PrefMiner: Mining Users Preferences for Intelligent Mobile - - PowerPoint PPT Presentation

prefminer mining user s preferences for intelligent
SMART_READER_LITE
LIVE PREVIEW

PrefMiner: Mining Users Preferences for Intelligent Mobile - - PowerPoint PPT Presentation

PrefMiner: Mining Users Preferences for Intelligent Mobile No5fica5on Management Abhinav Mehrotra Robert Hendley Mirco Musolesi University College London University of


slide-1
SLIDE 1

PrefMiner: ¡Mining ¡User’s ¡Preferences ¡for ¡ Intelligent ¡Mobile ¡No5fica5on ¡Management ¡

Abhinav ¡Mehrotra ¡

University ¡College ¡London ¡ University ¡of ¡Birmingham ¡

Robert ¡Hendley ¡

University ¡of ¡Birmingham ¡

¡ Mirco ¡Musolesi ¡

University ¡College ¡London ¡

slide-2
SLIDE 2

No5fica5ons ¡Inform ¡Users ¡ ¡ About ¡a ¡Variety ¡of ¡Events ¡ ¡

System, ¡Tools ¡and ¡Others ¡ Online ¡Social ¡Networks ¡ Communica5on ¡

slide-3
SLIDE 3

Are ¡All ¡No5fica5ons ¡Useful? ¡

slide-4
SLIDE 4

Are ¡All ¡No5fica5ons ¡Useful? ¡ NO ¡

¡ Users ¡are ¡not ¡interested ¡in ¡many ¡no5fica5ons ¡they ¡receive ¡ ¡ and ¡feel ¡annoyed ¡on ¡receiving ¡them. ¡

slide-5
SLIDE 5

Interrup5bility ¡Management ¡ System: ¡State-­‑of-­‑the-­‑art ¡

Predict ¡opportune ¡moments ¡by ¡using: ¡

  • context ¡of ¡the ¡user. ¡

¡

  • content ¡of ¡no5fica5on. ¡

Limita8on ¡ No ¡past ¡study ¡has ¡considered ¡stopping ¡or ¡detec5ng ¡ no5fica5ons ¡that ¡are ¡not ¡useful ¡for ¡the ¡user. ¡

slide-6
SLIDE 6

Interrup5bility ¡Management ¡ System: ¡State-­‑of-­‑the-­‑art ¡

Evaluated ¡in ¡the ¡offline ¡sePngs. ¡ ¡ Limita8on ¡ Never ¡evaluated ¡in ¡the ¡real-­‑world ¡scenario. ¡

slide-7
SLIDE 7

Contribu5ons ¡

  • 1. A ¡mechanism ¡for ¡learning ¡the ¡types ¡of ¡informa5on ¡users ¡

prefer ¡to ¡receive ¡via ¡no5fica5ons ¡in ¡different ¡situa5ons. ¡ ¡ ¡ ¡

  • 2. A ¡technique ¡to ¡make ¡the ¡predic5ve ¡model ¡transparent ¡to ¡
  • users. ¡ ¡
slide-8
SLIDE 8

Mining ¡User ¡Preferences ¡

slide-9
SLIDE 9

Mining ¡User ¡Preferences ¡

Construc5ng ¡ Associa5on ¡Rules ¡ ¡ ¡ Removing ¡ Reminder ¡ No5fica5ons ¡ No5fica5on ¡ Clustering ¡

slide-10
SLIDE 10

Mining ¡User ¡Preferences ¡

Construc5ng ¡ Associa5on ¡Rules ¡ ¡ ¡ Removing ¡ Reminder ¡ No5fica5ons ¡ No5fica5on ¡ Clustering ¡

slide-11
SLIDE 11

¡ If ¡an ¡applica5on’s ¡click ¡rate ¡is ¡zero ¡then ¡all ¡ no5fica5ons ¡from ¡that ¡applica5on ¡are ¡treated ¡ as ¡reminder ¡no5fica5ons. ¡ ¡ Click ¡Rate ¡ ¡ Number ¡of ¡accepted ¡no5fica5ons ¡ ¡ Total ¡number ¡of ¡no5fica5ons ¡ ¡ * ¡ Some ¡no5fica5ons ¡are ¡dismissed ¡because ¡they ¡do ¡not ¡require ¡ any ¡further ¡ac5on ¡from ¡the ¡user. ¡ ¡ ¡ = ¡ 100 ¡

S1: ¡Removing ¡Reminder ¡No5fica5ons ¡

slide-12
SLIDE 12

Mining ¡User ¡Preferences ¡

Construc5ng ¡ Associa5on ¡Rules ¡ ¡ ¡ Removing ¡ Reminder ¡ No5fica5ons ¡ No5fica5on ¡ Clustering ¡

slide-13
SLIDE 13

S2: ¡No5fica5on ¡Clustering ¡

  • 1. Cleaning ¡No5fica5on ¡Titles ¡
  • Conversion ¡of ¡the ¡text ¡to ¡lower-­‑case ¡ ¡
  • Removal ¡of ¡punctua5on, ¡numbers ¡and ¡stop ¡words ¡ ¡
  • Removal ¡of ¡the ¡sender ¡and ¡applica5on ¡names ¡ ¡
  • Stemming ¡of ¡words ¡ ¡
  • 2. Clustering ¡No5fica5ons ¡ ¡
  • Removing ¡sparse ¡terms ¡
  • Term ¡Frequency-­‑based ¡clustering ¡

No5fica5on ¡Title ¡ Applica5on ¡Name ¡

slide-14
SLIDE 14

S2: ¡No5fica5on ¡Clustering ¡

Removing ¡sparse ¡terms ¡

TFthreshold ¡ ¡ Number ¡of ¡par5cipa5on ¡days ¡ ¡ ¡ ¡ ¡ ¡ N ¡* ¡Total ¡number ¡of ¡no5fica5ons ¡ ¡ = ¡ This ¡ensures ¡that ¡at ¡least ¡one ¡no5fica5on ¡containing ¡ ¡ the ¡term ¡is ¡triggered ¡in ¡N ¡days. ¡

1000 2000 3000 1 2 3 4 5 6 7

N−value Terms Count

Filtered Remaining

Not ¡much ¡difference ¡in ¡the ¡ number ¡of ¡filtered ¡terms ¡with ¡ N ¡∈ ¡[2, ¡7]. ¡ ¡ We ¡use ¡N ¡= ¡2 ¡for ¡Tfthreshold ¡. ¡

slide-15
SLIDE 15

Mining ¡User ¡Preferences ¡

Construc5ng ¡ Associa5on ¡Rules ¡ ¡ ¡ Removing ¡ Reminder ¡ No5fica5ons ¡ No5fica5on ¡ Clustering ¡

slide-16
SLIDE 16

S3: ¡Construc5ng ¡Associa5on ¡Rules ¡

We ¡use ¡the ¡AIS ¡algorithm ¡for ¡mining ¡associa5on ¡rules. ¡ ¡ ¡ X ¡ ¡→ ¡ ¡Y ¡ ¡ ¡ X ¡is ¡the ¡antecedent ¡ ¡(never ¡contains ¡no5fica5on ¡response) ¡ Y ¡as ¡the ¡consequent ¡ ¡(only ¡contains ¡no5fica5on ¡response) ¡

slide-17
SLIDE 17

S3: ¡Construc5ng ¡Associa5on ¡Rules ¡

We ¡use ¡the ¡AIS ¡algorithm ¡for ¡mining ¡associa5on ¡rules. ¡ ¡ ¡ X ¡ ¡→ ¡ ¡Y ¡ ¡ ¡ X ¡is ¡the ¡antecedent ¡ ¡(never ¡contains ¡no5fica5on ¡response) ¡ Y ¡as ¡the ¡consequent ¡ ¡(only ¡contains ¡no5fica5on ¡response) ¡ Support: ¡

Number ¡of ¡no5fica5ons ¡of ¡the ¡no5fica5on-­‑type ¡covered ¡in ¡the ¡rule ¡ ¡ Total ¡number ¡of ¡no5fica5ons ¡in ¡the ¡dataset ¡

¡ Confidence: ¡

Occurrence ¡count ¡of ¡the ¡no5fica5on ¡pabern ¡covered ¡in ¡the ¡rule ¡ ¡ Occurrence ¡count ¡of ¡the ¡no5fica5on ¡pabern ¡covered ¡in ¡the ¡antecedent ¡of ¡rule ¡ * ¡ 100 ¡ * ¡ 100 ¡

slide-18
SLIDE 18

S3: ¡Construc5ng ¡Associa5on ¡Rules ¡

We ¡use ¡the ¡AIS ¡algorithm ¡for ¡mining ¡associa5on ¡rules. ¡ ¡ ¡ X ¡ ¡→ ¡ ¡Y ¡ ¡ ¡ X ¡is ¡the ¡antecedent ¡ ¡(never ¡contains ¡no5fica5on ¡response) ¡ Y ¡as ¡the ¡consequent ¡ ¡(only ¡contains ¡no5fica5on ¡response) ¡

{N1} ¡→ ¡{Dismiss} ¡ {N2,Home} ¡→ ¡{Accept} ¡ ¡ {N2,Other} ¡→ ¡{Accept} ¡ {N2,Work} ¡→ ¡{Dismiss} ¡

Let ¡us ¡consider ¡an ¡example ¡where ¡the ¡user: ¡ ¡

  • always ¡dismiss ¡N1 ¡no5fica5ons. ¡ ¡
  • does ¡not ¡accept ¡N2 ¡no5fica5ons ¡at ¡work. ¡
slide-19
SLIDE 19

Procedure ¡

  • 1. Evaluate ¡(offline) ¡the ¡proposed ¡intelligent ¡no5fica5on ¡mechanism. ¡

¡ ¡

  • 2. Funnel ¡the ¡findings ¡in ¡an ¡intelligent ¡no5fica5on ¡library: ¡MyPref. ¡

¡ ¡

  • 3. Implement ¡an ¡app ¡on ¡top ¡of ¡MyPref ¡and ¡deploy ¡it ¡to ¡evaluate ¡the ¡

library ¡in ¡a ¡ ¡real-­‑world ¡scenario. ¡

slide-20
SLIDE 20

Evalua5on ¡(Offline) ¡

slide-21
SLIDE 21

Dataset ¡

Study: ¡My ¡Phone ¡and ¡Me ¡ ¡ ¡ Par5cipants: ¡18 ¡ ¡ Minimum ¡ac5ve ¡days: ¡14 ¡ ¡ No5fica5on ¡samples: ¡11,185 ¡

Abhinav ¡Mehrotra, ¡Veljko ¡Pejovic, ¡Jo ¡Vermeulen, ¡Robert ¡Hendley, ¡and ¡Mirco ¡

  • Musolesi. ¡2016. ¡My ¡Phone ¡and ¡Me: ¡Understanding ¡People’s ¡Recep5vity ¡to ¡Mobile ¡
  • No5fica5ons. ¡In ¡CHI’16. ¡ ¡
slide-22
SLIDE 22

Types ¡of ¡Associa5on ¡Rules ¡

Assoc ¡Rule ¡1: ¡no5fica5on ¡response ¡with ¡no5fica5on ¡type. ¡ ¡ Assoc ¡Rule ¡2: ¡no5fica5on ¡response ¡with ¡no5fica5on ¡type ¡and ¡

  • ac5vity. ¡ ¡

¡ Assoc ¡Rule ¡3: ¡no5fica5on ¡response ¡with ¡no5fica5on ¡type ¡and ¡arrival ¡

  • 5me. ¡ ¡

¡ Assoc ¡Rule ¡4: ¡no5fica5on ¡response ¡with ¡no5fica5on ¡type ¡and ¡

  • loca5on. ¡ ¡

¡ Assoc ¡Rule ¡5: ¡no5fica5on ¡response ¡with ¡no5fica5on ¡type, ¡ac5vity, ¡ arrival ¡5me ¡and ¡loca5on. ¡

slide-23
SLIDE 23

Results ¡

Recall: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡and ¡ the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡actually ¡dismissed. ¡ ¡ ¡ Precision: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡ and ¡the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡predicted ¡as ¡dismissed. ¡ ¡

10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall

AR4 ¡ AR5 ¡

Minimum ¡Confidence: ¡65-­‑95% ¡ Minimum ¡Support: ¡Dpar5cipate ¡/ ¡( ¡2 ¡∗ ¡Ncount ¡) ¡ ¡

Dpar5cipate: ¡Days ¡of ¡par>cipa>on ¡ Ncount: ¡number ¡of ¡no5fica5ons ¡collected ¡for ¡a ¡user ¡ ¡ This ¡ensures ¡that ¡the ¡type ¡of ¡no5fica5on ¡covered ¡in ¡each ¡rule ¡ has ¡arrived ¡at ¡least ¡once ¡in ¡two ¡days. ¡

Assoc ¡Rule ¡1 ¡ Assoc ¡Rule ¡2 ¡ Assoc ¡Rule ¡3 ¡

slide-24
SLIDE 24

Results ¡

Recall: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡and ¡ the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡actually ¡dismissed. ¡ ¡ ¡ Precision: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡ and ¡the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡predicted ¡as ¡dismissed. ¡ ¡

10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall

Assoc ¡Rule ¡4 ¡ Assoc ¡Rule ¡5 ¡ Assoc ¡Rule ¡1 ¡ Assoc ¡Rule ¡2 ¡ Assoc ¡Rule ¡3 ¡

slide-25
SLIDE 25

Results ¡

Recall: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡and ¡ the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡actually ¡dismissed. ¡ ¡ ¡ Precision: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡ and ¡the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡predicted ¡as ¡dismissed. ¡ ¡

10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall

Best ¡ Performance ¡

Assoc ¡Rule ¡1 ¡ Assoc ¡Rule ¡2 ¡ Assoc ¡Rule ¡3 ¡ Assoc ¡Rule ¡4 ¡ Assoc ¡Rule ¡5 ¡

slide-26
SLIDE 26

Results ¡

Recall: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡and ¡ the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡actually ¡dismissed. ¡ ¡ ¡ Precision: ¡ra5o ¡between ¡the ¡number ¡of ¡no5fica5ons ¡that ¡are ¡correctly ¡predicted ¡as ¡dismissed ¡ and ¡the ¡total ¡number ¡of ¡no5fica5ons ¡that ¡are ¡predicted ¡as ¡dismissed. ¡ ¡

10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall 10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall

AR4 ¡ AR5 ¡

Best ¡ Performance ¡

User’s ¡preference ¡for ¡receiving ¡a ¡no8fica8on ¡depends ¡

  • nly ¡on ¡the ¡type ¡of ¡informa8on ¡it ¡contains ¡ ¡

and ¡the ¡loca8on. ¡

Assoc ¡Rule ¡1 ¡ Assoc ¡Rule ¡2 ¡ Assoc ¡Rule ¡3 ¡

slide-27
SLIDE 27

Op5mizing ¡the ¡System ¡ ¡ for ¡High ¡Precision ¡ ¡

10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall

Assoc ¡Rule ¡4 ¡

  • We ¡should ¡aim ¡to ¡have ¡fewer ¡false-­‑nega5ves ¡(i.e., ¡incorrectly ¡

predic5ng ¡a ¡no5fica5on ¡as ¡non-­‑interes5ng ¡for ¡the ¡user). ¡

  • This ¡can ¡be ¡achieved ¡by ¡ensuring ¡that ¡the ¡precision ¡remains ¡

close ¡to ¡100%. ¡

slide-28
SLIDE 28

Op5mizing ¡the ¡System ¡ ¡ for ¡High ¡Precision ¡ ¡

10 20 30 40 50 60 70 80 90 100 70 80 90

Confidence Value (%) Result (%)

Precision Recall

  • We ¡should ¡aim ¡to ¡have ¡fewer ¡false-­‑nega5ves ¡(i.e., ¡incorrectly ¡

predic5ng ¡a ¡no5fica5on ¡as ¡non-­‑interes5ng ¡for ¡the ¡user). ¡

  • This ¡can ¡be ¡achieved ¡by ¡ensuring ¡that ¡the ¡precision ¡remains ¡

close ¡to ¡100%. ¡ 90% ¡precision ¡ 35% ¡recall ¡ ¡

Assoc ¡Rule ¡4 ¡

slide-29
SLIDE 29

Online ¡Learning ¡

  • 90

100 1 3 5 7 9 11 13 15 17 19 21

Day Precision (%)

Confidence

  • 65%

70% 75% 80% 85% 90% 95%

  • 10

20 30 40 1 3 5 7 9 11 13 15 17 19 21

Day Recall (%)

Confidence

  • 65%

70% 75% 80% 85% 90% 95%

slide-30
SLIDE 30

MyPref ¡Library ¡

An ¡intelligent ¡interrup5bility ¡management ¡library ¡that ¡can ¡ predict ¡the ¡type ¡of ¡no5fica5ons ¡that ¡users ¡would ¡prefer ¡to ¡ receive ¡on ¡their ¡mobile ¡phones ¡in ¡specific ¡contexts. ¡ ¡ ¡ PlaQorm ¡ Implemented ¡for ¡the ¡Android ¡OS. ¡ ¡ Evalua8on ¡ It ¡takes ¡not ¡more ¡than ¡61 ¡seconds ¡for ¡mining ¡associa5on ¡rules ¡ from ¡1500 ¡no5fica5ons ¡for ¡any ¡features. ¡ ¡ Released ¡as ¡an ¡open ¡source ¡project ¡ hbps://github.com/AbhinavMehrotra/PrefMiner ¡ ¡

slide-31
SLIDE 31

In-­‑the-­‑wild ¡Evalua5on ¡ ¡

slide-32
SLIDE 32

PrefMiner ¡

1 ¡ 2 ¡ 3 ¡ 4 ¡

slide-33
SLIDE 33

PrefMiner ¡

1 ¡ 2 ¡ 3 ¡ 4 ¡

slide-34
SLIDE 34

PrefMiner ¡

1 ¡ 2 ¡ 3 ¡ 4 ¡

slide-35
SLIDE 35

PrefMiner ¡

1 ¡ 2 ¡ 3 ¡ 4 ¡

slide-36
SLIDE 36

PrefMiner ¡

1 ¡ 2 ¡ 3 ¡ 4 ¡

slide-37
SLIDE 37

In-­‑the-­‑wild ¡Evalua5on ¡Results ¡ ¡

Par5cipants: ¡18 ¡ ¡ Dura5on: ¡15 ¡days ¡(minimum) ¡ ¡ ¡ Data ¡of ¡2 ¡par5cipants ¡was ¡not ¡considered ¡because ¡they ¡never ¡ accepted/dismissed ¡any ¡suggested ¡rule. ¡ ¡

slide-38
SLIDE 38

Accepted ¡Rules ¡

80% 79% 77% 75% 73% 73% 67% 64% 57% 50% 50% 47% 45% 34% 25% 20%

5 10 15 20 25 30 35 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

User Rules Count

Accepted Rejected

Overall, ¡102 ¡out ¡of ¡suggested ¡179 ¡rules ¡were ¡accepted. ¡ 56.98% ¡acceptance ¡rate. ¡

slide-39
SLIDE 39

Accepted ¡Rules ¡

80% 79% 77% 75% 73% 73% 67% 64% 57% 50% 50% 47% 45% 34% 25% 20%

5 10 15 20 25 30 35 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

User Rules Count

Accepted Rejected

Overall, ¡102 ¡out ¡of ¡suggested ¡179 ¡rules ¡were ¡accepted. ¡ 56.98% ¡acceptance ¡rate. ¡ 70% ¡(11 ¡out ¡of ¡16) ¡users ¡accepted ¡50% ¡(and ¡above) ¡rules. ¡

slide-40
SLIDE 40

Filtered ¡No5fica5ons ¡

Average ¡filter ¡recall ¡for ¡16 ¡users: ¡45.81% ¡ ¡ ¡

20 40 60 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

User Filter Recall (%)

Filter ¡Recall ¡ N-­‑Auto ¡ N-­‑Auto ¡+ ¡N-­‑Manual ¡* ¡ = ¡ 100 ¡

N-­‑Auto: ¡Number ¡of ¡filtered ¡no5fica5ons ¡by ¡the ¡system ¡ N-­‑Manual: ¡Number ¡of ¡no5fica5ons ¡dismissed ¡manual ¡

slide-41
SLIDE 41

PrefMiner: ¡Exit ¡Ques5onnaire ¡

Triggered ¡aUer ¡15 ¡days ¡ Ques8on ¡ Average ¡ Response ¡ I ¡found ¡the ¡app ¡useful ¡for ¡learning ¡my ¡ preferences ¡to ¡filter ¡no5fica5ons. ¡ 4.25 ¡ The ¡app ¡filtered ¡most ¡of ¡the ¡no5fica5ons ¡ that ¡I ¡didn’t ¡want ¡to ¡receive. ¡ 4.33 ¡ The ¡app ¡incorrectly ¡filtered ¡no5fica5ons ¡ that ¡I ¡wanted ¡to ¡receive. ¡ 1.58 ¡ ¡ 1:Strongly ¡disagree ¡-­‑ ¡5:Strongly ¡agree ¡

slide-42
SLIDE 42
  • First ¡in-­‑the-­‑wild ¡study ¡for ¡automa5c ¡extrac5on ¡of ¡rules ¡that ¡

reflect ¡user’s ¡preferences. ¡ ¡ ¡

  • An ¡open-­‑source ¡library ¡(MyPref) ¡that ¡can ¡learn ¡the ¡type ¡of ¡

no5fica5ons ¡that ¡users ¡would ¡prefer ¡to ¡receive ¡on ¡their ¡mobile ¡ phones ¡in ¡specific ¡contexts. ¡ ¡ ¡

  • PrefMiner ¡represents ¡a ¡very ¡effec5ve, ¡yet ¡transparent, ¡solu5on ¡

for ¡interrup5bility ¡management ¡for ¡mobile ¡devices. ¡ ¡

Take ¡Aways ¡

42 ¡

slide-43
SLIDE 43

Ques5ons? ¡

Abhinav ¡Mehrotra ¡ ¡ Email: ¡a.mehrotra@ucl.ac.uk ¡ ¡ GitHub: ¡hbps://github.com/AbhinavMehrotra ¡ ¡ Play ¡Store: ¡hbps://play.google.com/store/ apps/details?id=com.nsds.prefminer ¡