Microsoft academic Services...MICROSOFT ACADEMIC SERVICES SEPTEMBER 26, 2016 KUANSAN WANG, MANAGING...

Post on 25-May-2020

5 views 0 download

Transcript of Microsoft academic Services...MICROSOFT ACADEMIC SERVICES SEPTEMBER 26, 2016 KUANSAN WANG, MANAGING...

MICROSOFT ACADEMIC SERVICESSEPTEMBER 26, 2016

KUANSAN WANG, MANAGING DIRECTOR AND PRINCIPLE RESEARCHER

MICROSOFT RESEARCH OUTREACH, REDMOND, WA USA

Search and Browse Cloud API

https://www.microsoft.com/en-us/research/project/academic/

Academic.microsoft.com Microsoft.com/cognitive-services/en-us/academic-knowledge-api

Intelligent Open

Reach Authoritative

Index Keywords Digest Knowledge

INTELLIGENT

Matching Keywords Understand User Intent

Hit-or-Miss 10 Blue Links Dialog Experience

Most relevant results do NOT

contain query terms

Most relevant results do NOT

contain query terms

Intent confirmation

Refinement suggestions

Dynamic

disambiguation dialog

INTERACTIVE DIALOG

Author (>110M)Publication(>160M)

VenueSeries

VenueInstance

Field ofStudy (50+K)Affiliation

Citations(Billions)

REACH

OPEN

Microsoft Academic

Academic Knowledge APIs

Knowledge Exploration Service (KES)

Microsoft Academic Graph

Entity Linking Intelligent Service (ELIS)

Microsoft

Cognitive

Services

(APIs)

Data

Experiences

OPEN

ACCESSING ACADEMIC GRAPH VIA CLOUD API

• HTTPS://WWW.MICROSOFT.COM/COGNITIVE-SERVICES/EN-US/ACADEMIC-KNOWLEDGE-API

• THREE METHODS:

• INTERPRET: QUERY COMPLETION AND SEMANTIC INTERPRETATION

• EVALUATE: RETRIEVE RESULTS BASED ON INTERPRETATION

• CALCHISTOGRAM: DYNAMICALLY COMPUTE HISTOGRAM FOR ANALYTICS

GET

https://api.projectoxford.ai/academic/v1.0/interpret?query=mic&complete=1&count=5&

model=latest

HTTP/1.1

Host: api.projectoxford.ai Ocp-Apim-Subscription-Key: ••••••••••••••••••••••••••••••••

Content-Type: application/json{ "query": "mic",

"interpretations": [

{ "logprob": -11.636,

"parse": "<rule name=\"#GetPapers\">

<attr name=\"academic#AA.AfN\">microsoft</attr>

</rule>",

"rules": [

{ "name": "#GetPapers",

"output": { "type": "query", "value": "Composite(AA.AfN=='microsoft')" }

} ]

},

{ "logprob": -11.956,

"parse": "<rule name=\"#GetPapers\">

<attr name=\"academic#AA.AfN\" canonical=\"university of michigan\">

michigan university</attr></rule>",

"rules": [

{ "name": "#GetPapers",

"output": {"type": "query",

"value": "Composite(AA.AfN=='university of michigan')" }

} ]

},

https://api.projectoxford.ai/academic/v1.0/evaluate?expr=Composite(AA.AfN=='microsoft')&count=10&attributes=Ti,D,C.CN,

J.JN,ECC

Content-Type: application/json {

"expr": "Composite(AA.AfN=='microsoft')",

"entities": [

{ "logprob": -14.427,

"Ti": "xen and the art of virtualization",

"D": "2003-10-19",

"ECC": 7570,

"C": { "CN": "sosp" }

},

{ "logprob": -14.523,

"Ti": "robust real time face detection",

"D": "2004-05-01",

"ECC": 11041,

"J": { "JN": "ijcv" }

},

]

}

https://api.projectoxford.ai/academic/v1.0/calchistogram?expr=And(Composite(F.FN=='machine

learning'),Y>2011)&model=latest&attributes=AA.AfN&count=10&offset=0&timeout=10000

Content-Type: application/json {

"expr": "And(Composite(F.FN=='machine learning'),Y>2011)",

"num_entities": 255762,

"histograms": [

{ "attribute": "AA.AfN", "distinct_values": 6985, "total_count": 189236,

"histogram": [

{ "value": "microsoft",

"logprob": -12.444,

"count": 1371

},

{ "value": "chinese academy of sciences",

"logprob": -12.992,

"count": 2256

},

{ "value": "carnegie mellon university",

"logprob": -13.069,

"count": 1060

}, …

}

AUTHORITATIVE

ISSUES

• INCUMBENT METRICS ARE OFTEN MISUSED/ABUSED

• JOURNAL IMPACT FACTOR

• H-INDEX

• READY FOR ALMETRICS?

DRAWBACK OF CITATION COUNTS

A BCitation #A = 5 Citation #B = 3

Is A > B?

• Counts alone are insufficient

• Need to consider credentials of

citing parties

RANKING BEHIND MICROSOFT ACADEMIC

• BASED ON PERON THEOREM (1904): PROBABILITY OF BECOMING WIDELY-KNOWN

• “WORK MORE IMPORTANT IF CITED BY IMPORTANT WORK”

• REDISCOVERED BY PINKSKI & NARIN, IPM (1976)

• MADE FAMOUS BY PAGERANK (1998), HITS (1998, ASK.COM), RANKDEX (1996, BAIDU)

• EXTENDED TO HETEROGENEOUS GRAPH

• “WORK MORE IMPORTANT IF CITED BY IMPORTANT <ENTITY> WEIGHTED BY <ENTITY TYPE>”

• <ENTITY> MORE IMPORTANT IF PRODUCING IMPORTANT WORK

• READY FOR SOCIAL SIGNALS!

• TEMPORAL DYNAMICS CONSIDERED

• OLDER WORK NEEDS SUSTAINED INFLUENCES TO MAINTAIN RANK

MOST SELF-CONSISTENT PARAMETERS

0.939

0.007

0.0230.013 0.012

0.005

0.991

0.948

0.005

0.019

0.009 0.0120.007

0.991

Citation Author Affiliation Venue Field of Study Keyword Time Decay

Jun-16 Mar-16

Overall 2012-2016 1997-2011

1 Nature Nature Nature

2 Science PNAS Science

3 PNAS Science PNAS

4 Cell Journal of Biological Chemistry Cell

5 Nucleic Acids Research Cell Nature Genetics

6 Nature Genetics PLOS ONE Nucleic Acids Research

7 Nature Medicine Nucleic Acids Research Nature Medicine

8 Nature Biotechnology Nature Genetics Bioinformatics

9 Bioinformatics Nature Methods Nature Biotechnology

10 New England Journal of Medicine Nature Medicine Blood

1 10 100

Log

Sa

lienc

y

H-index

Saliency vs. H-index for Authors

OPEN ACADEMIC SOCIETY

• LAUNCHED BY UNIVERSITY AND INDUSTRY PARTNERS IN JULY, 2016

• OPEN MEMBERSHIP

• CORE BELIEFS

• ACADEMIC DATA SHOULD BE OPEN

• ALGORITHMS PROCESSING THESE DATA SHOULD BE OPEN

• COMMUNITY EFFORT TO MAKE PROGRESS, BREAK DOWN PROPRIETARY WALLS

• OPENACADEMIC.AI FOR MORE DETAILS

Intelligent Open

Reach Authoritative

31