Persuasiveness of text messages generated by machine learning language model

The objective of this study is to examine the effectiveness of our algorithm that applies a machine learning language model to convert the message to be memorable and persuasive. We designed an algorithm that takes an input sentence, and by changing the sentence to be more general in the syntax level, and more distinctive at the lexical level with the masked language model, we will convert the input sentence to be more memorable and more effective at persuading people to change their attitude. We came up with two versions of the algorithm that have a slight difference of focus on the attributes of the output sentences, and we designed an experiment in Mechanical Turk to compare these two versions of the algorithm. In this study, we first introduce an algorithm that is consisted mainly of two steps. First, the algorithm will convert an input text message to be more general at the level of sentence structure, then the algorithm will compose a sentence with more distinctive words. We created two versions of the algorithm that have their advantages: one version of the algorithm focus on replacing original words with distinctive synonymous, thus keeping the original meaning of the input sentence during the generation process; the other version focuses on replacing words with more distinctive vocabularies, not only restricted to synonyms, thus making the sentence with more variations for vocabularies and more memorable to readers. To compare these two versions of an algorithm for their effectiveness at converting a sentence to be more memorable and more persuasive, we experimented on Mechanical Turk. Our experiment results show that sentences with more general sentence structure and more distinctive words contribute to memorability and persuasiveness of sentences, but the results also suggest several improvements should be made for the algorithms.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2021

Author dc:creator

• Zheng, Bo

Contributors dc:contributor

• Sundaram, Hari

• persuasive message generation
• machine learning

Statement dc:rights

• Copyright 2020 Bo Zheng

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/109448

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/109448