ibm watson technical deep dive swiss group for artificial intelligence and cognitive science

42
© 2013 IBM Corporation Swiss Group for Artificial Intelligence and Cognitive Science Intelligent Systems and Applications Workshop 2014, University of Basel Watson Technical Deep Dive @RomeoKienzler, IBM Innovation Center Zurich

Upload: romeo-kienzler

Post on 02-Jul-2015

535 views

Category:

Internet


2 download

DESCRIPTION

We are transitioning from the programmatic to the cognitive computing era.IBM Deep Blue won against the world champion in Chess 1996. IBM Watson won against the two world champions in the famous US quiz show "Jeopardy" 2011. Since then, the press heavily established the term "Cognitive Computing" to the public. I will explain how IBM Watson works internally and start with Algebraic Text Extraction. DeepQA is the heart of IBM Watson and I will explain each component of this pipeline, the linguistic preprocessor, hypothesis generation, hypothesis and evidence scoring, final matching based on supervised learning and confidence estimation. Finally, I conclude with an overview of actual use cases and outline the roadmap of future work.

TRANSCRIPT

Page 1: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation

Swiss Group for Artificial Intelligence and Cognitive Science Intelligent Systems and Applications Workshop 2014, University of BaselWatson Technical Deep Dive

@RomeoKienzler, IBM Innovation Center Zurich

Page 2: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation2

(part of) my role @IBM● Accelerate cognitive computing

● In Switzerland● Through

● Academia● Startups/ISV's● Cloud

Watson in the cloud: bit.ly/go4bluemix

Page 3: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation3

What Watson is not● Search Engine● Database System● HAL9000

Page 4: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation4

Page 5: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation5

What Watson is● Cognitive System (Marketing)● Combination of

● Information Retrieval● NLP

● Structured + Unstructured Data !● Runs on UIMA● Based on supervised learning

Page 6: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation6

What is a parser?● Annotate sentence with

● Tags● Relationships

● Probabilistic (e.g. Stanford)● Rule based

● (E) Slot Grammar

Page 7: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation7

Slot Grammar● Simplified● Lexicalist character

● High focus on words● Low focus on structure

● Assign words to slotsI can resist everything except temptation● Subject (I)● Verb (can resist)● Object (everything except temptation)

Page 8: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation8

PAS - Builder● Predicate-Argument Structure● Downstream to ESG● Reduces complexity of ESG“John opened Bill's door (with his key)

John's key opened Bill's doorBill's door openedBill's door was opened (by John)”

OPEN (John door key) | | |

Agent Theme Instrument

Many ESG trees reduce to same PAS

Page 9: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation9

Relationships● Relationship Extractor● Combination of

● Manual pattern specifications~30 types, high precision

● Statistical methods~7000 types, low precision

● SVM's on DBPedia/Wikipedia

Page 10: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation10

Relationships (2)“The Screwtape Letters” from a senior devil to an under devil are by this man better known for children’s books

author(“this man”,“The Screwtape Letters”)

Page 11: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation11

Ingestion

Page 12: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation12

Ingestion● Corpus creation● Input format: TREC

(Text Retrieval Conference)● Multiple HTML pages in one

HDFS file● Parallel ingestion process

(LiteScale)

Page 13: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation13

Dictionary● started w/ Wikipedia copus● Keyword → Text structure● Transformation of free text

● into Keyword → Text● optimization objective

Page 14: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation14

Knowledge Expansion● Follow links in content● Identify content keywords and link

to new content● → generate more content in

Keyword → Text form

Page 15: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation15

Question Analysis

Page 16: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation16

Question Analysis● Named entity recognition

● Type identification /Extract focus● ESG/PAS

● Relationship detection

Page 17: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation17

Question Analysis1) Extract focus2) Map to LAT3) Broad Type classification4) Detect if special handling is

needed (e.g. nested question)

Page 18: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation18

Query Decomposition

Page 19: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation19

Query Decomposition● Keyword identification● LAT (Lexical Answer Type)

● IBM Pat. US20120078890 for confidence estimation of LAT

● optimization objective: choosing keywords out of nontrivial set of words based on ML

Page 20: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation20

Query DecompositionIn 1894 C.W. Post created his warm cereal drink Postum in this Michigan city● Focus: this Michigan City● LAT: Michigan● Keywords: 1894, C.W. Post,

created, warm, cereal, drink, Postum, Michigan, City

Page 21: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation21

Query Decomposition

Page 22: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation22

Primary Search

Page 23: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation23

Primary Search● Lucene and Indri search engine● Preprocessing generated

keyword->text based documents● Keyword associated with found

document added to candidate answer list

Page 24: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation24

Hypothesis generation

Page 25: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation25

Supporting Evidence Retrieval

Unlike most sea animals, in the sea horse this pair of sense organs can move independently of one another

Question decomposition:Which [sense organ] of [Sea Horse] move independently?

Hypothesis generation:A Sea Horse can move its eyes independently.A Sea Horse can move its ears independently.A Sea Horse can move its skin independently.A Sea Horse can move its nose independently.A Sea Horse can move its tung independently.

Page 26: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation26

http://angelalmassey.com/SHC/about.html

Page 27: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation27

Supporting Evidence● Generated Candidate Answer is

● ESG'd● PAS'd● searched against corpus● LATs used to determine whether

a candidate answer is an instance of the answer types

Page 28: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation28

Supporting Evidence

Page 29: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation29

Scoring

Page 30: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation30

Scoring● Optimization objective

(confidence estimation framework)● Relational (PRISMATIC, Dbpedia)● Taxonomic,Geospacial● Temporal, Source Reliability● Gender, Name consistency● Passage Support● Theory consistency

Page 31: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation31

Scoring challenges● Feature significance different for

● Different questions ● Different question classes

● Very heterogeneous features● Normalization problem● Missing features

● Class imbalance

Page 32: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation32

Merging and ranking

Page 33: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation33

Merging and ranking1. John Fitzgerald Kennedy 2. Kennedy, 3. JFK● Different Scores● Merge to canonical form

● Morphological● Pattern-based● Table Lookup

● Partially generated from Wikipedia disabiguation pages

Page 34: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation34

ExampleMYTHING IN ACTION: One legend says this was given by the Lady of the Lake & thrown back in the lake on King Arthur’s death.

● Watson merged sword + Excalibur to “sword” (canonical form)

● Preserved relation● more_specific(sword)->Excalibur

Page 35: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation35

ML in Ranking● Experiments with logistic regression, support

vector machines, linear and nonlinear kernels, ranking SVM, boosting, single and multilayer neural nets, decision trees, locally weighted learning

● Finally: regularized logistic regression

Page 36: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation36

Normalization● Q set of all candidate answers● Feature x_ij

● j feature, i answer● missing values imputed

Page 37: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation37

Ranking● Based on training set n > 10K● IBM SPSS Modeler

Page 38: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation38

Evidence Sources

Page 39: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation39

Automatic Learning● Read through text semantically● Statistically rank annotated text● generate new knowledge

● Inventors patent inventions 0.8● officials submit resignations 0.7● people earn degrees at schools 0.9● fluid is a liquid 0.6● liquid is a fluid 0.5● vessels sink 0.7● people sink 8-balls (0.5) (in pool/0.8)

Page 40: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation40

Next steps● “Jeopardy!” - Watson was

● Open domain● Large training set

● New “Watsons” are● Closed domain● Small, but growing training set

Page 41: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation41

Demo● Bit.ly/go4bluemix

Page 42: IBM Watson Technical Deep Dive Swiss Group for Artificial Intelligence and Cognitive Science

© 2013 IBM Corporation42

References[1] Jeffrey Kabot, “Deep Parsing”[2] Richard Nordquist, “slot and filler”[3] The Journal of Research and Development, Vol 56, 2012