Home / Summaries / Human mind in the digital world / learning-rewards-social

Paper presentation

Q: What is the title of the presentation and the original source it is based on?

Presentation Title: A computational reward learning account of social media engagement Original Source: Lindström, B., Bellander, M., Schultner, D. T., Chang, A., Tobler, P. N., & Amodio, D. M. (2021). Nature Communications, 12(1), 1311.

39 important questions on Paper presentation

What are the steps to play the game illustrated in the given content?

Open Instagram.
Check the number of likes on the second to last post.
Note when the second to last post was made.
Open the last post on Instagram.
Determine the time difference between the last post and the second to last post.

What is the title of the presentation and the original source it is based on?

Presentation Title: A computational reward learning account of social media engagement
Original Source: Lindström, B., Bellander, M., Schultner, D. T., Chang, A., Tobler, P. N., & Amodio, D. M. (2021). Nature Communications, 12(1), 1311.

What is Reinforcement Learning (RL) Theory?

Involves an agent making decisions through learned behaviors.
Rewards are given for correct actions.
Punishments are administered for incorrect actions.
This approach is derived from animal learning theory.

What does post latency refer to?

It is the duration between two consecutive posts.
It assesses the wait time before posting again.

How is the Average reward rate R determined?

Calculated based on the number of likes a post receives.
Denoted as R and derived from the likes within a specific time period after posting, t−l.

What is the relationship between excessive social media engagement and addiction?

Excessive use of social media can lead to behavior akin to addiction.
Users are commonly motivated by the desire for positive feedback online.

How is social media engagement commonly portrayed and what psychological concept does it relate to?

Social media usage is often compared to operating a Skinner Box, indicating a reliance on social rewards.
It is associated with the psychological concept of reinforcement learning (RL).

How is social media behavior similar to a Skinner Box?

Seeks social rewards akin to likes, mirroring animals' learning in experiments.
Reinforcement learning theory states that increased frequency of posts occurs with more rewards.
Slower actions can result in fewer opportunities for rewards.

What challenges in research are outlined in the introduction?

Evidence regarding social media as a form of reward-based action is not conclusive.
Past research on online social incentives often relies on self-reported data.
There is a shortage of direct evidence supporting a social reward learning model for social media behavior.
It is uncertain if fundamental reward learning mechanisms can account for actual social media behavior.

What was the focus of the research by Lindstrom et al. in 2021 regarding social media behavior?

The research focused on using a computational approach to analyze large scale social media datasets.
It aimed to test whether and how reward learning mechanisms affect social media behavior.
The purpose of the study was to gain new insights into emergent human interaction modes and apply a learning theory model to real-life social behavior on a massive scale.

What method was applied to test social media engagement in the study?

Computational models based on reinforcement learning were applied.
These models examined if engagement is driven by social rewards like "likes".
Social media behavior's sensitivity to social rewards was tested by measuring response latencies, looking for reward learning signatures.

What does the computational model diagram illustrate about social media posts and rewards?

The diagram illustrates the timing of posts and receiving of social rewards.
Symbols represent a user's post (camera icon) followed by a reward (heart icon).
\( \tau_{Post1} \) and \( \tau_{Post2} \) indicate time intervals between posts and rewards.

How does the graph 'e' in the computational model relate to reward prediction errors and social rewards?

Graph 'e' depicts the relationship between reward prediction errors (\( T_{RPE} \)) and social rewards.
It compares low \( R \) (reward) and high \( R \) conditions.
Lower \( T_{RPE} \) values suggest fewer prediction errors with higher social rewards.

What influences the time between social media posts according to the provided information?

Social rewards like likes influence the time between posts (τPost).
Users adjust posting frequency to maximize social rewards.

How does posting frequency relate to effort and rewards based on the information given?

Longer wait between posts lowers the effort required.
Posting too quickly increases effort and can reduce rewards.

What happens when users wait too long to post on social media?

Users miss out on potential rewards.
Expectation of more likes leads to the feeling that they should post sooner.

What does the simulation reveal about posting frequency and expected rewards?

People tend to post more often when expecting higher rewards.
Simulation outcome confirmed by 1,000 synthetic users.

What is the quantitative law of effect in the context of social media posting behavior?

Behavior increases with more rewards like more likes.
Posting more slows after reaching a point, as depicted by the hyperbolic curve.
Hyperbolic vs linear function comparison shows the likes and response rates relationship.

What does the hypothesis suggest about social media behavior in relation to basic learning mechanisms and social rewards?

Suggests social media taps into basic learning mechanisms.
Social media behavior expected to show similar patterns between response latency and reward rate.
Response latency defined as time between successive social media posts.
Implies people maximize social rewards through reinforcement.

According to the research question, what aspect of social media behavior is being investigated?

Investigates whether social media behavior is affected by social rewards.
Queries sensitivity of social media actions related to social rewards.

What was the method used in Study 1 and what data was considered?

Analyzed a dataset of 851,946 posts from 2,039 participants in a photography contest on IG in 2014.
Removed accounts with fewer than 10 posts to focus on users with enough data to analyze learning patterns.
Acknowledged the possibility of engagement from fraudulent accounts and "fake likes".
Utilized a cross-sectional study design.

What does Study 1 Results suggest about the Reinforcement Learning (RL) model's explanation of behavior on Instagram?

The RL model is more effective in explaining behavior on Instagram than a model lacking a learning component where likes do not correlate with behavior.
Post latency (tPost) is less with high reward prediction error (R̂) than with low R̂.
Users with more followers place less subjective value on each like, possibly indicating desensitization to likes for those with many followers.

How does the number of Instagram followers seem to affect the perceived value of likes according to the study?

- People with a higher number of Instagram followers perceive less value in each like, which may suggest habituation to receiving likes among those with a large follower count.

Based on the graphs presented, what can be inferred about the comparison between data and model simulation in terms of post latency (tPost)?

- Graph B shows that both actual data and model simulations indicate a decrease in post latency (tPost) as reward prediction error (R̂) increases from low to high.

What prompted Study 2 to be conducted as an extension of Study 1?

- Possibility of fake likes in the first study led to the development of Study 2.

How many datasets were analyzed in Study 2, and what topics did they cover?

- Study 2 analyzed 3 datasets from forums focused on Men's fashion (543), Women's fashion (773), and Gardening (813).

What method was employed to collect data for Study 2?

- Data was collected using web scraping techniques on publicly accessible data.

How many posts and individuals were included in the final dataset of Study 2?

- The final dataset included 190,721 posts from 2,127 individuals.

What results did Study 2 find regarding the RL model and non-learning models in explaining behavior on social platforms?

RL model performed better than non-learning models for explaining behavior on three forums.
It was found to be supportive of RL theory.
Predicts social media use dynamics, independent of platform or topic.
Latency between posts was lower with high rewards, as predicted by RL theory.
Social comparison may influence reward learning on social media.

What individual differences were identified among participants in Studies 1 and 2, and how were participants classified based on these studies?

Participants in Studies 1 and 2 were identified to have individual differences in learning rate and effort cost sensitivity.
Learning rate differences are due to genetic and developmental factors.
Effort cost sensitivity is related to the dopaminergic system.
Participants were classified into four groups based on study results.

What does the 2D scatter plot on the left illustrate regarding the individual differences among the four clusters?

Represents the relationship between Learning Rate and Effort Cost
Identifies four unique clusters distributed across the plot
Demonstrates variation in Learning Rate and Effort Cost sensitivities among the clusters

What information is conveyed in the PCA plot and bar graph on the right?

PCA plot shows a distribution of individuals in a reduced-dimensionality space (PC1 vs PC2)
Bar graph compares standardized parameter values of α (learning rate), β (initial response policy), and C (effort cost sensitivity) across clusters

How are parameters α, β, and C related to the clusters according to the bar graph?

Parameters α, β, and C vary among clusters as shown by bar heights
α represents learning rate, β represents initial response policy, C represents effort cost sensitivity
Each parameter exhibits a distinct pattern of standardization across clusters

What characteristics define individuals in Cluster 1 according to the presented groups?

Low Learning Rate
Insensitive to social rewards
Unlikely to alter behavior based on received likes
Poor fit with the Reinforcement Learning (RL) model predictions

How do individuals in Cluster 2 respond to social rewards, and what is their effort cost?

Average Learning Rate
Highly responsive to social rewards
Tend to post frequently when they receive likes
Low effort cost making it easier for them to post

What are the attributes associated with individuals in Cluster 3?

Intermediate Profile
Exhibit moderate sensitivity to social rewards
Moderate effort regarding their posting activity
Positioned between Cluster 2 and Cluster 4 in terms of behavior

Describe the individuals in Cluster 4 in terms of their learning rate and cost associated with posting.

High Learning Rate
Respond well to social rewards
Often post in reaction to receiving likes
High effort cost signifies that it requires more effort for them to post

What was the method used in Study 3?

Conducted an online experiment with the manipulation of social rewards.
Participants' posting response times were observed.
A total of 176 subjects posted memes freely during a 25-minute session.
They collectively made 2,206 posts.
Participants received variable likes per post, categorized as low (0–9 likes) and high (10–19 likes) rewards.
For 25% of the participants, the reinforcement learning (RL) model fit was low, labelled as cluster 1.

What were the findings of Study 3 regarding post latency and social reward rates?

Post latency increased with lower social reward rates (0-9 likes/post).
A 10.9% difference in latency was noted between lower and higher social reward rates.
The findings supported the hypothesis that social rewards can influence response latencies.
The experiment was validated in its effort to reflect real-world psychological behavior.

The question on the page originate from the summary of the following study material:

Human mind in the digital world

View summary

A unique study and practice tool
Never study anything twice again
Get the grades you hope for
100% sure, 100% understanding

Remember faster, study better. Scientifically proven.