Analysing Parallel and Passive Web Browsing Behavior and...
Transcript of Analysing Parallel and Passive Web Browsing Behavior and...
![Page 1: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/1.jpg)
Analysing Parallel and Passive Web Browsing Behaviorand its Effects on Website Metrics
Christian von der WethEmail: [email protected]
![Page 2: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/2.jpg)
August 11, 2014
2
Online Browsing Behavior
Potential benefits● Improving design and usability of websites and browsers● Assessing the popularity of websites● Advancing ranking algorithms for search engines
Emerging and rising trends affecting browsing behavior
● Passive browsing (e.g., listening to online radio while cooking)● New Web technologies (e.g., Ajax, WebSockets)● Evolving Web demographics (e.g., “Facebook Generation”)● Browsing while on the go
![Page 3: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/3.jpg)
August 11, 2014
3
Related Work
Server-side studies● Analysis of Web server or search engine transaction logs● Limited to analyzing click streams or revisitation behavior
● Insufficient granularity and detail of collected data
Client-side studies● Special browsers or ass-ons to capture browsing behavior● Typically conducted as lab studies investigating specific tasks
● Unsuitable to elicit everyday browsing behavior
![Page 4: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/4.jpg)
August 11, 2014
4
DOBBS in a Nutshell
DOBBS = DERI Online Browsing Behavior Study● Client-side approach, but unsupervised field study
Core: Browser add-on● “install-&-forget” application● Logs wide range of events● Sends events to server
Important features● Non-intrusive● Anonymous● Privacy-preserving Central DB
...new tab openednew page loadedwindow maximizeduser inactiveuser activelink clicked...
![Page 5: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/5.jpg)
August 11, 2014
5
Privacy Preservation
Applied techniques to preserve users' privacy● Complete anonymisation (user = random integer)● Encryption of all sensitive information (i.e., URL data)● User in full control the stop logging at any time● No logging of key strokes and explicit user input
We have nothing to hide● Project website with all details and dataset for download● Add-ons are open-source under very open BSD license:
http://code.google.com/p/deri-dobbs/
![Page 6: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/6.jpg)
August 11, 2014
6
Technical Limitations
Problems beyond our means to avoid● Network / connection failures● Browser errors (crashes or other bugs)● Unexpected termination (e.g., to do SIGTERM / SIGKILL)
Incomplete data unavoidable
Two basic approaches to deal with incomplete data● Filtering out affected session information● Adding estimates for missing data
![Page 7: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/7.jpg)
August 11, 2014
7
Evaluation – Focus of This Work
Parallel browsing● Usage of tabbed browsing and/or multiple browser windows● Switching between different tabs
Passive browsing● Times user a inactive / idling while browsing the Web● Two means to measure idle times
Explicit: special events fired by browser Implicit: prolonged absence of logged events
Effects of parallel and passive browsingon quantifying the popularity of websites
![Page 8: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/8.jpg)
August 11, 2014
8
Parallel Browsing: Windows vs. Tabs
Parallel browsing as common phenomenon● Tabbed browsing particularly common● Degree of parallel browsing very different across users● For this data: multiple windows XOR multiple tabs
![Page 9: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/9.jpg)
August 11, 2014
9
Parallel Browsing: Re-using Open Tabs
Main results● Most tabs used for one or very few “rounds”● Not shown: 6% of loaded pages were never visible● Large difference regarding re-using tabs for multiple page loads
avarge number of page loads per session
![Page 10: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/10.jpg)
August 11, 2014
10
Passive Browsing (1)
Session duration vs. idle time● The longer a session, the longer a user is idling● Idle time quickly dominates over active time
Passive browsing very common phenomenon
avarge number of page loads per session
![Page 11: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/11.jpg)
August 11, 2014
11
Passive Browsing (2)
Idle time as interesting metric● Different methods to quantify users' idle time applicable● Different methods describe different aspects of behavior
Important: careful selection of method and careful interpretation of results
number of sessionswithin clock hour
average session length
![Page 12: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/12.jpg)
August 11, 2014
12
Website Popularity
Main results:● Often loaded does not imply long on display
How absorbing is a website?● Long on display does not imply the user was active
How engaging is a website?
Novel notions for defining website popularity
![Page 13: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/13.jpg)
August 11, 2014
13
Reranking of Websites
New metrics to quantify the popularity of websites● Client-side phenomenons – tabbed browsing, idling, etc. –
do significantly affect “classic” rankings● Expressiveness of metric often depends on type of service
Alexa Visit time Page Loads How absorbing? How engaging?
1 Google (1) Google Facebook Facebook LinkedIn
2 Facebook (2) Facebook Google Twitter Facebook
3 YouTube (3) YouTube YouTube YouTube Twitter
4 LinkedIn (11) LinkedIn LinkedIn LinkedIn Google
5 Twitter (12) Twitter Twitter Google YouTube
![Page 14: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/14.jpg)
August 11, 2014
14
Outlook
Graph-based analysisof browsing sessions
● Nodes = page loads● Node size = loaded time● Edges = page navigation
Application of graphalgorithms
● Out/in-degrees● Shortest paths● Diameter● ...
![Page 15: Analysing Parallel and Passive Web Browsing Behavior and ...christianvonderweth.org/publications/talks/vdw-DOBBS-SeSaMe-talk.… · August 11, 2014 2 Online Browsing Behavior Potential](https://reader036.fdocuments.net/reader036/viewer/2022071017/5fd11804d24fea56f23ef03a/html5/thumbnails/15.jpg)
August 11, 2014
15
Conclusions
Lessons learned – apart from current results● Incomplete data unavoidable, but valid ways to deal with it● Abundance of data requires careful analysis & interpretation● Still a challenge: spreading the word● Capabilities of DOBBS go far beyond available datasets
How to participate● Install DOBBS add-on ( 20 seconds) – that's it
How to get started with the dataset● Download dataset from http://dobbs.deri.ie● Bundle includes useful scripts and example queries● ...or just contact me