Our systems have detected that you are using a computer with an IP address located in the USA.
If you are currently not located in the USA, please click “Continue” in order to access our Website.
Local restrictions - provision of cross-border services
Swissquote Bank Ltd (“Swissquote”) is a bank licensed in Switzerland under the supervision of the Swiss Financial Market Supervisory Authority (FINMA). Swissquote is not authorized as a bank or broker by any US authority (such as the CFTC or SEC) neither is it authorized to disseminate offering and solicitation materials for offshore sales of securities and investment services, to make financial promotion or conduct investment or banking activity in the USA whatsoever.
This website may however contain information about services and products that may be considered by US authorities as an invitation or inducement to engage in investment activity having an effect in the USA.
By clicking “Continue”, you confirm that you have read and understood this legal information and that you access the website on your own initiative and without any solicitation from Swissquote.
The Lausanne-based lab generated a buzz at COP21 in Paris at the end of 2015. Tools developed by the lab made it possible to analyse and dissect opinions on climate change all over the world in real time. The goal was to provide the media, experts, NGOs and companies with a revolutionary way to measure public opinion on social media. For the 2017 French presidential elections, the Social Media Lab will provide its expertise to Swissquote, regularly delivering “ snapshots ” of French public opinion on various candidates. Director Jean-Luc Jaquier takes us behind the scenes of the lab.
Can you explain what your tool does and how it works?
We capture, analyse and publish, in real time, evolutions in public opinion on a given subject through what is expressed on the internet. We collect a huge amount of data on social networks to obtain an overview of opinions. To do so, we use what we call Natural Language Processing (NLP), which is the understanding of human language, as well as graph analysis, which determines the positions of various people, the connections between them, their respective influence, etc. These two technologies use artificial intelligence (AI), and by bringing them together we can gain a very precise understanding of public opinion shared on social media.
What are the main challenges when it comes to artificial intelligence?
The challenge is reaching a high level of clarity in understanding a text. The branch of NLP that does this is called Natural Language Understanding (NLU), and it is a vast research topic in AI. There is so much room for improvement. One of the main challenges is understanding posts that are ambiguous, ironic or sarcastic, which are very popular on social networks. A comment may seem positive taken at face value, but in reality it means exactly the opposite. We also have to identify the fake news that contaminates social media.
How does graph analysis fit into the process?
Not only do we have to decipher the text, but we also have to understand the context. Is this a credible source? How was the content shared and in which communities and places? What is the impact? Our algorithm builds a social-media graph based on the answers to these questions.
Let’s look at an example. The tool determines that person A posted in support of a candidate in the presidential election. If person B shares that post in a positive way, then AI can deduce that person B is likely to also support this candidate. Conversely, if someone posts a positive comment about a candidate but uses a sarcastic tone, the algorithm will understand that this person probably supports a different candidate.
By combining NLP and graph analysis in our algorithms, our tool is three times more precise than if we just used NLP. The results it produces are already very precise – and a quantum leap ahead of any other tool to date.
Where does human intervention come in?
to connect humans and machines. The machine will analyse the spread of opinion using the graph, and then we zoom in on parts of the graph and score the analysis using human knowledge. If the machine made a comprehension error, we will point it out. It will then recalculate and learn from its mistakes. This is called machine learning and deep learning.
The machine will then adjust the entire graph – which could include millions of opinions – based on our input. The machine is able to adapt and learn.
Just how much can artificial intelligence improve?
There is enormous room for improvement. In the next five to 10 years, we will make huge progress in understanding information expressed by humans. Tools are getting better every day and can analyse millions of data almost instantaneously, in any language all over the world.
Now that we can capture public opinion almost immediately, will traditional polling methods become obsolete?
We can now capture a change of opinion in just a few hours. A traditional poll takes much longer, around two or three days. But polls are the result of in-depth interviews and based on a representative panel. So I’d say that these two methods complement each other. On the internet and social media, we’re in rather uncharted territory. Some demographics, mainly young people, express themselves far more on social media. The advantage of our tool is that we can measure opinion in real time, directly from the source with no intermediary. In that sense, the French elections will be a fascinating case to observe.