Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Do Local Residents and Visitors Express the Same Sentiments on Destinations through Social Media?
1. Do Local Residents and Visitors
Express the Same Sentiments on
Destinations through Social Media?
M.Jabreel, A.Moreno, A.Huertas
Computer Science and Maths. Dept. + Communication Dept. -
Universitat Rovira i Virgili, Tarragona, Spain
Computer Science Dept., Hodeidah University, Hodeidah, Yemen
2. Motivation
Automated analysis of the polarity of
tweets sent by tourists and local citizens
Case study: 10 top destinations in Europe
Conclusions
Overview
3. Motivation
Automated analysis of the polarity of
tweets sent by tourists and local citizens
Case study: 10 top destinations in Europe
Conclusions
Overview
4. There is a huge volume of opinions about
destinations in social media, both from
citizens and from visitors.
DMOs are very keen on analyzing this
information, as it is very important on the
design of a proper destination brand.
This work proposes a new automated
sentiment analysis system that can retrieve
and categorize the opinions of locals and
tourists through social media.
Introduction
5. Motivation
Automated analysis of the polarity of
tweets sent by tourists and local citizens
Case Study: 10 top destinations in Europe
Conclusions
Overview
6. Methodology
Selection of the destinations to be
analysed.
Retrieval and pre-processing of tweets.
Extraction of a set of features from each
tweet.
Determination of the polarity of each
tweet.
Comparative analysis of the sentiment of
the opinions of locals and visitors in the
chosen destinations.
7. Methodology
Selection of the destinations to be
analysed.
Retrieval and pre-processing of tweets.
Extraction of a set of features from each
tweet.
Determination of the polarity of each
tweet.
Comparative analysis of the sentiment of
the opinions of locals and visitors in the
chosen destinations.
8. Retrieval of tweets
twiQuery crawler
English, radius of
15 kilometers from
the city centre,
specific period of
time.
Two sets of tweets:
locals (3,000) and
visitors (3,000).
Info. of home city in
Twitter profile
9. Do tweets actually refer to the
destination?
Manual analysis of 400 random tweets sent by
local people from London.
292/400 (73%) were related to local aspects
of the destination.
97/400 (24.2%) were conversational tweets
or personal life comments.
11/400 (2.8%) explicitly referred to other
locations or global news.
10. Pre-processing of tweets (I)
Normalization:
URLs and user
mentions are
removed, tweets
are lower cased.
11. Pre-processing of tweets (II)
Normalization
Tokenization &
POS-tagging
Each tweet is
tokenized &
POS-tagged
(Ark Tweet).
12. Pre-processing of tweets (III)
Normalization
Tokenization &
POS-tagging
Negation:
The suffix “_NEG” is
added to all the
words that appear in
a negated context.
15. Features Extraction (III-Semantic)
Each word is
mapped to a
predefined semantic
cluster.
1000 clusters in
Ark Tweet
4960 n-gram
clusters in
Word2vec
16. Features Extraction (IV-Lexicon)
7 opinion lexicons
Global polarity
of the tweet.
Average
polarity of the
+/- terms
Score of the
last +/- term.
Max/Min +/-
score.
17. Classification
SentiRich uses a
Support Vector
Machine to classify
each tweet.
It was trained using
sets of tweets from
SemEval2013.
Accuracy 68-72%,
outperforming state-
of-the-art sentiment
analysis systems.
18. Motivation
Automated analysis of the polarity of
tweets sent by tourists and local citizens
Case study: 10 top destinations in Europe
Conclusions
Overview
19. Top 25 European destinations in 2014, according to
Tripadvisor.
Destinations whose DMO had sent more than 3,000
English tweets were selected:
Amsterdam, Athens, Barcelona, Berlin, Budapest,
Dublin, Edinburgh, London, Rimini and Vienna.
For each destination we retrieved the tweets sent in
English from the city centre in the high tourism season of
2014 and 2015 (June 15th-Sept 1st) and the low tourism
season of 2014 (Oct 15th – Dec 30th).
3000 local tweets and 3000 visitors’ tweets were
randomly chosen for each destination.
Selection of Destinations
20. The 6000 tweets of each destination (3000 local +
3000 visitors) were classified as positive, negative
or neutral by SentiRich.
Results are shown in the form of radial plots,
where each of the concentric circles represents
10% of the tweets.
Results
21. Athens
Dublin
Barcelona
London
Amsterdam
Vienna
Berlin
Budapest
Edinburgh
Rimini
Positive Normal Negative
Positive views are quite
homogeneous in all the
destinations (51 ~ 57%).
Exception: Rimini (38%).
Neutral comments range
from 18% (Edinburgh,
Dublin) to 30% (Athens)
Except Rimini (over 50%).
The range of negative
views is even wider, from
only 9% in Rimini to more
than 27% in Edinburgh.
Results - Local Residents’ Tweets
25. Locals mostly make positive comments about their
city in social media, but there is a fair share of
neutral and negative comments.
There are significant differences among the opinions
of local residents in different cities (Positive >>>
Negative in Athens/Amsterdam, highly Neutral in
Rimini, Negative > Neutral in Dublin/Edinburgh).
The percentages of positivity, negativity and
neutrality of visitors’ tweets in the different
destinations are much more homogeneous than
those of the local residents.
Comparative analysis
26. Motivation
Automated analysis of the polarity of
tweets sent by tourists and local citizens
Case study: 10 top destinations in Europe
Conclusions
Overview
27. It is feasible to build a sentiment analysis
system that can analyze large amounts of
short messages on social media about a
given destination in real time.
It is possible to compare the opinions on
different destinations or the views of locals
and tourists.
Conclusions
28. Future work
Remove tweets that do not refer to the
destination.
Make a more fine-grained analysis of
positive/negative/neutral opinions on
different attributes of the destination.
Develop a web-based tool that can make
this analysis automatically in real time.
29. Do Local Residents and Visitors
Express the Same Sentiments on
Destinations through Social Media?
M.Jabreel, A.Moreno, A.Huertas
ITAKA-Int.Tech. for Advanced Knowledge Acquisition
Computer Sc. and Maths. Dept. + Communication Dept.
- Universitat Rovira i Virgili, Tarragona
Computer Science Dept., Hodeidah University, Yemen