#drmateur

Topic and research question

It is common sense that smoking is a serious public health problem worldwide. It is considered to be the largest single cause of preventable death and disease in the world (WHO Report on the Global Tobacco Epidemic, 2008). The more research is conducted, the more is known about the risks of smoking, hence we are certain in these days that smoking causes serious harms to people’s health and can even lead to death. Research has been conducted as long as scientist and public health experts have known about the risks of smoking. Moreover it seems that the interest in researching smoking risks (Catsburg et al., 2014; Pimhanam et al., 2013) and/or people’s smoking behavior (Noorzurani et al., 2013) is important for today’s society. It comes as no surprise that scientists and researchers started engaging themselves with the questions as to why people start smoking in the first place. Thus many studies are concerned with the perception of smoking and tobacco in society and how the image of smoking is being shaped and influenced.

In this context the role of the media has been questioned many times. Especially when it comes to the problem of smoking among teenagers the film industry and mass media in general are under high scrutiny. Since the 1990s research provides a better understanding of the proliferation of the image of smoking in popular media. Young people indeed associated certain meanings and values to the representation of smoking in media (McCool et. al, 2005). Especially with the rise of the movie star in the 1930s and 1940s, showing images of leading male as well as female actors smoking in films, the image of tobacco has been glamorized (McCool et. al, 2003).  With a growing number of studies examining the image of tobacco use in film and the media, more evidence has been found that these images endorse the motivation, especially among younger people, to start smoking. The media not only plays a significant role in influencing the behavior of teenagers and generally speaking the youth culture, different kinds of mass media (film, magazine, tv) are attributing favorable characteristics to the usage of tobacco (Watson et. al, 2003). Thus there is a growing concern of the media influencing young people’s attitude toward smoking.

With the rise of social media this concern becomes even more crucial, as many people depict themselves or others on pictures while smoking and disseminating them via the various social networks. It is argued whether those pictures have potential to influence others who see these pictures. A recent study conducted earlier this year investigates how friendship with peers who smoke and showing their substance use online could influence the teens’ own attitude towards tobacco (bitscience.com, 2014).

           However research on the representation of tobacco use and smoking in social media networks has been few and clearly shows a research gap within this field of study. Therefore this study investigates the social media platform Twitter and how smoking and the sentiment towards smoking can be identified here. The research project hereby follows the overall research question: What is the image of smoking on Twitter? We developed subquestions for the logical analysis:

What devices are used to write tweets contained #smoking?

What hashtags are connected with #smoking  on Twitter?

What is the sentiment of the tweets and retweets with the hashtag #smoking and hashtags related to it?

Methodology

In order to answer our research question, certain DRM methods were applied. First, to show the real amount of tweets connected with #smoking several online tools were used. Those allowed this research to explore findings in real time and get a more complete picture of the image of smoking. Towards the end, Gephi was used for analyzing the connections between hashtags and tweets.

Data

To have a better understanding of how smoking is being perceived on Twitter, two different datasets were used. Accordingly, the first dataset is analysed through the following tools:

For the online real time analysis made with Tweet Binder, we used 1981 tweets from October 2014, retrieved on 28 October.

For the online real time analysis made with Sentiment Viz, 419 tweets from October 2014 were retrieved on 28 October 2014.

The second dataset is analysed from the perspective of Gephi.

Hashtag_edges file, which consisted of 3676 edges, was used.

The file hashtag_edges as well as the file retweet_edges (242 edges) and smokingEnLIWC file were used for the LIWC analysis.  

Methods and results

Tweet Binder

The report for #smoking shows significant results. Overall, 1,981 tweets that contain this hashtag have been retrieved. It is interesting to notice that 6,43% of them are original tweets (120), while 35,99% are retweets (717). Thus the number of retweets is larger and there is a certain interaction between users, since they are willing to retweet the others’ tweets. In addition, the analysis shows that 32 of all tweets were replies. Most tweets contains additional links and pics (55,97% of all tweets, 1115 tweets). As these additional pictures and links are another type of content than text, it testifies the interest to the topic among users.

Most of the contributors (1006) tweeted once. The number of contributors who tweeted three times was smaller than the number that tweeted six times or more. It can be assumed that there may be a person or organization interested in #smoking, who posted more than 6 tweets as it can be considered as a daily (or with a certain purpose) rather than random contribution.

Among most active contributors, there are some twitter accounts connected with porn and videos for adults, for instance, @C4SUpdates, which is “#1 Downloadable Clip Site on the Net” (https://twitter.com/@C4SUpdates, 2014). This user is also among those who has the greatest impact.

As for the top 10 sources that were used to write and retweet tweets contained #smoking, the most popular one was Twitter for Web Client. The second one was Twitter for iPhone, the third one - IFTTT. The difference between the number of tweets from each client was not significant, though: 287 for Twitter for Web Client, 235 for Twitter for iPhone and 221 for IFTTT. Interestingly we noticed Instagram among the clients (5th place with 168 tweets).

Sentiment Viz

As for the hashtags connected with #smoking, those connected to porn (#porn, #latex, #xxx) are marked as positive, while #tobacco, #cancer, #nicotine are marked as negative. Interestingly, #quit was among the positive ones. This can be interpreted as the positive emotions people feel while quitting smoking, which they share with their followers.

Hashtags connected with porn (#fetish) and pot (#420) are also connected with #smoking.

As for the sentiment of the tweets, they are mostly positive. It could be assumed that it is either because of the positive emotions people feel about smoking or because the tweets are connected with porn (and users are positive about it). 

Gephi

Gephi was used for further examination. When examining the hashtag_edges data file in Gephi and conducting indegree and Force Atlas upon the twitter data, the following Gephi visualisation appeared.

The biggest cluster (the second one) in the Gephi visualisation represents #smoking. It can be observed that the smaller clusters are connected to the #smoking. Moreover, a lot of terms are related to this tweet. Some of the most popular ones are #nicotine, #longcancer, #LetsSmokeLoud and #ShakeTheHabit. However, most interesting of all, it is the cluster of tweets forming at the right of the cluster of #smoking. These are tweets that are related to #Smoking (with capital letter S). The findings of these tweet related messages are interesting since half of the messages are also related to #Nicotine, #Addiction and #KidsHealth, #Electroniccigs topics, which are smoking-related topics. However, the other cluster that is related to #Smoking represents porn related messages.

Examples of related tweets are #porn, #Sextweets, #Nudescene and #Slut. Therefore, an interesting perception and maybe a noticeable trend is that #Smoking is not only related to subjects concerned with smoking, but also to Porn and Sex.

We sorted the data in Excel and selected all tweets with the hashtags ‘sex’,’porn’ etc. and then marked them as “porn”. As it was done manually, it could be considered as a difficulty. Manual sorting of hashtags can lead to the error in identification, for example,#sex can appear in the non spam porn related tweet, although the probability of this error is quite small and it could be neglected; nevertheless, this can be considered a limitation. While red lines are for tweets connected with porn, green lines are for tweets which do not contain any kind of porn. It is interesting to notice, that majority of the hashtags related to the pornographic content are linked together, which means that they are used together frequently. Also #smoking with all letters in lowercase has less connection to such hashtags, meaning that porn related tweets mostly use all caps or first letter caps hashtags. It can be the case that some of the hashtag which were tagged as porn hashtags were used in regular tweets, but chances of that are rather small, and we see that such hashtags are rarely connected to the normal ones. When we looked at the tweets manually, we noticed that there could be some hashtags which were not necessarily related to porn, like #naked.

LIWC data (smokingEnLIWC) was added to hastag_edges file. Blue is for none, green is for positive, and red is for negative. It is clear, that there were a lot of tweets marked as ‘none’.

It is interesting that tweets contained hashtags connected with porn are marked either ‘none’ or ‘positive’ in terms of emotions.

Hashtag ‘#420’ related to pot is also noticeable and is marked as ‘positive’.

As for analysing the retweets, Fruchterman-Reingold algorithm was applied to retweets_edges file. Red colour is for tweets connected with porn while green is for tweets about everything else. There is only one big cluster of retweets connected with porn.

The emotions for retweets were analyzed using the same algorithm: red for negative emotions, green for positive emotions and blue for neutral or none. As it can be seen, the blue retweets dominate.

Conclusion

In order to answer the main question of this research (What is the image of smoking on Twitter?), these findings show that the used database has two main points of focus when using a hashtag related to smoking: (a) health concerns and smoking in general and (b) pornographic content. Moreover, an extended range of smoking-related hashtags were used. As previously seen visualizations show, it can be assumed that #smoking is primarily related to actual smoking and health issues concerning smoking. It is interesting to note that #Smoking is not only related to actual smoking but also to porn and sexual themes. Another interesting finding concerns #SMOKING, which proves to be mostly spam or related to pornographic content. Furthermore, this research shows that when users use a smoking-related hashtag, they do not necessarily refer to cigarettes but also to cannabis and pot consumption. Tweets connected with cannabis and pot are positive, tweets connected with porn are marked either as positive, or as none in terms of emotions. Most of retweets and tweets were marked as ‘none’ or ‘positive’ in terms of emotions. These findings can be useful for future research about the image of smoking on social networks.

P.s. Visualizations can be found here.

References

Catsburg C., Kirsh V.A., Soskolne C.L., Kreiger N., & Rohan T.E. (2014). Active cigarette smoking and the risk of breast cancer: a cohort study. Cancer Epidemiol, 38(4), 376-381.  doi: 10.1016/j.canep.2014.05.007.

McCool, J. P., Cameron, L.D. & Petrie, K. J. (2005). The influence of smoking imagery on the smoking intentions of young people: Testing a media interpretation model. Journal of Adolescence Health. 36, 475-485.

McCool, J. P., Cameron, L.D. & Petrie, K. J. (2003). Interpretations of smoking in film by older teenagers. Social Science and Medicine. 56. 1023-1032.

Noorzurani, R., Bond, A., Wolff, K. (2013). A comparison of smoking behaviour characteristics between Caucasian smokers in United Kingdom and Malay smokers in Malaysia, Preventive Medicine 2013, doi: 10.1016/j.ypmed.2013.04.010.

No author. (2014) How Party Pics on Social Media Stimulate Teens Smokinng and Drinking Behaviour, Retrieved October 24th 2014 from: http://www.bitescience.com/Knowledgedatabase/Article/tabid/81/ArticleID/251/ArticleName/HowPartyPicsonSocialMediaStimulateTeensSmokingandDrinkingBehavior/Default.aspx

Pimhanam, C., Sangrajrang, S., Ekpanyaskul, C. (2014). Tobacco Smoke Exposure and Breast Cancer Risk in Thai Urban Females. Asian Pacific Journal of Cancer Prevention, 15 (17), 7407-7411.

Sentiment Viz. Retrieved from: http://www.csc.ncsu.edu/faculty/healey/tweet_viz/tweet_app/

Tweet Binder. Retrieved from: http://www.tweetbinder.com/

Watson, N.A., Clarkson, J.P., Donovan, R.J. & Glies-Corti, B.(2003). Filthy or fashionable? Young people’s perception of smoking in the media. Health Education Research. Vol. 18, no. 5, 554-567.

Now that the online participation assignments are over it's time for a little breakdown of our subreddit. 

We have 31 subscriptions, making our subreddit rank 68,185 out of 509,573 according to: http://redditmetrics.com/r/subreddit_stats

Our traffic stats show the big impact of the online participation due date with weekly spikes, and declining total page views as the term went on.

The day of the week and hour of the day for posting to our subreddit (from: http://www.redditlater.com/analysis/) seem to be about what I'd expected:

I scraped the data from our reddit with import.io. In total we have 567 posts with 1.185 comments. The following is just based on posts, not comments.

Post with the most up votes: "Visualizations of my FB network & an easy guide how to create your own" by _lizlemon_ Tied for second place: "Data visualization new style: The age of buildings" by NienkeJ, "New interesting Ted Talk on Big Data <3" by josinebakkes and "Engaging my Facebook Friends to add me on ELLO" by npenchev

Users whose posts were most up-voted (in total):

412753ibeur

Vally_W

_lizlemon_

Users whose posts were most up-voted on average:

MonikaHlub

fanchuly1

iana_p

Most comments: "Is Facebook a good venue for political discussions?" by Esther1604 is tied with "Free Brownie Points or 'a Reddit questionnaire'" by NienkeJ

Users with the posts that had the most comments (total):

Vally_W

_lizlemon_

katyanagibina

Users with posts that had the most comments on average:

deankoend

Esther1604

gabrielagarcia

Here is a wordle made from your post titles.

One interesting question about the nature of news is how well it reflects the pattern of real events around the world. It’s natural to assume that people living in a certain part of the world are more likely to read, see and hear about news from their own region. But what of the international news they get—how does that compare to the international news that people in other parts of the world receive?

original research: http://arxiv.org/abs/1410.3710

If every online click is being registered in a global dataset also known as Big Data, our day to day habits can also be tracked. With the advent of digital technologies, human themselves become data; or at least, through the emergence of the Quantified Self movement and its tracking tools our behaviours, practices and daily routines can be projected into raw datasets, and then, closely scrutinised. This following experiment firstly focused on acquiring data from a mobile application called Sleep Time and developed by Azumio; secondly, online streaming platform Last.fm will also be referred to as device that tracks down and records the user’s music streaming activity. 

1. Sleep Time

Three days are going to be utilised in order to analyse my own sleep cycle: October 21st, 22nd and 23rd. Through the following graphs, the average bed time and the average duration will be analysed by comparing the aforementioned selected days. 

Afterwards, each day will be analysed individually. On October 21st, the set bed time was 23:47 and the waking time 07:23, so the entire sleeping activity lasted for 7h and 36 min. Out of this, I was awake for 44 minutes; Light Sleep counted for 3h and 16min and Deep Sleep for 3h and 36min drawing to a conclusive efficiency of 6h and 52 minutes. 

On October 22nd, the initial time of going to bed was 23:07 and the rising up time 8:56 so the entire activity tracked by the app lasted for about 9h and 49min. Out of this period of time, I was awake for 0:21 minutes; Light Sleep counted for 4h and 54min and Deep Sleep for 4h and 43min drawing to an eventual efficiency of 9h and 28 minutes of actual sleeping. 

On October 23rd, the set bed time was 22:53 and the waking time 8:39, so the entire sleeping activity tracked by the app lasted for about 9h and 46min. Out of this tracked period of time, I was awake for 20 minutes; Light Sleep counted for 5h and 1min and Deep Sleep for 4h and 24min which draw to a conclusive efficiency of 9h and 25 minutes.

2. Last.fm

Through Last.fm I tracked my most played songs that have been played on every personal device where the Last.fm application is installed on (my laptop and my phone). Under the conditions that I haven’t used the service that often lately, I will analyse a broader period of time: the last 12 months of activity. 

As the above graphs show, my most played bands are The Drums, The Hives and The Horrors. Both graphs contain an interesting trend regarding the band called The Drums. Under these circumstances, I will focus the rest of this analysis on this particular band.

Further on, if I click on the 340 plays The Drums has registered in the past year, I can see their most popular albums and most popular songs with their respective amount of plays for each track/record.

By clicking on the most popular song (What We Had with 37 plays), I can see the exact time and date when the song was played in the past year. As the application requires continuous Internet connection, the cluster of songs generated on the 4th of February is caused due to the fact that I listened to the song when I did not have an active Internet connection; so, when I went online for the first time, all the plays registered with the same date and time (4 February 2014, 15:58). 

It is interesting to look into the overall listening trends of the band itself. With over 17million plays and nearly 800k listeners, I contributed to The Drums’ official library with a total of 683 plays. Here, this projection tracked the trend of the band by taking into account the number of plays registered each month, starting with May 2014 and finishing with October 2014 (the low-high interval is roughly situated at 9k - 15k). 

As a platform that gives users the possibility to socialise and interact with other users, it is an engaging exercise to look into my relationship with other ‘user friends’ by taking into account The Drums as the connection point between users. 

Measuring my sleep cycle - now that was interesting. Although I have friends who do that on a regular basis, and I have to admit, I have thought about it myself, I was never too much into the quantified-self, tracking-my-every-move craze. Let alone post about it.

But here I am. Things change, I guess.

I decided to use the free app Sleep Time by Azumio for Iphone. I chose this app, because it seemed like I could get the raw data out of the app and send it to myself via email. However, after receiving a zip file full of data, I had no clue what to do with it, let alone know how to clean the data and use it in Tableau.

The data itself is quite detailed, but does not tell me much about the quality of my sleep itself. Here is a screenshot of the zip file containing the data:

I then opened the highlighted data set in Excel (I tried them all), which is displayed below:

As you can see, this data gives me information on what the app itself was doing while I was sleeping. Which is seen in ‘alarm is triggered..’ and ‘startClockupdates’ for example. While this information is pretty interesting, I am not able to use it in Tableau.

The app itself produced some visualizations as well, so I decided to just share those. The first one is the classic visualization of a sleep cycle including both light sleep, deep sleep and times I was awake. And yes.. I went to bed pretty late both times using this app. This week has been stressful!

I really liked this second visualization provided by the app as well, as it looks like something I would have been able to do in Tableau as well:

I do not know if I would use the app a lot after this experiment. I do quite like using self-tracking data like this and don’t really see a problem with it or find it creepy like I have read people say on Reddit. I really liked seeing the visualizations of my sleep cycle and thus being able to see what the quality of my sleep was like. But I do not know how accurate these findings would be coming from a phone lying close to me in my bed. I am sure this has been tested, but I still question the outcome. It will for sure give you a good indication of what your sleep cycle is like though.

A thing I didn’t like was the so called ‘wake-up phase’. This allows you to set a time frame in which you want to be waked up. This means that if you pick a time frame of 20 minutes and you set your alarm at 08:00, you can be woken up ranging from 7:40 to 8:20. In this way the app can wake you during your light sleep, which would ensure you to be less tired when waking up. When I tested it I set my alarm for 06:45 and got waken up at 06:32. Now I have to admit I did notice that I woke up fairly easily, but those fifteen minutes are still too precious for me to let go off. Not a morning person! So if I were going to use the app again and I would have to wake up early, I would get rid of the ‘wake-up phase’.

StockTwits uses a followers model, connecting investors who are interested in the same financial opportunities. Though the insights can be very timely and represent the sentiment of an informed group, the ‘group’ is just defined by who decides to follow who. There is no collaboration, because because no one is sharing their personal data.

The ‘Quantified-Self’ is a thrilling prospect for some: Massive datasets about oneself can be a new route to self-discovery. But for most of us, the idea of continuous self-tracking is a novelty that results in shallow insights. For the Quantified-Self movement to become truly useful, our gadgets will have to move beyond the novelty of gratuitous behavioural data, which we might call a ‘first degree of meaning.’ They’ll have to address a second degree of meaning, where self-tracking helps motivate people toward self-improvement, and a third degree of meaning, where people can use data to make better choices in the moments when a decision is actually being made. We’re moving closer to those goals, but we’re still not thinking rigorously about the challenges involved.

It so happens that the rise of the quantified self coincides with the rise of Big Data, which has become a buzzword rapidly adopted in targeted marketing campaigns and recommendation engines that push products. But in between Big Data and Small Data, between the Quantified-Self and the crowd lies a third way: the Quantified-Us. Imagine a future where self-tracking harnesses a whole population’s data to identify patterns and make meaningful recommendations. Imagine a future where we can see into the data of people just like us, to help us live better, and where we willingly give up a bit of privacy in exchange for vast benefits.

PatientsLikeMe allows people to share personal health records so they can compare ‘treatments, symptoms, and experiences.’ The site also supports personal connections with the community, as well as the ability to track your own health data and to make your records available to medical researchers. These data, however, are positioned as a tool for the medical community to review and gain clinical insights.

Crohnology is a social network centered on people who suffer from Crohn’s disease and colitis. The community revolves around the sharing and aggregation of information. But the scope and depth of data that the patient can access is limited, and, as a result, so are the insights.

StockTwits uses a followers model, connecting investors who are interested in the same financial opportunities. Though the insights can be very timely and represent the sentiment of an informed group, the ‘group’ is just defined by who decides to follow who. There is no collaboration, because because no one is sharing their personal data.

Imagine a person with epilepsy trying to understand an uptick in seizures. What if he could compare his triggers to those of people just like him? Such a user experience could address everything from Crohn’s disease to migraines. These need not be separate products: Indeed, they could be similar user experiences, tailored to individual use cases.

Now imagine a person with insulin-dependent diabetes whose blood sugars are running high at night, but who isn’t able or doesn’t feel motivated to understand why. What if she could see the profiles and data of other people like her, and see where she falls relative to the “norm”? What if she was able to start a dialog with other people like her, or to get emotional support when she needs it?

It’s easy to imagine a variety of scenarios in which self-tracking combined with collective data sharing can result in deeper understanding and heightened motivation. Ultimately the Quantified Us can help people take better care of themselves, more often—and feel more connected to each other in the process.