Argument Recommender Systems for Social Discussion

AI Quick Summary

There is a huge need for structured referendum discussion nowadays and even more in the future. In order to work on different technologies like algorithms, hci and more, there must be a first simple prototype. The objects are statements which have properties like source, reference, content and later on associations to a factcheck-rating and so on. Stances identification in Political Discourses through Arguments UsageUnderstanding and creating an informed opinion about facts of everyday life is becoming increasingly hard for users, due to the multiplicity of the sources and the crossing between traditional media and the user generated contents (UGC). This is particularly important in the context of elections, as citizens need to form an opinion to guide their voting based on partial coverage of subject in multiple interviews and debates. This thesis concentrates on finding a suitable approach to extract arguments used in politics to support or to oppose a certain position [4]. This implies extracting supposed relationships of causal nature from natural language [5-7].Method: the suggested approach is to understand and adapt the method used by “causenet” [1], thus employing pattern-based causal extraction from TV debates transcripts of the US presidential election (2020 [2] and 2016 [3]). Once validated the method, this should be applied to the Democratic presidential primary debates: by merging the political stance of each speaker, it is possible to find a view over the arguments used pro/against a specific subject, giving a fuzzy (membership degree) of pro/contra classes for each argument.Risks: there two main risks related to this proposal. 1) It is possible that the method used in causenet is unable to generalize for less structure data, such as transcript of dialogues. 2) The obtained results, in particular with respect to the argumentative structure into membership degrees, can be strongly affected by definition of the political stance, that is somehow subjective. This can result into a limited generalisation.Main objective of this work will be to develop a generic approach to analyse political debate transcripts/interviews and to characterize theembedded argumentative structure in term of pro and contra arguments. To simplify the context and offer a good coverage of multiple subjects, the benchmark will be performed using TV debates from US presidential elections (2020 and 2016). Subsequently, a fuzzy membership can be computed, using the stance of the politicians to obtain a broader characterization of general assertions adopted in the political argumentative panorama.The expected result is a small demonstrative (MVP) graphical tool allowing a generic user to interact and make sense out of the knowledge structure extracted. Useful features, such as the possibility for a user to input research keywords or subjects of interest should be included, together with filtering capabilities.

Project Idea Metadata

• Project Idea Name: Argument Recommender Systems for Social Discussion
• Date: 11/27/2021 8:35:37 AM
• Administrators:
  • Benjamin Emmenegger

Project Idea Description

Stances identification in Political Discourses through Arguments Usage

Understanding and creating an informed opinion about facts of everyday life is becoming increasingly hard for users, due to the multiplicity of the sources and the crossing between traditional media and the user generated contents (UGC). This is particularly important in the context of elections, as citizens need to form an opinion to guide their voting based on partial coverage of subject in multiple interviews and debates. This thesis concentrates on finding a suitable approach to extract arguments used in politics to support or to oppose a certain position [4]. This implies extracting supposed relationships of causal nature from natural language [5-7].

Method: the suggested approach is to understand and adapt the method used by “causenet” [1], thus employing pattern-based causal extraction from TV debates transcripts of the US presidential election (2020 [2] and 2016 [3]). Once validated the method, this should be applied to the Democratic presidential primary debates: by merging the political stance of each speaker, it is possible to find a view over the arguments used pro/against a specific subject, giving a fuzzy (membership degree) of pro/contra classes for each argument.

Risks: there two main risks related to this proposal. 1) It is possible that the method used in causenet is unable to generalize for less structure data, such as transcript of dialogues. 2) The obtained results, in particular with respect to the argumentative structure into membership degrees, can be strongly affected by definition of the political stance, that is somehow subjective. This can result into a limited generalisation.

Main objective of this work will be to develop a generic approach to analyse political debate transcripts/interviews and to characterize the

embedded argumentative structure in term of pro and contra arguments. To simplify the context and offer a good coverage of multiple subjects, the benchmark will be performed using TV debates from US presidential elections (2020 and 2016). Subsequently, a fuzzy membership can be computed, using the stance of the politicians to obtain a broader characterization of general assertions adopted in the political argumentative panorama.

The expected result is a small demonstrative (MVP) graphical tool allowing a generic user to interact and make sense out of the knowledge structure extracted. Useful features, such as the possibility for a user to input research keywords or subjects of interest should be included, together with filtering capabilities.

There is a huge need for structured referendum discussion nowadays and even more in the future. In order to work on different technologies like algorithms, hci and more, there must be a first simple prototype. The objects are statements which have properties like source, reference, content and later on associations to a factcheck-rating and so on.