Information Nudge and Its Effects on Douyin's Disabling Personalized Recommendation Algorithm Function (Pilot)

How to cite

Lyu, Yang; Meng, Juanjuan; Gao, Yu. 2025. "Information Nudge and Its Effects on Douyin's Disabling Personalized Recommendation Algorithm Function (Pilot)." AEA RCT Registry. AEARCTR-0015159. https://doi.org/10.1257/rct.15159-1.2

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Yang Lyu* , Juanjuan Meng , Yu Gao

* Primary investigator

Registry ID
AEARCTR-0015159
DOI
https://doi.org/10.1257/rct.15159-1.2
First registered
Last updated
Status
In development
Trial dates
2025-05-18 – 2025-07-06
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/15159

Abstract

Douyin utilizes personalized recommendation algorithms to provide users with more relevant content and services tailored to their preferences. However, this feature may also lead to negative effects such as overuse and the creation of "information cocoons." The Douyin app offers a "One-Click Disable Personalized Recommendation Algorithm" function, after which users no longer receive content recommendations based on personalized factors.

This project aims to explore how providing information can more effectively improve users’ willingness to disable personalized recommendations. It also investigates changes in users' algorithm settings, usage patterns, and experiences after receiving such information. The findings will provide insights into helping digital platform users prevent overuse and avoid "information cocoons" while contributing to a deeper understanding of user psychology and behavior.

Intervention

Our intervention is to provide information to Douyin users about the potential positive effects if they disable the personalized recommendation algorithm on Douyin, and then we observe whether they actually disable it and how their usage time changes. There are two kinds of information as treatment in our experiment: (1) Disabling the personalized recommendation algorithm can help mitigate overuse (based on Gao et al., 2024); (2) Disabling the personalized recommendation algorithm can broaden information channels (based on Levy, 2021). In control group, we provide irrelevant information with comparable text length in Chinese, as active control.

Experimental design

The core experimental module will randomly assign participants to three groups: Control, Treatment 1, and Treatment 2. Each group will receive two rounds of information, spaced one week apart. Participants will upload screenshots of their Douyin app homepage and the "Disable Personalized Recommendation Algorithm" interface to verify algorithm settings. Additionally, there will be questions for other relevant data.
Process Overview:
1. Control: Receives two rounds of unrelated information of comparable length to the experimental groups.
2. Treatment 1: In the first round, informed about the anti-addiction effects of disabling the algorithm; in the second round, informed about its effect on broadening information channels.
3. Treatment 2: In the first round, informed about the effects of broadening information channels; in the second round, informed about its anti-addiction effects.
This crossover experimental design will help identify the upper and lower bounds of the impacts of the two types of information.

Additionally, some participants may disable the personalized recommendation algorithm after receiving the experimental information but later re-enable it. Potential reasons for this behavior include present bias, expectation bias, and projection bias related to their utility functions. Therefore, we include questions related to these mechanism in our surveys to tease out the channels.

As for the timeline, there will be firstly a recruitment survey. One week later, baseline survey + wave 1 information. Another week later, wave 2 information. Another week later, endline survey. And there will be also a long-run tracking survey one month after the endline survey.

Design details

Randomization method
Randomization by computer
Randomization unit
Individuals
Clusters
0
Observations
1200 Douyin users
Arms
400 users control, 400 users receive treatment 1, 400 users receive treatment 2
Primary outcomes
(1) Whether the Douyin user disables the personalized recommendation algorithm; (2) The usage time on Douyin.
Secondary outcomes
(1) Opinion polarization; (2) Subject well-being; (3) Working efficiency; (4) Preference for different video topics on Douyin; (5) Attitudes towards Douyin platform and personalized recommendation algorithm.

Cited by

No citation data available for this registration yet.

Keywords