Most probably, you are not instantly aware of any emerging and unexpected social media crisis. In this world, we do know though that anyone can write anything about our company, our product or even – just mention us on online. So how to exploit real-time social media monitoring?
I met a product – called SentiOne – that helps brands track online discussions about their products and services and engage in conversations with customers across the entire Hungarian social and online web – or if needed, on an international level.
SentiOne is a language-sensitive tool with natural language processing algorithm for finding and analyzing all text contents published on the Internet.
I wanted to try it out and analyze a subject that has created huge online buzz in Hungary. Lots of articles and comments were posted for example about the migration and refugee issue last year. Let’s have a closer look, shall we?!
Basically, they monitor all online sources where contents related to the refugee topic have been mentioned, including Facebook, blogs, review sites, Instagram, YouTube, Twitter and all other online platforms (news portals, other portals). SentiOne works with and registers all online historical data from 3 years back, monitors every possible online source and continuously scans these sources in real-time for newer hits. It gathers all public data from the specified source (e.g: kuruc.info’s every related article, comment and their context). The contents published online are usually accessed and the info gathered through API. In the collected database, we find complex context, interactive charts and a visually enhanced, data-based user experience for richer results and deeper insights.
As part of a larger analysis, we revealed the hottest web-topic in Hungary over the last year with this online listening tool. We specified these topics, and then we launched new queries focusing on each hot topic to measure their effect and reach. In Hungary, the winner was the refugee issue. All findings – including the analysis based on the refugee problem, asylum seekers, etc. – were published in Hungarian.
The analysis of emotional attitude (sentiment analysis) is SentiOne’s own development, yes, but no specific algorithms have been made public. The academic research of John R. Crawford and Julie D. Henry’s is its basis. This was called Positive and Negative Affects Schedule (PANAS). The emotional analysis results in 70% accuracy – beyond this, all numbers and stats based on gathered data have 100% accuracy.
Researches carried out to map up emotional attitudes towards different topics are extremely subjective. If you for example place real people into a room, make them read an article and then ask them about their emotional attitude towards the topic, it would also result in just an approximate 80% of accuracy. Depending on what people perceive as positive and/or negative. For example, on this topic there would be no balance with supporting any political party, even though its algorithms imitate positive and negative attitude with regard to related mentions. At present, there are numerous ongoing researches on this field and the development of artificial intelligence is also of utmost importance: to get perfect results in sentiment analysis.
We also included posts and mentions that even though did not explicitly mention the keywords, but are part of the related online discussion and context. It is, of course, a decision and a question of the methodology underlining the topic set up. Naturally, you can also refine any SentiOne research and monitor only those posts that have the specific and previously set keywords and expressions in it. Contextual researches will always bring about more irrelevant results (eg. a smiley that someone posted as a comment to your examined topic), though it might provide a more thorough understanding of your online subject most of the time.
Queries related to quotations are not accurately interpreted but e.g.: if someone comments on Facebook it is a special reference to the system.
It is important for data protection to analyze public data – only and exclusively. From social media channels, data is gathered in accordance with the API relevant to the specific channel (posted in a public group, event or a fan page content). We do not see anybody’s writing he or she published on his/her private side, even if it’s public. Other website contents are public in a different way: eg. anyone can see registered users’ profiles when they check the articles, who have ever commented on the topic. SentiOne also sees those mentions that have been deleted.
If you want to know more about SentiOne please check it out yourself here: https://sentione.com/hu
Now, you know how I gathered all the refugee data. Let’s see my visual results -interactive version – Refugee problem in Hungary, 2016 .
I focused on the negative and positive sentiment occurrences with regard to the topic. The trend line shows outstanding periods and online peaks when people discussed about the refugee issue with an extremely high volume. Like on the 2nd October when Hungary’s refugee referendum was made public. It was initiated by the government, and it was commonly referred to as the quota referendum in the Hungarian media.
It is no surprise that the histograms of all media platforms show more negative comments than positive.
My favorite part of this dashboard is the ‘wheel chart’.
The annual ring shows the monthly breakdown of your choice (blogs, portals or Facebook users) and the emotional background of people’s reactions online about the asylum question in 2016. A point means a comment/post. One circle slice represents one month’s posts. If you look at the dots starting from the center and you go outwards, you can read data of an hour level breakdown of a given day.
If you have any more technical questions just write me.
You can read the Hungarian version of this blog post on INDEX site : Tudományosan is bizonyított: működik a kormány agymosása