Social Media Use is Predictable from App Sequences: Using LSTM and Transformer Neural Networks to Model Habitual Behavior

Overview

• This paper introduces a novel approach to studying social media habits by predicting sequential smartphone user behaviors.
• Rather than relying on self-reported data or simple behavioral frequency measures, the researchers examine the predictability of social media use within and between individuals.
• The study leverages Long Short-Term Memory (LSTM) and transformer neural networks to model patterns in user behavior.

Plain English Explanation

The researchers wanted to understand how people use social media on their smartphones in a more detailed way than previous studies. Instead of just asking people how often they use social media or tracking how much time they spend on it, the researchers looked at the sequences of actions people take on their phones.

By using advanced machine learning techniques like LSTMs and transformers, the researchers were able to predict people's social media use patterns. They found that social media use is quite predictable, both at the individual level and across all the study participants.

Interestingly, the researchers discovered that how predictable someone's social media use is doesn't necessarily correlate with how much they use social media overall. This suggests their approach is capturing a distinct aspect of smartphone habits beyond just frequency of use.

Technical Explanation

The researchers developed several different modeling approaches to predict social media use:

1. Global models: Training a single model on data pooled from all participants
2. Idiographic person-specific models: Training individual models for each participant
3. Fine-tuned global models: Starting with the global model and further training it on each participant's data

Surprisingly, the global models performed just as well as the person-specific models, indicating they were able to capture a variety of individual behavioral patterns. The researchers also found that the predictability of an individual's social media use was not strongly related to their overall frequency of smartphone or social media use.

Critical Analysis

A strength of this research is the use of sequential behavioral data and advanced neural network models to gain insights beyond what self-report or simple frequency measures can provide. This aligns with calls in the literature to move beyond usage metrics and explore the dynamics of social interaction.

However, the paper does not delve into potential privacy concerns around modeling individual smartphone behaviors in such detail. There are also open questions about how generalizable these findings are, as the study was conducted with a relatively small, homogeneous sample.

Additionally, the researchers do not provide much theoretical grounding or interpretation for why some individuals' social media use is more predictable than others. Further probing of these large-scale behavioral models could yield valuable insights.

Conclusion

This paper presents a novel approach to studying social media habits that goes beyond simple usage metrics. By modeling the sequential nature of smartphone behaviors, the researchers were able to uncover predictable patterns both within and between individuals.

The finding that global models can capture individual differences suggests these techniques may be useful for understanding player behaviors in other domains as well. Overall, this research highlights the potential of advanced machine learning to shed new light on technology-mediated human behaviors.