Political Barometer is a software estimating political public opinion, by using political tweets, which are collected and analysed daily. This analysis is based on Artificial Intelligence techniques (text sentiment analysis). Tweets collected refer to political parties that are currently forming Greek Parliament, in order to ensure data sufficiency (alphabetically ELLINIKI LISI, KINAL, KKE, MERA25, ND, SYRIZA). This software could substitute polls, offering the benefits below:
The Political Barometer research and Development work has been carried out by Computational Politics Special Interest Group (SIG) of the Computer Vision Machine Learning (AIIA.CVML) R&D group of the Artificial Intelligence and Information Analysis (AIIA) Laboratory of the Department of Informatics, Aristotle University of Thessaloniki.
'Political Barometer' software is still under evaluation.
- Daily political public opinion analysis with immediate response to political stimulus.
- Low cost.
- Poll results since the beginning of our analysis do not differ significantly from our results.
The Political Barometer research and Development work has been carried out by Computational Politics Special Interest Group (SIG) of the Computer Vision Machine Learning (AIIA.CVML) R&D group of the Artificial Intelligence and Information Analysis (AIIA) Laboratory of the Department of Informatics, Aristotle University of Thessaloniki.
'Political Barometer' software is still under evaluation.
The voters who have not decided their votes yet are distributed proportionally to all parties.
en.png)
The current estimation on voting intention, based on the past weeks' political tweets.
The results of voting intention estimation per day for every political party is shown below. Our estimation uses a combination of PREVIOUS poll data and political tweet sentiment analysis. Polls are also provided with the symbols listed. Non-continuous line represents the period that the model DOES NOT takes new polls under consideration.
By using the arrows you can change viewing time period: a) whole analysis period (since 5/2022), b) past week, c) past month, d) past three months.
Past week
Past month
The graphs below show the estimation of voting intention merely from twitter data. Essentially, the negative tweets of every party are distributed to the rest of the parties according to their popularity (as defined by the share of their positive and neutral tweets).
This estimation presents significant sensitivity and ongoing political events have direct impact on it. It is emphasised here, that the individuals using twitter does not necessarily represent the general voting intention.
By using the arrows you can change viewing time period: a) whole analysis period (since 5/2022), b) past week, c) past month, d) past three months.
Past week
Past month
The graphs below show the number of positive-negative-neutral tweets per day for the past month (30 days). They are generated by automated political tweet analysis. The sentiment classification accuracy is approximately 82% after being compared to ground truth.
You can view the corresponding graph, for every party, by using the arrows.
26/2/2023: Greek MP Pavlos Polakis case (discussion for possible deletion)
1/3/2023: Vale of Tempe railway tragedy
28/3/2023: Announcement of the general election date (21/5/2023)
10/5/2023: Political leaders' debate.
1/3/2023: Vale of Tempe railway tragedy
28/3/2023: Announcement of the general election date (21/5/2023)
10/5/2023: Political leaders' debate.
Political Barometer mean deviation from the general elections at 25/6/2023
Deviation for ND: -0,95%
Deviation for SYRIZA: 2,16%
Deviation for KINAL: -0,55%
Deviation for KKE: -0,19%
Deviation for ELLINIKI LISI: -0,24%
Deviation for MERA25: 0,41%
Deviation for OTHER: 3,14%
Mean deviation for all political parties: 1.1%
Mean Political Barometer deviation compared to opinion polls during 5/2022-5/2023
Mean deviation for ND: 1,46%
Mean deviation for SYRIZA: 1,24%
Mean deviation for KINAL: 0,63%
Mean deviation for KKE: 0,35%
Mean deviation for ELLINIKI LISI: 0,38%
Mean deviation for MERA25: 0,26%
Mean deviation for OTHER: 1,09%
Total mean deviation for all political parties: 0.95%
Mean deviation is small for smaller political parties and bigger for bigger ones, but very low on general.
By using more data the mean deviation was reduced significantly.
Deviation for ND: -0,95%
Deviation for SYRIZA: 2,16%
Deviation for KINAL: -0,55%
Deviation for KKE: -0,19%
Deviation for ELLINIKI LISI: -0,24%
Deviation for MERA25: 0,41%
Deviation for OTHER: 3,14%
Mean deviation for all political parties: 1.1%
Mean Political Barometer deviation compared to opinion polls during 5/2022-5/2023
Mean deviation for ND: 1,46%
Mean deviation for SYRIZA: 1,24%
Mean deviation for KINAL: 0,63%
Mean deviation for KKE: 0,35%
Mean deviation for ELLINIKI LISI: 0,38%
Mean deviation for MERA25: 0,26%
Mean deviation for OTHER: 1,09%
Total mean deviation for all political parties: 0.95%
Mean deviation is small for smaller political parties and bigger for bigger ones, but very low on general.
By using more data the mean deviation was reduced significantly.
I. Pitas – pitas@csd.auth.gr