Skip to content

German Twitter Stance Detection With BERT on the Russo-Ukrainian Conflict

License

Notifications You must be signed in to change notification settings

joh-ga/german-tweetstance-russiaukrainewar

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

German Twitter Stance Detection With BERT on the Russo-Ukrainian Conflict

To gain insights into the stances of Twitter users in Germany towards the Russo-Ukrainian conflict, a German BERT model was fine-tuned for cross-target stance detection and applied on 2022 Twitter data for prediction and analysis.

General Information

Fine-tuned model
The model is available on joh-ga/german-tweetstance-bert-uncased-russiaukrainewar. It achieves an F-macro score of 0.9024 on test-1 and 0.7450 on test-2.

Targets
The model was fine-tuned on four targets:

  • Arms Delivery (AD): Delivery of (heavy) arms to Ukraine
  • NPPs Operation Continuation (NOC): Repeal of the planned nuclear phase-out by the end of 2022 and the continuation of NPP operations in Germany
  • Speed Limit Implementation (SLI): Implementation of a temporary speed limit on German highways
  • Ukraine Support (US): General support of Ukraine in the ongoing Russian-Ukrainian conflict 2022

Data

  • Fine-tuning: Tweets were auto-labeled using stance-exposing hashtags.
  • Evaluation: test1 (auto-labeled), test2 (manual-labeled).
  • Analysis: Unlabeled 2022 Twitter data (02/24/2022 until 12/31/2022).

The complete dataset is made available for the research community. Due to privacy restrictions of the Twitter API, only the text IDs, the target as keywords as well as in the form of a German question, and stance labels (against, favor) are provided. The complete data can be retrieved via the Twitter API with appropriate access. Note: For underrepresented classes a back translation was performed that is not included in the dataset files.

Class Distribution of Auto-Labeled Dataset

Preprocessing
The model receives as input the target formulated as a question along with the corresponding tweet: target question [SEP] tweet. Information on the text preprocessing can be found here.

CC BY 4.0

Exploratory Analysis 2022 Twitter Data

Stance Prediction Results

Tweet Volume Timelines

Ukraine Support Arms Delivery
Nuclear Energy & NPPs Speed Limit Implementation

Takeaways

The classified data provide clues to potential arguments and reasons of a stance-group identified with word frequencies and keyword-in-context consideration.

Target Ukraine Support
German-speaking Twitter users have been debating Germany's political course in the conflict. Users who are in favor or against towards the general support of Ukraine in the conflict often use the same arguments as of the following socio-political related topics that have been discussed and found in 2022 Twitter data. Throughout 2022, it can be seen that the number of users who have a favorable stance towards supporting Ukraine in the conflict outweighs the number of users with an opposing stance (tweet volume timeline).

Favor-Arms Delivery Against-Arms Delivery
The delivery of weapons to Ukraine does not make Germany a participant in the war. The delivery of weapons to Ukraine means the active participation of Germany in the war against Russia.
Military support measures can ensure a faster process of peacekeeping. Arms deliveries prolong the war. Peace negotiations and diplomacy should begin.
Ukraine needs more heavy weapons. Ukraine is a corrupt country that does not represent German values.
The German Armed Forces have no heavy weapons to supply at all.
Favor-Nuclear Energy Against-Nuclear Energy
Germany needs nuclear energy for a secure future. There is no reason to extend the operation time of NPPs, as the energy supply in Germany is secure.
Renewable energies are not (yet) safe enough to exclude nuclear energy as an alternative energy source. NPPs bring with them further problems, such as with regard to the disposal of nuclear waste.
LNG terminals are also not sustainable. Nuclear power is not an environment-friendly technology.
Nuclear energy is a source of danger.
Favor-Speed Limit Against-Speed Limit
A speed limit significantly reduces the emission of pollutants and is thus an important contribution to climate protection. A speed limit does not significantly reduce pollutant emissions.
A speed limit increases safety on the road. A speed limit does not increase road safety.
A speed limit reduces the noise level caused by motor vehicles. A speed limit is a restriction on personal freedom.
A speed limit reduces fuel consumption and thus supports Ukraine against Russia.

About

German Twitter Stance Detection With BERT on the Russo-Ukrainian Conflict

Topics

Resources

License

Stars

Watchers

Forks