ACADYS announced on March 3rd that, via its Beyond Data Sciences business unit dedicated to data intelligence, it will deliver a weekly analysis of the speeches, statements and writings of the main declared candidates until the second round of the presidential election. The firm will dissect the speeches and political programs of the main candidates in the French presidential election in order to identify common themes but also their specificities thanks to Artificial Intelligence.

Created in 1996, ACADYS’ mission is to help organizations create value through business optimization and innovation. The Information System, an essential factor of success for competitiveness, has quickly become ACADYS’ best known area of intervention. Present in France, Switzerland and Africa, ACADYS now has more than 25 employees and more than 300 client companies since its creation. The firm has just launched its Business Unit around Data Intelligence, Beyond Data Sciences. Through this unit, ACADYS implements diagnostic and consulting strategies in Data Intelligence but also studies and training.

Better understand the proposals and values of the main candidates with the help of AI

This week, the statements of 6 candidates have been analyzed by ACADYS: Anne Hidalgo, Eric Jadot, Marine Le Pen, Jean-Luc Mélenchon, Valérie Pécresse and Eric Zemmour, Emmanuel Macron and Fabien Roussel, will be included in the next analysis. In addition, the analysis will evolve over the weeks, focusing on or integrating other prisms, for example the evolution of the discourse between now and the inter-trial speech.

Data sources for the analysis

ACADYS only studied sources that came directly from the candidates: official programs, official websites, social media posts by the candidate himself, and political speeches at official campaign rallies. Interviews that may be biased by the types of questions asked by journalists or the information reported by the media were not considered in this analysis.

Methodology

ACADYS first identified all the official sources of the declared candidates, whether oral or written, and injected them into a data lake using big data techniques. Then, the speeches were “scripted” through NLP (Natural Language Processing) methods in order to be compared to the written texts. This includes speech recognition techniques to transcribe the recorded speech into usable text. Machine learning algorithms were trained to extract the most frequent terms and themes (simple 1-gram and compound 2-gram) from a base of 10000 words. Pre-processing was performed to “clean” the corpora (pre-processing). A vector analysis using the Euclidean distance allows to better understand the level of “proximity” between the different candidates.

The first results of the survey

Most used terms: France and Europe.

Individual analyses

• The terms France and Europe are by far the most cited terms and appear systematically in the Top 5 for each of the candidates, except for Marine Le Pen, where Europe only comes in 9th position.
  systematically in the Top 5 for each of the candidates except for Marine Le Pen, where Europe is only in 9th position.
• The French term appears in the top 10 for each candidate except for Y. Jadot where it appears beyond the 20th position.
• The terms state, country and politics are cited by all the candidates among their 30 most used words.
• The term purchasing power is frequently cited by Valérie Pécresse, but also by Eric Zemmour and Marine Le Pen…

Cumulative frequency analysis:

In cumulative frequency, the words Europe, France and French are by far the most used. The other most cited terms are: country, ecology, law, state, politics, public, then come the terms social, law, nation, work, president E. Macron, rights, health, republic, justice, business, economy, climate.

Level of proximity between the candidates

A similarity analysis was performed to identify the level of proximity between the candidates. Using the Euclidean distance, and based on the last output of the algorithm, we note that:

• The candidate closest to Jean-Luc Mélenchon is Anne Hidalgo
• The candidate closest to Marine Le Pen is Eric Zemmour, but he himself is, through his statements, closer to the candidate Valérie Pécresse
• Yannick Jadot is the furthest from the other candidates.

The three terms out of 10,000 most cited by the candidates.

VALERIE PECRESSE: France, French, Country.
ERIC ZEMMOUR: France, French, Country.
YANNICK JADOT : Europe, Ecology, Law.
ANNE HIDALGO : Europe, France, Public.
J-LUC MELENCHON: France, Europe, Measure.
MARINE LE PEN : France, French, Law.

Translated from Election Présidentielle : L’Intelligence Artificielle va-t-elle influencer les votes ?