes  to  op:mise   reasoning  tasks  where  data  is  ordered  by  recency  and   can  be  forgoBen  a€er  a  while   –  C-­‐SPARQL  Engine  prototype   –  IMaRS  conQnuous  incremental  reasoning  algorithm   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   54
  55. 55. OpQmizing  for  stream  reasoning   alternaQve  approaches   •  DyKnow   –  How:  logical  models  of  an  observed  dynamic  system  +  metric  temporal  logics     –  Fredrik  Heintz,  Jonas  Kvarnström,  Patrick  Doherty:  Bridging  the  sense-­‐reasoning  gap:   DyKnow  -­‐  Stream-­‐based  middleware  for  knowledge  processing.  Advanced   Engineering  InformaQcs  24(1):  14-­‐26  (2010)   •  MorphStream   –  How:  rewriQng  in  DSMS  languages  (one  at  a  Qme)   –  Ref:  Calbimonte,  J.-­‐P.,  Corcho,  O.,  &  Gray,  A.  J.  G.  Enabling  ontology-­‐based  access  to   streaming  data  sources.  In  ISWC,  2010,  pages  96–111.     •  TR-­‐OWL   –  How:  Truth  maintenance  for  EL++  with  syntacQc  approximaQons   –  Ref:  Y.  Ren,  J.  Z.  Pan.  OpQmising  ontology  stream  reasoning  with  truth  maintenance   system.  In  CIKM  (2011)   •  ETALIS   –  How:  rewriQng  in  prolog   –  Ref:  Anicic,  D.,  Rudolph,  S.,  Fodor,  P.,  &  Stojanovic,  N..  Stream  reasoning  and   complex  event  processing  in  ETALIS.  SemanQc  Web,  3(4),  2012,    397–407.       (conQnues  in  the  next  slide)   SR  2015,  Austria  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   55
  56. 56. OpQmizing  for  stream  reasoning   alternaQve  approaches   •  Sparkwave   –  How:  extended  RETE  algorithm  for  windows  and  RDFS   –  Ref:  Sparkwave:  ConQnuous  Schema-­‐Enhanced  PaBern  Matching  over  RDF  Data   Streams.  Komazec  S,  Cerri  D.  DEBS  2012   •  DynamiTE   –  How:  Truth  maintenance  for  ρDF  (a  fragment  of  RDFS)   –  J.  Urbani,  A.  Margara,  C.  J.  H.  Jacobs,  F.  van  Harmelen,  H.E.  Bal:  DynamiTE:  Parallel   MaterializaQon  of  Dynamic  RDF  Data.  ISWC  (1)  2013:  657-­‐672   •  STARQL   –  How:  rewriQng  on  a  scalable  DSMS  with  Qme-­‐series  support   –  Ref:  ÖL  Özçep,  R  Möller.  Ontology  Based  Data  Access  on  Temporal  and  Streaming   Data.  Reasoning  Web,  2014   •  ASP-­‐based   –  How:  opQmizing  ASP  for  incremental  and  Qme-­‐decaying  programs   –  Ref:  hBp://arxiv.org/abs/1301.1392   •  The  Backward/Forward  Algorithm   –  How:  opQmizing  DRed   –  B.  MoQk,  Y.  Nenov,  R.E.F.  Piro,  I.  Horrocks:  Incremental  Update  of  Datalog   MaterialisaQon:  the  Backward/Forward  Algorithm.  AAAI  2015:  1560-­‐1568   SR  2015,  Austria  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   56
  57. 57. Sub-­‐research  quesQons   1.  Is  it  possible  extend  the  Seman:c  Web  stack     in  order  to  represent  heterogeneous  data  streams,   conQnuous  queries,  and  conQnuous  reasoning  tasks?   2.  Does  the  ordered  nature  of  data  streams  and  the   possibility  to  forget  old  enough  informaQon  allow  to   op:mize  con:nuous  querying  and  con:nuous  reasoning   tasks  so  to  provide  reac:ve  answers  to  large  number  of   concurrent  users  without  forsaking  correctness  or   completeness?     3.  Can  SemanQc  Web  and  Machine  Learning  technologies  be   jointly  employed  to  cope  with  the  noisy  and  incomplete   nature  of  data  streams?   4.  Are  there  prac:cal  cases  where  processing  data  stream  at   semanQc  level  is  the  best  choice?     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   57
  58. 58. Cope  with  the  noisy  and    incomplete  data   •  "Noise"  is  reduced  using  DSMS  techniques   •  Deduc:ve  stream  reasoning  copes  with  incompleteness  deducing  implicit  facts   •  Induc:ve  stream  reasoning  copes  with  "irrepairable"  incompleteness  inducing   missing  facts   D.F.  Barbieri,  D.  Braga,  S.  Ceri,  E.  Della  Valle,  Y.  Huang,  V.  Tresp,  A.  Re•nger,  H.  Wermser:   Deduc:ve  and  Induc:ve  Stream  Reasoning  for  Seman:c  Social  Media  Analy:cs.     IEEE  Intelligent  Systems  25(6):  32-­‐41  (2010)     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   58
  59. 59. Findings   •  A  combina:on  of  deduc:ve  and  induc:ve  stream   reasoning  techniques  can  cope  with  incomplete  and   noisy  data     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   59
  60. 60. AlternaQve  approaches   •  Stream  Reasoning  with  ProbabilisQc  Answer  Set   Programming   –  MaBhias  Nickles,  Alessandra  Mileo:  Web  Stream  Reasoning   Using  ProbabilisQc  Answer  Set  Programming.  RR  2014:  197-­‐205   –  Anastasios  SkarlaQdis,  Georgios  Paliouras,  Alexander  ArQkis,   George  A.  Vouros:  ProbabilisQc  Event  Calculus  for  Event   RecogniQon.  ACM  Trans.  Comput.  Log.  16(2):  11:1-­‐11:37  (2015)   –  Anni-­‐Yasmin  Turhan,  Erik  Zenker:  Towards  Temporal  Fuzzy   Query  Answering  on  Stream-­‐based  Data.  HiDeSt@KI  2015:   56-­‐69   SR  2015,  Austria  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   60
  61. 61. Sub-­‐research  quesQons   1.  Is  it  possible  extend  the  Seman:c  Web  stack     in  order  to  represent  heterogeneous  data  streams,   conQnuous  queries,  and  conQnuous  reasoning  tasks?   2.  Does  the  ordered  nature  of  data  streams  and  the   possibility  to  forget  old  enough  informaQon  allow  to   op:mize  con:nuous  querying  and  con:nuous  reasoning   tasks  so  to  provide  reac:ve  answers  to  large  number  of   concurrent  users  without  forsaking  correctness  or   completeness?     3.  Can  SemanQc  Web  and  Machine  Learning  technologies  be   jointly  employed  to  cope  with  the  noisy  and  incomplete   nature  of  data  streams?   4.  Are  there  prac:cal  cases  where  processing  data  stream  at   semanQc  level  is  the  best  choice?     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   61
  62. 62. ContribuQon:     Streaming  Linked  Data  Framework   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   62 Stream Bus Recorder Re-player AnalyserDecorator Adapter Publisher VisualizerStream HTTP HTTP Data Source Streaming Linked Data Server HTML5 Browser Marco  Balduini,  Emanuele  Della  Valle,  Daniele  Dell'Aglio,  Mikalai  Tsytsarau,  Themis   Palpanas,  CrisQan  Confalonieri:  Social  Listening  of  City  Scale  Events  Using  the  Streaming   Linked  Data  Framework.  InternaQonal  SemanQc  Web  Conference  (2)  2013:  1-­‐16  
  63. 63. ContribuQon:  RSP  services   •  RSP  services:  a  RESTful  interface  for  RSP  engines   –  hBp://streamreasoning.org/download/rsp-­‐services     UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   63
  64. 64. PracQcal  cases   •  10+  deployments  in  Sensor  Networks  &  Social  media  analyQcs,  e.g.           BOTTARI Winner of Semantic Web Challenge 2011   City Data Fusion Winner of IBM faculty award 2013   M.  Balduini,  I.  Celino,  D.  Dell’Aglio,  E.  Della  Valle,  Y.  Huang,  T.  Lee,  S.-­‐H.  Kim,  V.  Tresp:     BOTTARI:  An  augmented  reality  mobile  applica:on  to  deliver  personalized  and  loca:on-­‐based   recommenda:ons  by  con:nuous  analysis  of  social  media  streams.  J.  Web  Sem.  16:  33-­‐41  (2012)     Social Listener M.Balduini,  E.Della  Valle,  M.Azzi,  R.Larcher,  F.Antonelli,  and  P.Ciuccarelli:     CitySensing:  Fusing  City  Data  for  Visual  Storytelling.  IEEE  MulQMedia  22(3):  44-­‐53  (2015)   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   64
  65. 65. Findings   1.  The  Seman:c  Web  stack  can  be  extended  so  to  incorporate   streaming  data  as  a  first  class  ciQzen   –  RDF  stream  data  model   –  Con:nuous  SPARQL  syntax  and  semanQcs   –  Con:nuous  deduc:ve  reasoning  semanQcs           2.  Stream  Reasoning  task  is  feasible  and  the  very  nature  of   streaming  data  offers  opportuniQes  to  op:mise  reasoning   tasks  where  data  is  ordered  by  recency  and  can  be  forgoBen   a€er  a  while   –  IMaRS  conQnuous  incremental  reasoning  algorithm   –  C-­‐SPARQL  Engine  prototype   3.  A  combinaQon  of  deduc:ve  and  induc:ve  stream  reasoning   techniques  can  cope  with  incomplete  and  noisy  data     4.  There  are  applica:on  domains  where  Stream  Reasoning  offers   an  adequate  soluQon   UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   65
  66. 66. Open  issues   1.  The  Seman:c  Web  stack  can  be  extended     –  "NavigaQng  the  Chasm  between  the  Scylla  of  PracQcal  ApplicaQons   and  the  Charybdis  of  TheoreQcal  Approaches"   A.  Bernstein,  2015   2.  Stream  Reasoning  task  is  feasible     –  It's  Qme  to  start  removing  assumpQons   •  knowledge  does  not  change   •  background  data  does  not  change   –  OBDA  for  SQL  ≠  OBDA  for  conQnuous  querying   3.  Stream  reasoning  can  cope  with  incomplete  and  noisy  data   –  Theory  is  needed!     4.  There  are  applica:on  domains  where  Stream  Reasoning  offers   an  adequate  soluQon   –  Rigorous  quanQtaQve  comparaQve  research  is  needed       UiO,  Norway  -­‐    3.11.2015     @manudellavalle    -­‐    hBp://emanueledellavalle.org   66
  67. 67. AdverQsements  :-­‐P   •  Check  out  my  PhD  thesis   – hBp://dare.ubvu.vu.nl/handle/1871/53293     – Chapter  1:  IntroducQon   •  The  content  of  this  presentaQon   – Chapter  8:  conclusions   •  A  review  of  stream  reasoning  approaches  updated  in   spring  2015   •  Put  an  "I  like"  to  Stream  Reasoning  on  Facebook   – hBps://www.facebook.com/streamreasoning     @manudellavalle    -­‐    hBp://emanueledellavalle.org  UiO,  Norway  -­‐    3.11.2015     67
  68. 68. Thank  you!   Any  QuesQon?   Emanuele  Della  Valle   DEIB  -­‐  Politecnico  di  Milano   emanuele.dellavalle@polimi.it   hBp://emanueledellavalle.org     University  of  Olso,  Norway  -­‐    3.11.2015  

×