Upload
brian-ulicny
View
977
Download
3
Tags:
Embed Size (px)
DESCRIPTION
Doctrinally, Priority Intelligence Requirements (PIRs) represent information that the commander needs to know in order to make a decision or achieve a desired effect. Networked warfare provides the intelligence officer with access to multitudes of sensor outputs and reports, often from unfamiliar sources. Counterinsurgency requires evaluating information across all PMESII-PT categories: Political, Military, Economic, Social, Infrastructure Information, Physical Environment and Time. How should analysts evaluate this information? NATO's STANAG (Standard Agreement) 2022 requires that every piece of information in intelligence reports used to answer PIRs should be evaluated along two independent dimensions: the reliability of its source and the credibility of the information. Recent developments in information retrieval technologies, including social search technologies, incorporate metrics of information evaluation, reliability and credibility, such as Google's PageRank. In this paper, we survey various current approaches to automatic information evaluation and explore their applicability to the information evaluation and PIR answering tasks. (Presented at Fusion 2010)
Citation preview
Fusion 2010 13th International Conference on Information Fusion EICC, Edinburgh, UK Thursday, 29 July 2010
Current Approaches to Automated Information Evaluation and their Applicability to
Priority Intelligence Requirement Answering
Outline
• Overview • Priority Intelligence Requirements• Doctrine: Reliability/Credibility• Question-Answering Technologies• Conclusion/Research Gaps• Disclaimer
VIStology | FUSION 2010 | Edinburghwww.vistology.com 2
Overview
Priority Intelligence Requirement (PIR) Answering requires STANAG 2022 assessments of information reliability/credibility/independence. Each element of information in a PIR should have an assessment of: the accuracy of the information provided, how credible it is, and how reliable the source is. STANAG 2022 explicitly adopted in US and NATO doctrine.
There are currently no real tools for making these assessments, or for reasoning with STANAG-assessed data.
Contemporary commercial question-answering tools partially address some of the necessary reliability/independence/credibility issues.
Here we survey the state of the art technology and evaluate the research gaps.
www.vistology.com VIStology | FUSION 2010 | Edinburgh 3
Priority Intelligence Requirements: Doctrine
Priority Intelligence Requirements (PIRs) are “those intelligence requirements for which a commander has an anticipated and stated priority in his task of planning and decision making” (FM 2-0 “Intelligence”, section 1-32)
• Ask a single question. • Are ranked in importance.• Are specific: Focus on a specific event, fact or activity.• Are tied to a single decision or planning task the commander has to
make.• Provide a last time by which information is of value (LTIOV).• Are answerable using available assets and capabilities.
– McDonough, LTC W. G., Conway, LTC J. A., “Understanding Priority Intelligence Requirements”, Military Intelligence Professionals Bulletin, April-June 2009.
VIStology | FUSION 2010 | Edinburghwww.vistology.com 4
NATO STANAG 2022
Question-Answering Technologies by Source Data Format
VIStology | FUSION 2010 | Edinburghwww.vistology.com 6
Information Source Format
FamiliarApplication
AdvancedApplication
Tables(Relational DBs,Spreadsheets)
Structured Query Language (SQL)
Wolfram Alpha(Mathematica)
Text Web Search Engines (Google, Yahoo!, Ask)
Systems from AQUAINT (IC) competition; IBM Watson
Tagged Text Google Patent Search Metacarta;Palantir
Logic Statements Prolog Powerset (acquired by MS Bing); Cyc
Trusted Teammates
Personal Communication
Yahoo! Answers;Vark (acquired by Google);US Army Intelligence Knowledge Network Shoutbox
Structured Data Q-A: Wolfram Alpha
www.vistology.com VIStology | FUSION 2010 | Edinburgh 7
Wolfram Alpha identifies Tupelo as where Elvis was born (Elvis
disambiguated as Elvis Presley) and provides map overlay and additional
info, like current city population. Reference sources listed by title on another screen, no access to source
data.
Query “Where was Elvis born?” automatically translated to Mathematica query: Elvis
Presley, place of birth.
Text: Google
VIStology | FUSION 2010 | Edinburghwww.vistology.com 8
Google PageRank disambiguates query:
Elvis = Elvis Presley by PageRank.Top-ranked snippets can easily be
scanned for consensus answer from independent sources:
Tupelo, MS. PageRank less useful in MI context because reports are
not hyperlinked.
Text-Based Q-A: IBM Watson
www.vistology.com VIStology | FUSION 2010 | Edinburgh 9
IBM’s text-based algorithms identified these
phrases as top potential “Jeopardy” answer, with
scores displayed.In “Jeopardy”, answer is in
form of question.
Query in “Jeopardy” format (including category “Musical Pastiche”)
Tagged Text: Metacarta
VIStology | FUSION 2010 | Edinburghwww.vistology.com 10
Query identifies documents that contain
“elvis”, “born” and a location. Answers literally
all over the map. Consensus answer not obvious from location
clusters. Documents are recent news articles.
Query: “Where was Elvis born?”
Logic-Based Q-A: Powerset
VIStology | FUSION 2010 | Edinburghwww.vistology.com 11
Answers involve
multiple “Elvises”.
Source data is Wikipedia
only.
Social Question-Answering: Vark
VIStology | FUSION 2010 | Edinburghwww.vistology.com 12
Routed to unknown user
in my ‘network’ computed as
likely to provide answer; Answer returned in less
than minute.Optimized for
mobile environment.
Feedback
Vark queries need to be over certain length,
hence this phrasing.
Comparison by TechnologySTANAG
RequirementTables:Wolfram Alpha
Text:GoogleIBM Watson
Tagged Text:MetacartaPalantir
Logic Statements:Powerset
Teammates:VarkY! Answers
Source Wolfram: Reference document title (no url)
URL of document in which info appears (usually: not Watson). No further attempt to match info to source. I.e. not: 1000 demonstrators according to police.
Teammate known. May not say where info originates.
Source Reliability Curated data: Reference works, Government data.
Centrality measures: Google PageRank (eigenvector centrality); Technorati Authority (inlink centrality);VIStology blogger authority (centrality + engagement)
Curated data: Wikipedia.
Wikipedia has PageRank: 9 out of 10 (reliable)
Track record, Reputation.Votes on answers.Longevity.# of answers
Source Independence
No. One unified datastore.
Duplicate document detection;Explicit source tracking (href; bit.ly); Leskovec meme tracking.SNA metrics of independence.
No. Single data source.
UserAuthentication.
InformationCredibility
Partial Integrity constraints.Can’t easily verify info.
Consensus answers; same answer identified in multiple distinct sources.
Could check integrity constraints; URI co-ref a problem. Contradictions halt inference.
Demonstrated area of expertise
www.vistology.com VIStology | FUSION 2010 | Edinburgh 13
Research Gaps(1) How best to map network-based reliability metrics to
STANAG 2022 reliability codes?(2) How to make reliability metrics derived from networks of
different scales comparable with non-estimated reliability metrics?
(3) How to automatically reason with information that has been assigned STANAG 2022 evaluation codes?
(4) How to efficiently identify independent confirmation of reports in social media and other networked sources?
(5) How to tractably identify inconsistent new reports?(6) How to adjudicate inconsistencies among reports
automatically?
www.vistology.com VIStology | FUSION 2010 | Edinburgh 14
Conclusions
In contemporary environments, direct evaluation of source reliability may be impossible given the proliferation of OSINT and other sources relevant to COIN fight across all PMESII-PT categories.
Networked sources make judging independence of sources and identifying influence more difficult.
Analysts may have to rely on correlated network-based metrics of reliability, credibility and inde-dence rather than evaluate many sources/reports as “Reliability cannot be judged”/”Truth cannot be judged”.
www.vistology.com VIStology | FUSION 2010 | Edinburgh 15
Thank You
Note: This paper does not represent an endorsement by the Army Research
Laboratory of any of the commercial products discussed.
VIStology | FUSION 2010 | Edinburghwww.vistology.com 16