Let’s Make Your Request More Persuasive: Modeling Persuasive Strategies via Semi-Supervised Neural Nets on Crowdfunding Platforms

Proceedings of NAACL-HLT 2019, pages 3620–3630 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 3620 Let’s Make Your Request More Persuasive: Modeling Persuasive Strategies via Semi-Supervised Neural Nets on Crowdfunding Platforms Diyi Yang∗, Jiaao Chen∗, Zichao Yang, Dan Jurafsky, Eduard Hovy Georgia Institute of Technology, Carnegie Mellon University, Stanford University diyi.yang@cc.gatech.edu {jiaaoc, zichaoy, hovy}@andrew.cmu.edu jurafsky@stanford.edu Abstract Modeling what makes a request persuasive— eliciting the desired response from a reader— is critical to the study of propaganda, behav- ioral economics, and advertising. Yet current models can’t quantify the persuasiveness of re- quests or extract successful persuasive strate- gies. Building on theories of persuasion, we propose a neural network to quantify persua- siveness and identify the persuasive strategies in advocacy requests. Our semi-supervised hi- erarchical neural network model is supervised by the number of people persuaded to take ac- tions and partially supervised at the sentence level with human-labeled rhetorical strategies. Our method outperforms several baselines, uncovers persuasive strategies—offering in- creased interpretability of persuasive speech— and has applications for other situations with document-level supervision but only partial sentence supervision. 1 Introduction Crowdfunding platforms are a popular way to raise funds for projects. For example, Kiva, a peer- to-peer lending platform, has crowd-funded more than a million loans, totaling over $1 billion since 2005. Kickstarter, another online crowdfunding platform, successfully funded 110,270 projects with a total of over 2 billion dollars. Yet most projects still suffer from low success rates. How can we help requesters craft persuasive and suc- cessful pitches to convince others to take actions? Persuasive communication has the potential to shape and change people’s attitudes and behaviors (Hovland et al., 1953), and has been widely re- searched in various fields such as social psychol- ogy, marketing, behavioral economics, and politi- cal campaigning (Shrum et al., 2012). One of the ∗Equal contribution. This work was done when the first two authors were students at CMU. most influential theories in the advertising liter- ature is Chaiken’s systematic-heuristic dual pro- cessing theory, which suggests that people pro- cess persuasive communication by evaluating the quality of arguments or by relying on inferential rules. Some such heuristic rules are commonly used in consumer behaviors; commercial websites may highlight the limited availability of their items “In high demand - only 2 left on our site!” or em- phasize the person in authority “Speak to our head of sales—he has over 15 years’ experience sell- ing properties” to attract potential consumers. Al- though numerous studies on persuasion have been conducted (Chaiken, 1980), we still know little about the way how persuasion functions in the wild and how it can be modeled computationally. In this work, we utilize neural-network based methods to computationally model persuasion in requests from crowdfunding websites. We build on theoretical models of persuasion to operational- ize persuasive strategies and ensure generalizabil- ity. We propose to identify the persuasive strat- egy employed in each sentence in each request. However, constructing a large dataset with persua- sion strategies labeled at the …

Content truncated to 500 words