ÇağatayDemiralp,PeterJ.Haas,SrinivasanParthasarathy,TejaswiniPedapati

IBMT.J.WatsonResearchCenter,NewYork

FORESIGHT Recommending Visual Insights

Rapid visual exploration of high dimensional data

Visual insights are grouped into classes

Bookmark & search related insights

Users can bookmark interesting

insights. Foresight supports search and ranking of

insights related to a given insight —

only those insights in the

neighborhood of a given insight will

be displayed within each class.

Overview visualization for a class

Insights within a class can be ranked and filtered using statistical measures.

Overview of linear relationships —

clustered heatmap of pairwise

correlations.−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

CnOR

EdcA

StdS

QOSN

SlRH

LfSt

EmpR

WtrQ

LfEx

HNFW

RmPP

HNAD

PrsE

VtrT

YrIE

TDTL

HsnE

JbSc

LnUR

AssR

HmcR

DWBF

ArPl

EWVL

CnOREdcAStdS

QOSNSlRHLfSt

EmpRWtrQLfEx

HNFWRmPPHNADPrsEVtrTYrIETDTLHsnEJbScLnURAssRHmcRDWBFArPl

EWVL

Interactive response timesForesight Architecture

Dispersion

Outliers

Insight Microservices

Heterogeneous frequencies

Multimodality

• Random Hyperplane Sketch • Random Projection Sketch • Entropy Sketch • Frequent Item Sketch • Quartile Sketch

User Interaction

Sketch request

Sketch response

Ranking Metric & Filters

Ranked Insights

File UploadSketch Microservice

File store

File response

File request

Skew

Heavy tails Linear relationship

Monotonic relationship

Statistical dependence

Segmentation

Foresight employs sketching

algorithms to speed up insight

queries