Dit is een webapplicatie die de evolutie van populaire tags kan voorspellen die op Twitter en Instagram zullen passeren. Zo kan men weten welke tags men dient te gebruiken om extra populariteit te kunnen genieten. Dergelijk onderzoek kan ingesteld worden voor een bepaalde periode en een bepaalde regio. Op die manier kan men kijken of er overlappingen van tags zijn op deze twee populaire platformen tijdens de ingestelde periode. Met deze applicatie kan men ook de mate van externe invloeden onderzoeken.
Dit is een Engelstalig project dat bestaat een een onderzoek, webapplicatie en evaluatie. Onderstaand vindt u enkele belangrijk stukken van het rapport. Het volledige document is beschikbaar via de document knop.
Nowadays it seems unthinkable to live in a world without constantly being one click away from information, interesting people and all kinds of popular media. There is so much content on the internet that it can be overwhelming at times and one could start to wonder if there is any method to the madness that is social media. Many social media users are actually interested to know what exactly they can do to increase the popularity and visibility of their posted content on those media platforms.
In that context of predicting popularity on social media it is plausible to believe that, given two fairly similar social media platforms, the popular tags on one social media platform are overall different from the popular tags on the other social media platform. In other words: the set of tags that could be considered popular on both platforms is relatively small. Furthermore there is also reason to believe that that set is strongly influenced by important or major seasonal events. It makes sense to expect that for a short period of time event-bound tags will rise to popularity on both platforms and therefore temporarily increase the general overlap of popular tags on both platforms.
This research project seeks to support the above beliefs by studying the evolution of popular tags on Twitter and Instagram for a period of fourteen days during which two major seasonal events occur, namely Christmas and new year. Using the data gathered during that period the project seeks to answer the following two research questions: (1) How large is the overlap of tags that are considered popular on both platforms? (2) How significant is the influence of the two major events on that overlap?
The analysis provides enough information to give a proper answer to both of the research questions. Regarding the question whether the two platforms have a large overlap of popular tags, the prior beliefs were confirmed. The overlap is rather small and limited to only 2 to 4 tags every day even when taken into account the ability to use spaces in tags on Twitter and lack of that feature in Instagram. The question if seasonal events have a significant impact on the popular tags, has also been answered. It is clearly noticeable on Twitter that during the Christmas and new year period, tags regarding those events were strongly present in the top 100. It was less observable on Instagram. That can be explained by the fact that Twitter’s trending tags were used. Twitter uses a special ranking algorithm that filters out tags that have been popular for a longer period of time. That way they encourage new tags to become trending. The old, but actually quite popular tags such as #JustinBieber won’t occur in the trending tags but new tags such as #christmas will appear as soon as they know a sudden rise and become popular enough. On Instagram there’s no such thing as trending tags so simply tags that were recently used in posts by users in the UK were used. Therefore no filtering was done on the Instagram data and also tags which have been popular for quite a while were retrieved. The main reason, however, is that the algorithm which calculates the Instagram tag weights is based on the worldwide occurrence of a tag. That makes it even more difficult for new, temporal tags such as #christmas to be in the top 100. It is also worth mentioning that a lot people on Instagram already use specific tags to increase the popularity of their posts. These tags range from #love to #followme and are clearly dominant every day in the top 100. That kind of behaviour is a lot less recurrent on Twitter.
Some improvements could be made to the current project which haven’t been done due to time constraints. A first optimization would be to implement a better weighting algorithm for Instagram. It can be more accurate by taking the population of a city into account.
Secondly, until now the Java code has always been run manually, three times a day at more or less random times. It would be more convenient if we could automate the data runs. More convenient because you don’t have to do it yourself and better because the time would be fixed, which provides more accurate data. A third improvement which could be made, is to remove all whitespaces in tags before they’re written to the database. At the moment, the database contains tags such as ‘New year’ and ‘Newyear’, which should be considered the same. The frontend tries to work around this little flaw but it is not possible to do so everywhere. Doing this in the backend is quite easy, would save a lot of time in the frontend and, top of that, it would improve the results. Another and perhaps also least important improvement, would be to write more information to the database. The backend only writes away a bare minimum of information such as the tag itself and its rank. The rank was calculated based on a lot more properties and it’s possible to simply write all these properties to the database. The frontend can then retrieve all these properties and do more calculations as it pleases. For example, we haven’t written away the location of where a certain tag was retrieved and this makes it impossible for the frontend to do a geographical comparison. A last improvement would be letting the resulting values, from the tag weighing formulas, match more. This is a very complex problem that may not even be possible to solve. If the resulting values could be even worthy, it would be much easier to compare tags from different platforms.