This presentation talks about the user behaviours and the trend of mobile applications. It also talks about the behaviour of users downloading most common application on the market.
BDSM⚡Call Girls in Sector 71 Noida Escorts >༒8448380779 Escort Service
Mobile apps-user interaction measurement & Apps ecosystem
1. Mobile apps-user interaction measurement
Prepared by
Salah Amean Ahmmed Saeed
Source: 1-Falling Asleep with Angry Birds, Facebook and
Kindle –
A Large Scale Study on Mobile Application Usage, by Matthias
Böhmer
&
Apps ecosystem
Presented at
Prof. Choi Kae Won-Lab
1
2. Swiss Army knife
Communication
Social networking
Productivity
Sports
News
Games
Settings
Browsing
Travel
Shopping
Finance
Camera: video, etc
Etc.
2
3. Big Deal ?
Millions of Devices require apps.
Apps require developers
Developers Need to be paid
Apps need to be stored somewhere where
Huge demand on knowledge of usage behavior…(why?)
Essential for understanding usage pattern
In order to customize services
Provide specific users with specific apps recommendations
This knowledge could be utilized by app stores to cluster apps
according to the contextual knowledge
3
4. User Behavour
4
How long does each interaction with an app last?
Less than a minute
More than a minute and less than 5 minutes
Does this vary by application category?
Social networking versus alarm or weather app
If so, which categories inspire the longest interactions with their
users?
5. Contextual description
Example:
News app is highly probable to be checked morning
Games in the evening
From this we know that certain apps are likely to be used at specific
time
So far there is no user behavour study
5
6. Contextual description
How does the user’s context – e.g. location and time of day – affect
her app choices?
What type of app is opened first?
Does the opening of one application predict the opening of
another?
The interest of Böhmer et al.:
Is to provide data from a large-scale study that begins
to answer these basic app usage questions (especially those related to
contextual usage.)
6
7. Appazaar and AppSensor
7
Previous needs are translated as Usage pattern which are
gathered by Appazaar
Which was studied by Böhmer et al
Is used to recommend apps based on the Appstores gathered data
Based on the user’s current and past locations and app usage, the
system recommends apps that might be of interest to the user.
Within the Appazaar app AppSensor, that does the job vital to this
research of measuring which apps are used in which contexts.
9. Life cycle of mobile app
9
Determined by five states:
When a user Needs and App, then the user Install it
Remove it, if the user does not like it.
The app has two states in the eyes of users
Used or in the background
Switching is also an important factor to consider
weather to games
Weather to communication
Updates often are done automatically and most
Apps do not update frequently
10. What is the benefits of App life cycle
10
Since android OS system can report the most recently started
application
They enable us to observe app usage on a more fine-grained level
,and
Provide a much more accurate understanding of context’s effects on
app usage.
This OS report functionalities to be captured by the AppSensor
11. Formal Description of AppSensor
11
Let A is the available application on a devices
And all the application the user can interact with
Where A is the application in the devices and e is the rest of apps in
the App store
AppSensor provide the values:
AppSensor can know when a certain app is being used if
If app is change then the
18. Mobile Ecosystem -2nd paper
18
Mobile phones have gained a lot of popularity because of the
continues advancement of telecommunication Technologies
640 million device (active July 2012)
Apps for android seven fold increase
Hundreds of thousands of Apps on the App stores
This huge number, urges:
Understanding about the users grouping
Providing a systematic approach for pricing
19. Motivation of App Ecosystem
19
Analysis of Apps popularity
Measurement of Apps popularity
Modeling Apps popularity
Explore pricing
20. Zipf Distribution- Wikipedia
20
Brown Corpus
The most frequently occurring word ( 7% out million 69,971)the
Then the second most frequent is half (3.5%344985.5)
x is rank of a word in the frequency table; (diagram in next slide)
y is the total number of the word’s occurrences.
Most popular words are "the", "of" and "and",
Zipf's law corresponds to the upper linear portion of the curve
24. Pareto Effect-wiki
24
80% of a company's complaints come from 20% of its customers
80% of a company's sales are made by 20% of its sales staff
Microsoft noted that:
Fixing the top 20% most reported bugs, 80% of the errors and crashes
would be eliminated
90% of apps download comes for only 10% of Apps( Really )
Big chance that users in same area tends to use same apps
Sharing of new and good apps
Trusting others review make a big difference
26. Data collection
26
Data are gathered from four android market place
AppChina and Anzhi
are very popular appstores located in China, with more than 50,000
apps each
27. Data collection
27
Data are gathered from four android market place
AppChina and Anzhi (Chinese) with more than 50,000 apps each
SlideMe is one of the oldest Android marketplaces, founded in 2008,
containing more than 20,000 apps
1Mobile is one of the largest third-party appstores,
with more than 150,000 apps.
29. Clustering
29
Appstores are organized into groups of related apps
E-book
News
Games
Etc.
User focus on a single or few categories
Downloading news app and ignore finance app
This implies the clustering effect
App download is affected by the popularity of this App
Number of downloads, comments
E.g., one could download few games more than tools
30. Temporal affinity
30
To validate the clustering effect
This study is based on users comments and rating gathered from
Anzhi Appstore
During the crawling process from the Appstore,
stream of comments were recorded for each user on each app
They suppressed successive comments so
if(a1,a2,a3,a3,a1,a4)(a1,a2,a3,a4)
(a1,a2,a3,a4) is called App string
In the Appstores, there are categories like 푎푖
So (a1,a2,a3,a3,a1,a4) is categorized into category string c(a1), c(a2),
c(a3), c(a4).
S is String category of n elements c1,c2,c3,… 푐푛
with respective n comments for the user( in chronological order)
31. 31
Temporal affinity- continued
We define Temporal affinity metrics Aff as the number of elements
in the same category, with the previous element divided by n-1
from the formula if Aff is 1,
then they are in the same category
Example:
category string c1c1c1c1, the Aff is 3/3, the user tends to comment on
App from the same category
string c1c1c1c2, the Aff is 2/3
string c1c1c2c3, the Aff is 1/3 user switched from category to another
32. 32
Temporal affinity- continued
Pattern like c1,c2,c1,c2 are solving using affinity depth
To make sure all the elements from the same category
Are classified together
33. 33
Temporal affinity- Random walk
In practice, App are not evenly distributed among categories
calculate the accurate affinity probability of a random walk in the
Anzhi marketplace
They used the actual distribution of apps to the C different
categories
Let A be the total number of apps in the appstore
A(i) the number of apps that belong to category i.
Given this distribution, the random walk affinity probability
Continued..
34. 34
Temporal affinity- Random walk-continue
퐴푓푓푟푎푛푑표푚 푤푎푙푘 , i.e., the probability that two random app choices
belong to the same category, is equal to:
A*(A -1) possible random app choices
The number of app choices where these two apps belong to the
same category is
35. User behavour based on APP-Clustering
35
1. Download the first app according to the ZG distribution.
2. Download another app:
2.1. with probability p the app will be downloaded from the same
cluster c of a previously downloaded app.
The cluster c is randomly chosen from previous downloads with a
uniform probability.
The app from cluster c is drawn from distribution Zc.
If the app has been downloaded go to 2.1.
2.2. with probability 1 − p the app will be drawn from ZG.
If the app has been downloaded go to 2.2.
3. If user’s downloads are less than d go to 2.
36. 36
Predicted downloads for app with total rank i and rank j in its
cluster:
The overall downloads for an app equal to
The probability of all users downloading this app
For a single user equals:
The probability 1-(zipf based(1-p)*d) times p*d cluster