4. What
> social profiles
> user posts
> user played music
Data set 1
Facebook user
statuses and posts
Data set 2
Last.fm listened
tracks
5. How
> sentiment analysis
> filtering
> cross-correlation
Sentiment analysis
Colours encode
user’s mood
Listening prefs
Tracks played are shown
for each time slot
Playlist generation
Playlist generated
according to moods
6. Evaluation process
> user study
Preliminary studies
User profiling
Information needs
Low-fi prototypes
Hi-fi prototype
User evaluation
On a working prototype
● Design evaluation
● Information gains,
user relevance
● Functionality
evaluation
7. Conclusions
> critical aspects
> future work
Moods detection
Minimum amount of data needed to
reliably extract emotional patterns
Single sign on
At present, signing in each of the two
SNSs is needed
Moods detection
Datasets could be further expanded
and more elements analysed to detect
users’ moods
Single sign on
Authentication through OpenId or
similar services should be implemented
8. Organisation
> individual work
Graham Hardie
Programming, data collection and data visualization
Viola Pinzi
Theoretical analysis, visual design and data analysis
Alessandro Piscopo
Theoretical analysis, visual design and data visualisation
10. The Social Thermometer
The Social Web - VU University Amsterdam
Group 2: Adnan Ramlawi, Sindre Berntsen, Yaron Yitzhak
11. Introduction
● Weather issues:
○ Too hot, too cold, too wet, et cetera
○ Does the weather affect people’s mood?
● Is there a correlation between:
○ Weather
○ Twitter sentiment
12. The application:
● Data used:
○ Tweets
○ Weather data (temperature, precipitation, cloudiness)
● Analysis:
○ Classification of tweets
○ Filtering
● Virtualization:
○ Average sentiment of tweets vs. weather elements (per
day)
○ ChartJS, Bootstrap
13. Code:
● How does the application work:
○ Long, Lat retrieval via Google Maps API
○ Weather data - World Weather Online (JSON).
○ Tweets - Twitter API (filtered by long,lat,lang,date)
■ Tweets re-formatted (JSON)
■ Tweets sent to Sentiment140 API
● Returned data is displayed in graphs using a
ChartJS script.
18. Introduction & Context
o Willingness to travel around the world
• Expensive
• Time to plan the trip (finding accommodation)
o Alternatives
• Couch surfing (accommodate to a stranger’s house)
o Our application:
• Leverage the hospitality of your friends
19. Goals
o Reduce the financial cost of exploration
o Motivate the traveler to explore new places
feeling safer
20. Approach & Method
o Extract data from user’s Facebook account
• User’s friends
• User’s friends name
• User’s friends photo
• User’s friends current location
• Personal friends lists
o Visualization
• Google Maps API
• Map
• Markers
o Provide travel details
• Google flights
• Skyscanner API
34. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5
Help user to find people with similar routes to their workplace
• Allows car pooling which saves fuel, reduces carbon dioxide emission and helps to
reduce traffic jams
• More social to ride with somebody else or use the car in case of bad weather
Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Purpose
38. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Approach
Magic
+
Travel Together Control Center
Building a community + reuse of existing data
Friendlist
Working and living place
Opening hours
Realtime updates
39. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Approach
Magic
+
Travel Together Control Center
Building a community + reuse of existing data
Working place
Opening hours
40. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Approach
Magic
+
Travel Together Control Center
Building a community + reuse of existing data
Realtime Updates
41. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Approach
Magic
+
Travel Together Control Center
Building a community + reuse of existing data
Working and living place
Workinghours
42. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Screenshots
Search-
and
Displayoptions
Resultsection
Option to share
on Facebook
and Twitter
X-Ray Mode
46. Achraf Belmokadem
Bernd Themann
Sheldon Pijpers
Group 5Evaluation
Burdon to join cummunity decreased
due to prefilled information and access
via Facebook account
Higher value for the user because even
not registered users are participating
„missuse“ of information
NLP techniques are really weak and
have a low accuracy
59. #motivation
#useful
An easy way to find free things via Twitter
You don’t need to search for Twitter accounts about free things
You don’t need to have a Twitter account at all!
#unique
There are several Twitter accounts that tweet about freebies
Gratweet collects all new tweets about freebies for you.
Unique in The Netherlands
60. #data
#what
Dutch tweets that contain the keyword ‘gratis’
Geographic coordinates of the tweets
Alternative: social web data from other resources such as Facebook
#pre-processing: filtering
Explicit tweets
Identical (re)tweets
Stopwords, meaningless words, personal pronounces
Timestamps, URLs
61. #approach
#algorithms
Assign specific weights to words surrounding the keyword ‘gratis’
#backend
Cache tweets using Twitter API and Tweet.JS
#frontend
Visualizations made with D3.JS, Jquery, CSS, HTML
65. Why this solution?
Our goal: Inform people on specific topics and how they
developed over time.
• People may not know what trending – or certain other –
topics are about on Twitter.
Our solution: Visualization of trending topics as word clouds
combined with insight on the explosion of tweets over time with
sentiment analysis if the tweets are about good or bad news.
66. Analysis of existing tools
• Twistori (sentiment keyword search) à
• We feel fine (feeling analysis) à
• I-logue (trending topic word cloud)
68. Approach
1. Use Twitter API
• GET search/tweets (Matthijs)
2. Use Python packages
• Textblob (sentiment analysis - Ans)
• Visualize sentiments of tweets over time in a cloud
• Pytagcloud (word cloud visualization - Lia)
• Extract tags based on word frequencies
• Important words are displayed larger
69. Smart part
• Filter out ‘meaningless’ words (e.g. ‘of’, ‘that’) and process
the ones that really matter
• Provide a condensed view of a trending topic in a word cloud.
• Sentiment over time: shows changing opinions
72. Our application
● Odd one out game using “likes” from Facebook.
● Retrieve small list of likes for a selection of
Facebook friend.
● Random pages(potential likes) are added to each
list.
● Player has to pick the odd one(s) out.
73. Our application
● Type: - Entertainment
- Raise awareness to other possible likes.
- Give insight to what friends like in an interactive and fun way.
● Scoping: - Only usable with a Facebook account.
- Facebook users who’s friends have enough likes.
79. Evaluation / Improvements
● Measurables: - Amount of users / games played per day
- Variations in users per day
- Users’ scores
● Future work: - Clustering for better matching of “likes”
○ Creates more variety in difficulty
- Add scores
○ Percentage correct on daily basis
○ Leaderboards, shared between friends
○ Makes users come back
80. Individual work
- Explore possibilities
Omer, Mustafa
- Retrieving and analysing Facebook data
Lennert, Omer
- Programming
Lennert, Mustafa
- Testing
Everyone
92. CARSIDEROR: Car Perception
Feature 1:
Positive/negative/neutral
classification (tweets)
For (potential) buyers & car manufacturers
G11
By Andreas Karadimas
100. Goal & Added value
Group 13: Thom Boekel, Rianne Nieland, Maiko Saan
Goal:
Helps you to find places to go to
based on popular places among your
friends.
Added value:
Information of friends might be more
interesting to you than reviews
available on the internet.
101. Data
Group 13: Thom Boekel, Rianne Nieland, Maiko Saan
Data source:
Facebook locations of friends
Wikipedia location information, future work
Size of data:
Information of all your friends, in our case:
140 friends (1819 locations) and 215 friends (2517 locations)
Type of data:
JSON files containing friends and locations (latitudes and longitudes)
102. Approach
Data collection
Gather friend
locations from
Facebook
Process
Categorize data on year
Filter out locations
without latitude and
longitude
Visualization
Heatmap with markers
Heatmap → number of
friends
Markers → all locations
Group 13: Thom Boekel, Rianne Nieland, Maiko Saan
103. Visualization (1/2)
Group 13: Thom Boekel, Rianne Nieland, Maiko Saan
Visualization type:
Google heatmap with location markers
Visualization of places:
Locations marked with
markers
Popularity of locations
indicated with colors and
radius
104. Visualization (2/2)
Group 13: Thom Boekel, Rianne Nieland, Maiko Saan
Options:
Filter locations by year
Heatmap options (e.g. radius)
Infobox with:
● information about the
location provided by
Wikipedia
● friend visits per year
105. Critical reflection
Pro’s:
● Filter on year
● Indication of popularity of a
location (heatmap)
● Able to perform pattern
analysis, e.g. Ziggodome
(number of visits increases
every year)
Con’s:
● Only locations your friends
have checked in or were
tagged
● Cannot see the names of your
friends
● Only information for locations
available on Wikipedia
107. Predicting the local elections
with Twitterdata
GROUP 14
Mabel Lips
Marco Schreurs
Wouter van den Hoven
108. Data & Approach
• Our data
• Collection of tweets of political parties and prominent politicians
• Size of data: ~15.000
• Approach
• Sentiment analysis
• Normalisation
109. Purpose of WebApp
• Predict the outcome of the local elections
• People of Amsterdam interested in politics
• Unique:
• Using realtime Twitter data
• Normalisation
110. Algorithms
• Sentiment analysis
• Pattern: python package with functionality for sentiment analysis
• SentiWordNet: Dutch sentiment lexicon (De Smedt and Daelemans, 2012)
Source image: http://jmlr.org/papers/volume13/desmedt12a/desmedt12a.pdf
111. Individual work
• Wouter: Twitterdata retrieval
• Marco: Sentiment analysis of Twitter data
• Mabel: Algorithm sentiment analysis and normalization process
118. General Features
• Memory-based collaborative filtering.
• Naive Bayes classifier to train on user’s timeline.
• Linear discriminant analysis: interesting vs. uninteresting.
• Continuous loop: retrieve Tweets and let user rate.
120. Feature Sarah
● Discovering and extracting recurring terms (i.
e. common subjects)
● Categorization and visualization of interesting
and uninteresting Tweets