MobileMiner: Mining Your Frequent Behavior Patterns On Your Phone

Slides:



Advertisements
Similar presentations
FixtureFinder: Discovering the Existence of Electrical and Water Fixtures Vijay Srinivasan*, John Stankovic, Kamin Whitehouse University of Virginia *(Currently.
Advertisements

D u k e S y s t e m s Sensing Meets Mobile Social Networks: The Design, Implementation and Evaluation of the CenceMe Application Emiliano Miluzzo†, Nicholas.
Understanding Human-Smartphone Concerns: A Study of Battery Life Denzil Ferreira, Anind K. Dey, Vassilis Kostakos Pervasive 2011.
ACE: Exploiting Correlation for Energy-Efficient and Continuous Context Sensing Suman Nath Microsoft Research MobiSys 2012 Presenter: Jeffrey.
Mining Motion Sensor Data from Smartphones for Estimating Vehicle Motion Tamer Nadeem, PhD Department of Computer Science NSF Workshop on Large-Scale Traffic.
Failure Avoidance through Fault Prediction Based on Synthetic Transactions Mohammed Shatnawi 1, 2 Matei Ripeanu 2 1 – Microsoft Online Ads, Microsoft Corporation.
Energy Model for Multiprocess Applications Texas Tech University.
A Social Help Engine for Online Social Network Mobile Users Tam Vu, Akash Baid WINLAB, Rutgers University May 21,
A Survey of Mobile Phone Sensing Michael Ruffing CS 495.
SoundSense: Scalable Sound Sensing for People-Centric Application on Mobile Phones Hon Lu, Wei Pan, Nocholas D. lane, Tanzeem Choudhury and Andrew T. Campbell.
Sensing Meets Mobile Social Networks: The Design, Implementation and Evaluation of the CenceMe Application Emiliano Miluzzo†, Nicholas D. Lane†, Kristóf.
Ambulation : a tool for monitoring mobility over time using mobile phones Computational Science and Engineering, CSE '09. International Conference.
Type of Software There are two main types of software They are System software Application software Hardware System Software (OS) Application Software.
Information-Based Building Energy Management SEEDM Breakout Session #4.
Ontology-Driven Automatic Entity Disambiguation in Unstructured Text Jed Hassell.
Mobile Middleware for Energy-Awareness Wei Li
© 2008 Quest Software, Inc. ALL RIGHTS RESERVED. Perfmon and Profiler 101.
Collaborative Information Retrieval - Collaborative Filtering systems - Recommender systems - Information Filtering Why do we need CIR? - IR system augmentation.
Contextual Ranking of Keywords Using Click Data Utku Irmak, Vadim von Brzeski, Reiner Kraft Yahoo! Inc ICDE 09’ Datamining session Summarized.
WEST VIRGINIA UNIVERSITY Lane Department of Computer Science and Electrical Engineering CROWDSOURCED TRAFFIC MAP Team Members: Faculty Mentor: David Williams.
Rule based Context Sensing. Background Context sensing – Sensors in smartphone – Reacts based on operating condition Example – Location based reminder,
Web Mining Issues Size Size –>350 million pages –Grows at about 1 million pages a day Diverse types of data Diverse types of data.
Sensing Meets Mobile Social Networks: The Design, Implementation and Evaluation of the CenceMe Application Emiliano Miluzzo†, Nicholas D. Lane†, Kristóf.
The Future of Mobile E-Health Application Development Exploring HTML5 for Context-aware Diabetes Monitoring Speaker: Nishant Chettri.
July 2013 Elastic Offloading by Dale Denis. Dale Denis The Elastic Offloading of Computationally Intensive Tasks to the Cloud to Augment the Computing.
Predicting the Location and Time of Mobile Phone Users by Using Sequential Pattern Mining Techniques Mert Özer, Ilkcan Keles, Ismail Hakki Toroslu, Pinar.
Identifying “Best Bet” Web Search Results by Mining Past User Behavior Author: Eugene Agichtein, Zijian Zheng (Microsoft Research) Source: KDD2006 Reporter:
Power Guru: Implementing Smart Power Management on the Android Platform Written by Raef Mchaymech.
Enhancing Mobile Apps to Use Sensor Hubs without Programmer Effort Haichen Shen, Aruna Balasubramanian, Anthony LaMarca, David Wetherall 1.
Tom Lovett and Eamonn O’Neill Department of Computer Science University of Bath Bath BA2 7AY UK +44 (0) Social sensing:
Item Based Recommender System SUPERVISED BY: DR. MANISH KUMAR BAJPAI TARUN BHATIA ( ) VAIBHAV JAISWAL( )
Introduction to Machine Learning, its potential usage in network area,
WP2 - INERTIA Distributed Multi-Agent Based Framework
Lecture 1: Getting Ready
Android Mobile Application Development
Understanding and Improving Server Performance
Recommender Systems & Collaborative Filtering
EMERALDS Landon Cox March 22, 2017.
Success Stories.
SAP SuccessFactors extension with SAP HANA Cloud Platform Innovation Use Case SAP & Partner Confidential
Green cloud computing 2 Cs 595 Lecture 15.
Outline Introduction Related Work
OPERATING SYSTEMS CS3502 Fall 2017
Walk n’ Play Group #8 - Team Murali Krishna Goli Viswanath Patimalla
Introduction to Computers
Vijay Srinivasan Thomas Phan
Presented by: Vijay Srinivasan (Samsung Research America)
Presented by: Vijay Srinivasan (Samsung Research) Collaborators:
Mining and Analyzing Data from Open Source Software Repository
Microsoft Ignite /14/ :21 AM BRK2101
HyperLoop: Group-Based NIC Offloading to Accelerate Replicated Transactions in Multi-tenant Storage Systems Daehyeok Kim Amirsaman Memaripour, Anirudh.
Hui Chen, Shinan Wang and Weisong Shi Wayne State University
Semiconductor Manufacturing (and other stuff) with Condor
I don’t need a title slide for a lecture
What's New in eCognition 9
CPU SCHEDULING.
Characterizing Smartwatch Usage In The Wild
Technical Capabilities
We Need Your Feedback Please complete your learning evaluations
John H.L. Hansen & Taufiq Al Babba Hasan
Actively Learning Ontology Matching via User Interaction
M. Kezunovic (P.I.) S. S. Luo D. Ristanovic Texas A&M University
RTC RIDE Service Improvement Recommendations
MAPO: Mining and Recommending API Usage Patterns
Volunteer Impact Database Training
Yining ZHAO Computer Network Information Center,
What's New in eCognition 9
What's New in eCognition 9
Power Consuming Activity Recognition in Home Environment
Case Study: Choosing an Exercise Mode in a Heart Rate Monitor
Presentation transcript:

MobileMiner: Mining Your Frequent Behavior Patterns On Your Phone Vijay Srinivasan, Saeed Moghaddam, Abhishek Mukherji, Kiran K. Rachuri, Chenren Xu, Emmanuel Munguia Tapia Advanced Software Platforms Lab Samsung Research America - Silicon Valley Replace samsung logo

Patterns From Your Phone “What are you doing now?” Detect Current Context “What do you typically do?” Mine Co-Occurrence Patterns “Which contexts typically happen around the same time?” “Typically, when I am home on Sunday nights, I call my parents” App Usage Activities and Places Lots of research in ubc comm on wot u r doing now Much faster Pre-condition Post-condition Association rules on multi-modal mobile context Call / SMS

Why Mine Co-Occurrence Patterns? Intuitive UI Pre-Loading Project Better explain pattern vs automation Show arrows one by one Automate frequent actions Users can review and select patterns to automate

Why Mine On the Phone? 2. Regions with no cloud or network access 1. Personal data privacy Network access? Think about population? 45 hours idle time per week on average Octa core & quad core processors 3. Powerful phone processors 4. Lengthy phone idle times

Outline MobileMiner system design Evaluation Is it feasible to mine patterns on the phone? How do we use co-occurrence patterns? Graying out Think about getting rid?

Rule Mining on Mobile Context Data Time Mobile Context Logs Support: Proportion of input context baskets in which a frequent itemset occurs Lowering support threshold increases mining time Morning Afternoon Home Work Jazz Confidence: P( post-conditions | pre-conditions ) CNN Call Bob Morning, Home Morning, Home, Jazz, CNN … Basket Extractor Context Baskets Morning, Home ( 10% ) Jazz, CNN ( 1% ) Afternoon, Work, Call Bob ( 5% ) … Frequent Itemset Miner Use milder black Frequent Itemsets Morning → Home ( 80% ) Jazz → CNN ( 90% ) Afternoon, Work → Call Bob ( 70% ) … Rule Generator Association Rules Key: ( Support , Confidence )

Optimized Apriori Algorithm Time Mobile Context Logs Morning Afternoon Home Work Jazz 1-3 months of data CNN Call Bob 𝑩 Morning, Home Morning, Home, Jazz, CNN … 𝑪 𝟏 ∗𝐁 Basket Extractor Context Baskets 𝑪 𝟐 ∗𝐁 𝑪 𝐧 ∗𝐁 … Optimized Apriori Algorithm Morning Home Gmail … Morning Home Jazz … Morning, Jazz Morning, Home Jazz, CNN … Morning, Home Jazz, CNN … Give some avg numbers for ck b … Candidate 1-itemsets 𝑪 𝟏 Frequent 1-itemsets Candidate 2-Itemsets 𝑪 𝟐 Frequent 2-Itemsets Frequent n-Itemsets 𝒌 (𝑪 𝒌 ∗𝐁) is expensive! ( 𝑩 𝒂𝒗𝒈 =𝟏𝟏𝟒𝟐𝟕𝟓, 𝑪 𝒌 ~ 𝟏𝟎−𝟏𝟎𝟎𝟎𝟎) More than 4 hours to mine patterns

Weighted Basket Mining Time Mobile Context Logs Morning Afternoon Home Work Jazz CNN Call Bob 𝑩 ′ Weighted Context Baskets Morning, Home (2%) Morning, Home, Jazz, CNN (0.2%) … 𝑪 𝟏 ∗ 𝑩 ′ 𝑪 𝟐 ∗ 𝑩 ′ 𝑪 𝐧 ∗ 𝑩 ′ Weighted Basket Extractor … Weighted Basket Mining Morning Home Gmail … Morning Home Jazz … Morning, Jazz Morning, Home Jazz, CNN … Morning, Home Jazz, CNN … Animate this better … Candidate 1-itemsets 𝑪 𝟏 Frequent 1-itemsets Candidate 2-Itemsets 𝑪 𝟐 Frequent 2-Itemsets Frequent n-Itemsets 𝑩 ′ << 𝑩 (92.5% size reduction) Mine patterns in 16.5 minutes ( 15 times improvement ) ( 𝑩′ 𝒂𝒗𝒈 =𝟖𝟓𝟓𝟗)

Context-Specific Mining Mobile Context Logs 1 minute Daily log 8 hours 6 hours Weighted Context Baskets Weighted Basket Extractor App Usage Baskets Outgoing Call Baskets … Filtering API Weighted Basket Mining Weighted Basket Mining Weighted Basket Mining (1/(24*60))*100 = 0.07% support 1% support 1% support 1% support Infeasible mining time of several days

Next Context Prediction Predicted Apps 92% Context Prediction Engine 90% Current Context Predicted Outgoing calls 80% 70% MobileMiner Association Rules Matching Rules Confidence Ranked app predictions → Outlook Add support value 70% 92% → 90% 90% 70% → 92% Prediction algorithm also ranks based on: Support value Number of pre-conditions in matching rule

Outline MobileMiner system design Evaluation Is it feasible to mine patterns on the phone? How do we use co-occurrence patterns?

Data Collection Details Mobile context data from 106 users 1-3 months of data from each user Start time & duration of context events 440 unique context events per user Call & SMS App usage Location Home, work, outside, … Getting rid of text under headings? Getting rid of status events Moving/not moving Charging Battery levels Location provider Moving / not moving Cell ID

Feasibility of Mining on the Phone Metrics Execution Time Memory CPU utilization Mobile Context Logs Weighted Context Baskets 1.4 seconds 11.6 MB 20.8% Basket Filtering 1.7 seconds 9.9 MB 22.9% Weighted Basket Extractor App Usage Baskets Weighted Basket Mining Weighted Basket Mining 16.5 minutes 44.2 MB 24.3% 21.2 seconds 1.0 MB 21.9% Combine extractor & mining? Remove energy? Just highlight 16.5 & 44.2 mb? Cpu utilization.. Rule Generation Rule Generation 1% support 1% support

Sample User Patterns & Potential Uses Post-conditions Put back animation Pre-conditions Charging or low power mode reminders Sensing place using cheaper attributes

Group Patterns & Potential Uses Post-conditions All 106 users Use color if time permits! Pre-conditions Bootstrapping with common patterns Group activity scheduling & recommendations

Next Context Prediction Accuracy App Prediction Service User preferences over 42 external users Tizen explain in 1st slide Tizen logo Explain numbers better Recall Precision Number of recommendations Similar trends for next contact prediction in paper

Conclusions Novel MobileMiner mobile system Weighted basket mining Context-specific mining Next context prediction Feasible to mine context patterns on the phone 16.5 minutes to mine 1-3 months of data Several applications of patterns Evaluated app & call prediction

Device Intelligence @ Samsung Research America Samsung Research America - Silicon Valley http://www.sisa.samsung.com/ ~1000 Researchers and Engineers 16 Labs (and growing!) Advanced Software Platform Lab Software innovations Core deliverables to Android and Tizen OS Open-Source Innovation Center Device Intelligence Group Core Expertise in Activity Recognition, Context-Awareness, Machine Learning, Mobile Computing, Device-Cloud coordination, Wearables University collaborations: Run MSM and find some interesting patterns. Cambridge CICESE Cornell EFPL Stanford Soft sensing Data Collection Crowd-sourced data Data fusion Mobile-Social

Thanks! Questions & Feedback?