Twitter Cities: a study of Twitter as a city's attention barometer

Team members: Alexander Guian-Illanes,Max Cantellow, Leonid Goryachev, Anissa Jousset, Ashraf Monzer, Jonas De Meulenaere, Milo van Bokkum, Nadia Dresscher, Carlo De Gaetano and Xabier Landabidea


  1. Introduction
  2. Research Question(s)
  3. Methodology
  4. Limitations
  5. Findings
  6. Discussion
  7. Conclusion
  8. Bibliography


“Twitter is your window to the world” - Twitter

Technological evolution has allowed for the essence of a city to permeate across the globe, with a cityscape’s physical infrastructure being replicated in the virtual world through the medium of Twitter. This causes a rise for a greater understanding of how the physical and the virtual world can be conceptualised in relation to one another, as social media networking sites are increasingly entangling urban space . This hybridisation of the urban space sees the city being recreated as ‘a digital double’ (Owen, Imre) creating a (re)reproduced construct through quintessent images, descriptions and a perspicacious eye of an individual on their social media channel. The benefits of these modes of expression can provide us with an invaluable insight of socio-cultural influence within a highly dense and populated environment. Harnessing the real time nature of Twitter provides us with a strong foundation in which we can use it to understand the relationship between key events and how they are documented on the Tweetosphere.

Initially, we had chosen three cities, Chicago, Brussels and Paris; however, due to tool limitations we were unable to continue with Chicago and changed our third city to Mexico City. These cities were chosen primarily due to their variety in language (English, French & Flemish), while two of which were then refined based on their geographical proximity. Given the real time nature of Twitter, our selection of cities was influenced by events taking place during a selected timespan, initially trying to find global events in which we could see them being (re)reproduced through the medium of Twitter. Therefore we chose 13th November 2015 to 13th December 2015, which encompasses numerous incidents between its thirty day span, including the Paris Attacks, COP21 and other various political and entertainment events.

The objective of this research is to identify how key events over the period 13th November 2015 - 13th December 2015 are replicated on Twitter in Brussels, Paris and Mexico City. Measuring the engagement of users on Twitter within the metropolis, we hope to gain an invaluable insight on the role Twitter can play as a social tool, and as a barometer to further our understanding of a city in which it is a junction of both the the niche and social (Severo & Kergosien, 2015). In this, we will be relying on Twitter’s characteristics as a radio scanner and dispatch enthusiasm (Rogers,2013) which translate the immediacy and real time nature of Twitter. In order to do this, individual posts with our chosen hashtags, that are geographically located in specific cities will be extracted and analysed, to draw connections between what topics, places and events may be shaping communication via the social network, Twitter. By doing this, we hope to begin to construct a means to understand the way occurrences interconnect a city in the digital sphere. Twitter will be examined as a social tool in the sense that it places itself in the center of the social sphere, by showing what topics are of high interest to users tweeting about a particular city.

Research Question(s)

How are cities mapped on Twitter through patterns of user engagement?

Using Twitter as a data source and the specific city as a dataset, we aimed to learn about the attention of our chosen cities in terms of:
  • Which topics are generating user activity?

  • How does this city relate to other places?

  • What events are mentioned in the tweets?


We identified three metropoles: Chicago, Mexico City and Paris. However, due to tool limitations we replaced Chicago with Brussels and decided to measure the level of users’ interests on Twitter between two geographically close cities (Paris and Brussels) and a distant city (Mexico City). Therefore, the cities that are being studied in this paper are: Brussels, Paris and Mexico City and the time period is between November 13th, 2015 and December 13th, 2015.


Overall Tweets

Distinct Users







Mexico City




For this project, we used TCAT to retrieve data for our three cities. Taking into consideration the overall amount of tweets during the 30 day period for every city, we decided to retrieve the monthly hashtag frequency for Brussels and Mexico City, with a minimum frequency of 10 while for Paris with a minimum frequency of 50. Similarly, we retrieved weekly hashtag frequency, selection of tweets and users activity using the same method.

Visualizing Geo-located Hashtags with Google Maps

We utilized Google Maps to visualize the list of hashtags retrieved with the mentioned frequencies in the previous section. The hashtags were divided to two different categories: Places and Events. The “Places” category refers to the precise latitude and longitude of a physical place, while the “Events” category refers to an event that took place in one of these three cities. Google Maps cannot identify the places of the events, therefore, we had to manually create an extra column in the data folder to feed Google Maps with the places these events took place.

Opinion Identification and Analysis

We attempted to identify the opinions of users through the hashtags retrieved for the three cities under study (Brussels, Paris and Mexico City). Two data folders were utilized, the first data folder contained four inputs: Geo-Hashtags; Frequency; Category and Real Location, while the second data folder contained Content of Tweets. The method to identify users’ opinions was based on two English lexicons: one containing positive terms (love, win, etc.), another containing negative terms (attack, war, loose, etc.). The ratio in which tweets were labeled negative, neutral or positive can be visualised as follows: a positive term = +1 while a negative term = 0. For each tweet there was an OpinionValue which is the average of positive and negative terms. For each Geo-Hashtag there was an OpinionValue which is the average of OpinionValue of the tweets containing that particular geo-hashtag. The scale goes as follows: A positive geo-hashtag is between 0.60 and 1. A neutral geo-hashtag is between 0.40 and 0.60. A negative geo-hashtag is between 0 and 0.40

Topic Modeling

The topic modeling tool is a graphical user interface tool in which researchers can harness to identify topics within small and big texts. The tool learns and connects topics related to each other and tags each document or set of data with a small number of topics ( Topic Modeling Tool). This tool provides a way to understand and analyze large volume of text, then the tool labels topics in cluster of related words that are constantly present together in the same sentence. The tool connects words with similar meanings and differentiates between words with multiple connotations through contextual clues.

We utilized the topic modeling tool to get an idea of the most common topics discussed in a specific city as a way to identify the top trending topics during a particular timeframe. For the Topic Modeling tool, we extracted the set of tweets from TCAT using the same settings as those for the monthly/weekly frequency datasets. The Tweets were then cleaned from hyperlinks and the Twitter handle sign (@), upon completion the file was uploaded to the topic modeling tool.

Noise Cancellation

During the first stages of data gathering and analysis there was a need for filtering the data set. One of the specifics of Twitter as a platform is that it has a high amount of noise, spam and promotional information; so in order to study particular topics, trends or issues those distractions needed to be accounted for, both with regards to the validity of the dataset and for the convenience of conducting further research. There are two ways of excluding noise, irrelevant tweets and spam utilising quantitative (using tools such as Microsoft Excel) and qualitative methods (manually categorised into themes in small groups). In this research we utilized both, so it may be called a quanti-quali approach.

Quantitative — judging by the ratio of tweets for a certain hashtag/category divided by the number of unique users, it is possible to identify entries which are either spam or promotional posts. If the same content is posted by a single user numerous times, this content is most likely irrelevant to the research (self-promotion, random tags etc.); if it is posted by a small group of users, it is likely a spambot network. For the dataset used in this research, the usual normal ratio of tweets/users fluctuates around 1.3, while for the spam/noise hashtags the ratio may be as high as 10.


Analysis of tweets/users ratio per category for geo_paris dataset for the week 05/12 — 13/12

As a result, some of the categories and groups of hashtags were excluded from the analysis as they did not provide any useful data for the research. In the studied dataset those categories are Consumerism (#reduction, #promo, #discount, #deal) and Sex/Spam (#webcam, #teen, #massagerotic).


Graph of tweets/users ratio per category for geo_paris dataset for the week 05/12 — 13/12

While manually determining categories within the highest frequencies of tweet/user ratios, the topics identified as noise broadened from promotional commercial content and sexism towards other morally and culturally ambiguous hashtags. Groups of hashtags such as “Geo” (featuring the names of geographical objects: #Paris, #France, etc.) or “Business” (#emploi, #job, #tech) may contain both noisy entries as well as some that are relevant for the research. This is the reason two methods of exclusion were used and qualitative analysis was utilized in order to further reduce the noise and make the dataset more useful and researchable.

The categories mentioned above were only used during the preliminary analysis in order to identify the most noisy topics/parts of dataset and were not a part of the further research.

Qualitative — this method requires manual investigation of sets of hashtags and user data. By reviewing all the most popular entries for the studied period,important issues are identified through the use of hashtags, while the least important ones (as well as some statistical outliers) were excluded from the dataset. Those are usually represented by the twitter-slang hashtags (#lrt, #rt, etc.) or the meta-entries (#trnd, #trndnl). While such entries do not seriously interfere with the research and in some cases are not really noise, some simply present the specifics of the studied platform. Their exclusion allows for a cleaner and more usable dataset in relation to our research question, especially considering the high frequency of such hashtags and their omnipresence.


Untitled 2.png


Our limitations varied from a lack of consistency between the groups’ work, a lack of unified data extraction and data processing to tool limitations forcing us to divert our initial scope of studies.


3.1 Monthly Hashtag Frequencies

3.1.1 Brussels

Initial findings for the city of Brussels show that the most frequent hashtags are #Bruxelles in French, or #Brussels in English. However, these hashtags are frequently used to gain exposure. #Brusselslockdown and #ParisAttacks were the most tagged hashtags after the geo-hashtags which showcases the users’ attention towards specific matters during this period. A noticeable factor within the top trending hashtags is the presence of Brussels identified in 4 different languages: English, French, Spanish and German.

3.1.2 Paris

The initial findings of Paris clearly demonstrate the presence of key events within the trends with highest frequencies of circulation. The attacks in Paris are represented through two major hashtags (#PrayForParis and #ParisAttacks), and the COP21 initiative equally carried momentum across the Tweetosphere. Hashtags such as #Paris and #trndnl offered little contextual background, as they simply function as a method for individuals to attract more traffic on their posts. As hashtag frequencies were delimited into weeks, interesting trend shifts become observable even within the highest percentile of utilised hashtags.

3.1.3 Mexico City

The initial findings of monthly hashtag frequencies demonstrate a clear presence of entertainment media in the Mexico City Tweetosphere. Major hashtags such as #MTVStar and #PremiosTelehit is frequency generated by television; notably of shows asking their spectators to participate in competitions, giveaways and providing opinion about the shows’ continuity. #1DMX is in relation to a protest having taken place in the capital to free a group of missing students.

3.2 Weekly Hashtag Frequencies

3.2.1 Brussels

From the 13th to the 19th of November 2015, the top three hashtags for Brussels were #ParisAttacks, #Brussels, and #Paris in reference to the immediacy of breaking news and the events that took place in Paris. The topic of the attacks in Paris is the most prominent hashtag, and demonstrates a common rise in solidarity for the city of Paris. #TodosSomosParis demonstrates a noticeably high proportion of Spanish speaking users present in Brussels, engaged in current events. #Molenbeek is a hashtag related to the journalist abused in Sint-Jans Molenbeek.

From the 21st to the 27th of November 2015, there is a drastic change in hashtag topics. #Bruxelles, #Brussels, and #BrusselsLockdown have become the dominant three hashtags whereas the hashtag #ParisAttacks has decreased in frequency. The focus has shifted to Brussels and the repercussion of the Paris attacks on the Belgian city specifically the lockdown which began that week. #Molenbeek is still present within the top ten highest trending hashtags.

From the 28th of November to the 4th of December 2015, #BrusselsLockdown, #Bruxelles, and #Brussels were the most three frequent hashtags. This expresses the level of concern of the Belgian population related to the lockdown of the city in Brussels and the repercussions of the Paris attacks. However, one can see the noticeable reduction in hashtag frequency for #Paris & #ParisAttacks.

From the 5th to the 11th of December 2015, the hashtags linked to the Paris attacks and the Brussels lockdown have become filtered out of the Tweetosphere and do not appear in the weekly top 10 hashtags. Similar to previous weeks #Brussels, #Belgium, and #Bruxelles remain in as the top three most cited hashtags.

3.2.2 Paris

From the 13th to the 20th of November 2015, a great deal of Paris attack-related hashtags are visible within the top 10. The week following the tragic event in the Bataclan, generated a large level of collective engagement which overwhelmed Twitter with over 8000 mentions referring to the attacks. Trending topics of discussion relate to messages, thoughts, and questions trying to make sense of the tragedy happening in Paris. Hashtags #JeSuisParis and #PrayForParis and other related hashtags begin to demonstrate the conscious solidarity amongst online users in the city where they dominate the top 10 hashtags list.

From the 21st to the 27th of November 2015, the repercussions of the attacks in Paris are still present at the top of the Tweetosphere. #ParisAttacks, #PrayForParis and #AttentatsParis are still within the top ten hashtags of the highest frequency of circulation. The #SaintDenis hashtag equally rises to the podium of virally trending topics, which was caused by the police raid taking place there in the past week. There is nevertheless the presence of entertainment circulation over the course of week 47 within the top trending hashtags. #1dfr, a hashtag celebrating pop band One Direction’s arrival to France, and #LPDLA3, devoted to a french reality TV show airing a new season, witness a considerable amount of interest on Twitter.

From the 28th of November to the 4th of December 2015, the Tweetosphere a relatively important shift in the frequency of trending hashtags can be witnessed. Hashtags mourning and discussing the past months events in Paris have almost entirely dropped from the top 10 list. Only #hommagenational remains within the top 10, a hashtag carrying the lingering weight of France’s tragedies across Twitter. Week 48 is when the COP21 initiative begins to rise in popularity amongst users on the social network, in relation to the decrees made public in an effort to regulate climate change. #LPDLA3 remains a popular trend in discussion, and #DALSDisney rises in popularity as the entertainment program Danse Avec Les Stars invades national television with a Disney special. In terms of entertainment, the #HavasVillage hashtag also promoted itself as a touristic and national place of gathering through Twitter. A gift distribution machine was placed in Havas Village that would reward people with a gift after having geo-locally posted a tweet linked to the Village’s official twitter page.

From the 5th to the 11th of December 2015, hashtag frequencies visibly shift towards political and ecological incentives. The trending references to the Paris Attacks have subsided from dominating the Parisian Tweetosphere to being practically nonexistent. 2 hashtags, #COP21 and #climatechange, demonstrate the digital resonance of media coverage promoting ecological incentive as part of the COP21 initiative. The trending of the #Régionales2015 hashtag is in relation to the imminent regional political elections happening in France. #MTVStars has been present within the top 50 monthly hashtags throughout the month of research, and in week 49 amasses enough frequency to become present within the most popular weekly hashtags. #1DnoticeSandy is a public incentive on Twitter to promote a disabled woman’s wish to contribute to a One Direction video clip; the motion helped her transmit her birthday wish to the celebrities.

3.2.3 Mexico City

From the 13th to the 20th of November 2015, there is a shift in hashtags with a high amount of frequency that remain focused on topics related to entertainment and television. Interestingly, the highest circulated hashtag of week 46 is #YoNoTiroLaToalla, a motivational hashtag circulated amongst users as a reminder never to give up. The #OYELAZONADECODER hashtag is related to a Mexican radio show that triggers a lot of user-producer interaction via Twitter. #Viernes13 is related to the tragedies having occurred in Paris. #SuperCopaTelcePuebla highlights a car racing event taking place in the country.

From the 21st to the 27th of November 2015, there is another wide shift in frequency of popular hashtags. The previous trending of car racing and a radio show is reiterated towards more televised events and commercial promotion. #jaryLlevameALosPremios and #NataliaRegálameBoletos are promotional competitions held by the national television channel TeleHit through particular celebrities. #MTVStars also bears strong presence on Twitter during this week in Mexico City. The #CoronaCapital15 is related to Corona, one of the country's domestic beers.

From the 28th of November to the 4th of December 2015, there is yet again a shift in the major trending hashtags, mostly reiterated the environment of entertainment media. #RetoKit1D, #PremiosTelehit, #IreConJaryAPremiosTelehit and #RetoPMTelehit exemplify the dominant positioning of televised promotion through Twitter. Through various celebrities and giveaways, Telehit maintains a strong stance within Twitter circulation. Other considerable trends in the week include #BryanBallZ, a teenage musician/celebrity being photoshopped into a variety of Dragon Ball Z images; and #PorUnPlatinoYo, which is a Telehit promotion to drive users to imagine what incentive would encourage them to earn a music award.

From the 5th to the 11th of December 2015, hashtags with the highest frequency of usage fluctuated towards new trends on Twitter, but conserved a majority of frequency related to television and radio entertainment. #MTVStars climbed back up the ladder to be the 2nd most utilised hashtag during week 49. At the top, with practically the double of hashtags devoted to MTV, #QuieroIrACozumel witnessed a sharp rise in usage. The hashtag was related to an event sponsored by AMEX and Aero Mexico, inviting Twitter users to compete online in order to have the chance to be sorted for tickets to travel on vacation. #TQMM and #sincontrato are hit songs written by local artists, bearing meaning as identifiers of cultural representation within individuals.

3.3 Geographical Opinion View

3.3.1 Brussels


This geographic opinion view showcases the impact of overall tweets and measures the opinion of each tweet containing specific hashtags. The scale ranges from 0 to -1: the closer the score is to -1 the more negative the opinions are, while the closer the score is to 1 the more positive they are. This geographic opinion view showcases the impact the overall tweets and measures the opinion of each tweet containing the hashtags over frequency 10 for Brussels. As seen in the graph, most of the opinions related to tweets that permeated through Brussels during the period of Paris Attacks were negative as they score 0.25. While the negative opinions visible also in the graph related to Brussels belong to the #BrusselsLockdown and the aftermath of the Paris Attacks.


Furthermore, this map shows the negative opinion of the tweets which can be geo-located to Brussels. Brussels experienced a large majority of negative tweets relating to Syria and a further negative referral to the presidential elections that were occurring in Venezuela.

3.3.2 Paris


Looking at the graph ( link for the interactive graph) if we take Syria as an example it has a score of 0.2, which means most of the tweets which included #Syria have a negative opinion. This can be linked to #ParisAttacks and how people judged Syria during the Paris Attacks period, knowing that the attacks were planned and executed by the Islamic State who is active in Syria.


While on the other hand, if we consider Disneyland within the same time frame, opinions mentioning #Disneyland as a hashtag scored 0.7 - which means most of the tweets have a positive opinion.


Furthermore, a third example shows that tweets posted from Paris have negative opinions regarding the #BrusselsLockdown’ with a total of -1 being the ultimate score for negative opinionated tweets.

3.3.3 Mexico City

Due to the lack of raw data and logistical mishaps we were unable to recreate a similar geographical opinion map for Mexico City.

3.4 Topic Modeling

In order to determine semantic categories linked to the top hashtags for each city, we made use of the Topic Modeling Tool provided by David New. It allowed us to process the set of geo-located hashtags in order to determine which words were associated with tweets from specific cities. We also used a list of stop words for each language such as pronouns, articles, and connecting words in order to refine and clarify our results. We used 300 iterations of the algorithm with 15 topic words displayed for each of the 10 topics and topic proportion threshold of 0.05.

Output settings of the Topic modeling ToolUntitled.png

The result of topic modeling is a set of keywords arranged in categories, which required further manual cleaning and arranging in order to present a clear set of topics. Here is the output of the Topic Modeling Tool:


It can be seen that the topics are not automatically arranged in a way that can be used in research, which calls for further manual investigation. After the rearranging and combining the keywords into the semantic categories, the following topics were identified:

Environment — entries associated with COP21, environmental activism and generally ecology concerns: world, energy, emissions, carbon, water, green, fuel, climate, talks, solutions;

Paris Tourism — usual Paris things, names of landmarks, museums, attractions and stereotypical French entities: paris, eiffel, louvre, city, tour, dame, triomphe, champs, palais;

Self-expression — very wide and meaningless category, which is apparently prevalent in all social media: good, love, happy, hope, work, time, together, great, night, ambition, best;

Business/Economy — words on topic of finance, career, etc.: business, development, job, apply, team, economy;

(Sl)activism — “change the world” category: global, world, peace, love, future, start, youth, tomorrow, united;

Reaction to Attacks — smaller category featuring the entries, associated with the Paris attacks: victims, attack, support, help, strong, pray.


4.1 Brussels

4.1.1 Fluctuation in Hashtag Frequency

Over the course of the month following the Paris attacks in 2015, online user engagement shifted towards messages of solidarity, concern and disarray. In Brussels, this was primarily exemplified through the hashtags #ParisAttacks and #BrusselsLockdown. A curfew was instilled in Belgium as a follow up to the terrorist threat in Paris, and the fact this threat trickled towards Brussels. This succession of occurrences disseminated a widespread atmosphere of tension and insecurity, which resonated throughout the Tweetosphere in Brussels. The Belgian capital was equally destabilized by events, for example the journalist assaulted in Molenbeek. Quantifying the sum of hashtags into categories enabled a clearer view of how dominant these topics of discussion were during the month of November 2015. Amidst this circulation of information related to terrorism and government intervention, Brussels also demonstrated a strong presence of multilingualism in its’ top trending hashtags. Although the corpus of tweets for Brussels contained only 155,000 entries, the top 10 monthly hashtags contained hashtags identifying the city in four different languages. This was the only city researched in this project that demonstrated a distinct lack of monolingual hegemony within a country. Breaking down hashtag frequencies into weekly quarters, the fluctuation of trending events coincides with the happening of current events at the time. The first quarter is highlighted by the tragedies in Paris and the assault in Molenbeek. By the second week, the Brussels Lockdown initiative has introduced another hashtag encapsulating the underlying tension in Belgian societies. The last week of the month, its research shows the rapid decrease in user engagement regarding such events; hashtags identifying events related to terrorist threat have become obsolescent.

4.1.2 Geo-located Events and Opinion Analysis

The geographic opinion view for Brussels showcased the response of users based in Brussels to several events that took place during the assigned period of time. In the period of Paris attacks, the tweets which came out from Brussels were mainly negative and mentioned several key hashtags such as Syria, Iraq, Turkey and Yemen. Syria and Iraq could have been mentioned due to the fact that the Islamic State who adopted the attacks in Paris currently has strong military activity in Iraq and Syria. However, Turkey could have been mentioned for being an active component of the war in Syria which the Islamic State has launched a few years ago. We can see clearly that the geographic proximity of Brussels to Paris has diverted the attention of Twitter users living in Brussels from their usual topics to discussing the incident that took place in Paris. Nonetheless, #BrusselsLockdown also had mainly negative opinionated tweets, as it was discussing topics that were perceived as negative. Furthermore, the Venezuelan election also had negative opinionated tweets coming out from Brussels due to the rallies that took place between the Mesa de la Unidad Democrática and Partido Socialista Unido de Venezuela.

4.2 Paris

4.2.1 Fluctuation in Hashtag Frequency

The wave of despair and sadness that overwhelmed France following the attacks in Paris became an important portion of the online circulation of information between users on Twitter. A number of top trending hashtags over the month amplified the topic, carrying its graveness and temporarily permeating it throughout digital space. On Twitter, user communication occurred in the first half of the month demonstrated a clear feeling of solidarity and involvement. Twitter users identified a number of hashtags to bear messages aimed at the atrocities having occurred in the city. As the weeks move on into December, the attention devoted to the attacks shift towards #hommagenational, a hashtag identifying tweets as mournful and in remembrance of the tragedy. The topic shifted since its first initial conception, a lapse of time, moved it from being a highly topical and contemporary event to one that lacked user attention. By the final week, such discussions on Twitter had almost subsided from the most popular hashtags in Paris. Instead, concern about climate change and the COP21 initiative built up a strong online presence in user interaction. The regional elections in France also reoriented discourse on Twitter towards political issues not immediately targeted at the war on terror. For such a major event, having occurred in a major Western capital, the frequency of debate and discussion arising in its’ succession appeared to have a limited range of projected significance, as everyday interaction on Twitter is triggered by current news and entertainment media.

4.2.2 Geo-located Events and Opinion Analysis

Through the analysis of geo-located tweets and opinion analysis with Google Maps, it became clear that Paris’s tweets reflected a negative opinion regarding the attacks on the city. These tweets mentioned similar countries as the ones mentioned in Brussels’ tweets, in addition, Tunisia and Algeria were added to the list, which reflects the cultural tension present in the city. Furthermore, tweets from Paris were also negatively opinionated when mentioning the #BrusselsLockdown hashtag. However, during the same period of time we were able to measure positive opinions surrounding DisneyLand Paris. The connection between the place, event and opinion becomes clearer. The content of the tweets is influenced by events taking place in the city, in other words, users are being influenced by the events taking place in their city which in turn is reflected in their tweets.

4.3 Mexico City

4.3.1 Fluctuation in Hashtag Frequency

Trending topics of discussion on Twitter in Mexico City demonstrated interaction mainly from entertainment media, such as television and radio. Mexico City has a media culture which is heavily influenced by radionovelas (stories broadcast on radio) and later telenovelas (soap operas); interestingly it was reaffirmed through the high frequencies of entertainment related hashtags. Throughout weeks 46 to 49 of 2015, hashtags referred to events such as Telehit promotional advertisements and radio shows. The events in Paris were identified solely through one major hashtag, #Viernes13, which only permeated through the first week of the research following the attacks. The 1DMX protest also had a poignant but short-lived place in the Mexican Tweetosphere. Observing a dominance of consumer based hashtags, Twitter engagement in Mexico City visibly has different implications than Brussels and Paris for example. Topics of interest and discussion online are much less representative of individual opinion but indicative of more self-centered forms of online activity.


Throughout this project we have aimed to explore the ways in which Twitter can be harnessed as more than just a social media channel but as a barometer for a city. We delved into the question of whether cities can be mapped on Twitter through patterns of user engagement. From our research on three core cities: Brussels, Paris & Mexico City, we can gauge that the top trending hashtags directly coincide with contemporary events in each city (i.e. #ParisAttacks, #COP21, #1DMX, #BrusselsLockdown). This implies that the foundation on which Twitter has been built upon, “Twitter is a window to the world”, still holds true as Twitter accurately reflects what is happening in the physical world into the virtual world of the Tweetosphere (Twitter).

We can determine that there are two types of engagement: one based on geographical proximity, and another is “emotional” or “topical” engagement. There is a clear influence of geographical proximity on hashtag frequency, in which a city’s propinquity to one another will influence user engagement surrounding a particular event. For example, Brussels and Paris have more in common than Mexico and Paris or Mexico and Brussels regarding the top 10 hashtags. In contrast, the pattern of Latin engagement vs. European engagement varies, as Mexico City demonstrate a different form of participatory culture due to the dominance from entertainment and media outlets. We found that the topics which generated user activity were related mainly to current events ( for example the Paris Attacks, Venezuela elections, Brussels lockdown). These topics were the most talked about, which led us to conclude that events influence Twitter users and divert their attention to discuss topics related to these events. With regards to geo locating opinions, we found ourselves faced with a data related limitation for Mexico City, which prevented us from expanding our analysis of geographic proximity in relation to opinions. Another limitation we faced was linked to the Topic Modeling tool which is, in essence, subjective and reflects what it is linguistically fed. This report took into consideration all the replicable data which could validate our research.

As social networks continue to infiltrate our daily lives, we as users will become increasingly more accessible, which can aid our future research. An in depth analysis focused on a larger, and more diverse sample of cities, with complete data sets, would allow researchers to study how Twitter reflects users’ engagement with respect to events taking place in their cities. This could eliminate further limitations and elucidate a cross metropolitan comparative study of Twitter user engagement.


About Twitter. 2016. Twitter. Web. URL: Date of access: 21.01.16

Agre, P. “Surveillance and capture: Two models of privacy.” The Information Society: An International Journal 10.2 (1994).

Owen, S., Imre, R. Little mermaids and Pro-sumers: The dilemma of authenticity and surveillance in hybrid public spaces. Callaghan: The University of Newcastle, 2013.

Mitchell, A., Hitlin, P. Twitter Reaction to Events Often at Odds with Overall Public Opinion. Web. URL: Date of access: 21.01.16

Topic Modeling Tool. 2012. Google Code. Web. URL: Date of access: 20.01.16.

Twitter Capture and Analysis Toolset (DMI-TCAT). 2014. Digital Methods Initiative. Web. URL: Date of access: 20.01.16.

Topic revision: r2 - 26 Feb 2016, JonasDeMeulenaere
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback