EEG-SVRec: A Dataset with User Multidimensional Affective Engagement Labels for Short Video Recommendation

Core Concepts

This dataset provides EEG signals, user behavior logs, and multidimensional affective engagement scores (valence, arousal, immersion, interest, visual, and auditory) for short video interactions, enabling a deeper understanding of user preferences and cognitive activity in short video recommendation scenarios.

Abstract

The EEG-SVRec dataset was constructed through a user study involving 30 participants who browsed short videos in personalized, randomized, and mixed session modes. During the browsing stage, participants' EEG and ECG signals were continuously collected, and their interactions (liking and viewing duration) were logged. After each session, participants provided self-assessments of their multidimensional affective engagement (valence, arousal, immersion, interest, visual, and auditory) for the videos they had viewed.
The dataset contains 3,657 interactions with corresponding EEG/ECG signals, user behavior logs, and multidimensional affective engagement scores. Statistical analysis of the data reveals interesting insights, such as the strong correlation between user liking/viewing and their reported interest and immersion levels. Benchmark experiments also demonstrate the potential of incorporating EEG signals to enhance recommendation performance.
The EEG-SVRec dataset presents opportunities for various research directions, including developing human-centric evaluation metrics, uncovering the relationship between user behavior and cognitive activity, designing EEG-guided recommendation algorithms, and improving accessibility for users with disabilities in short video streaming.

Stats

The dataset contains 3,657 interactions from 30 participants involving 2,636 short videos.

Quotes

"EEG, as a neuroelectrical signal, containing rich spatial, temporal, and frequency band information about human experience, can be used to study the underlying neural mechanisms and can reflect relevant information about user cognition, emotion, and attention."
"Providing high temporal resolution data, the application of EEG technology in the Information Retrieval (IR) domain has been proven to be useful."

Key Insights Distilled From

EEG-SVRec

by Shaorun Zhan... at arxiv.org 04-02-2024

https://arxiv.org/pdf/2404.01008.pdf

Deeper Inquiries

How can the insights gained from the EEG-SVRec dataset be leveraged to develop more inclusive and accessible short video recommendation systems for users with disabilities?

The insights from the EEG-SVRec dataset can be instrumental in developing more inclusive and accessible short video recommendation systems for users with disabilities. By analyzing the EEG and ECG signals alongside multidimensional affective engagement annotations and user behavior data, researchers can gain a deeper understanding of how individuals with disabilities interact with short video content. This understanding can lead to the following advancements:

Personalized Recommendations: Leveraging EEG signals can help tailor recommendations to the specific cognitive and emotional responses of users with disabilities. By incorporating EEG data into recommendation algorithms, systems can better adapt to individual preferences and needs.

Accessibility Features: Insights from the dataset can guide the development of accessibility features in recommendation systems. For example, recommendations can be optimized for users with visual or auditory impairments by considering their unique affective experiences and cognitive activities.

User-Centric Design: Understanding the cognitive and emotional states of users with disabilities can inform the design of user interfaces and recommendation interfaces that are more intuitive and user-friendly for this demographic.

Inclusive Content Curation: By considering the affective engagement scores and EEG signals of users with disabilities, recommendation systems can curate content that is more inclusive and diverse, catering to a wider range of preferences and needs.

Overall, the insights from the EEG-SVRec dataset can pave the way for the development of recommendation systems that are not only more personalized but also more inclusive and accessible for users with disabilities.

What are the potential limitations and biases in the current dataset, and how can future research address them to improve the generalizability of the findings?

The current dataset, while valuable, may have limitations and biases that could impact the generalizability of the findings. Some potential limitations and biases include:

Sample Size: The dataset was constructed with a relatively small sample size of 30 participants, which may not fully represent the diversity of users in real-world scenarios. Future research can address this by expanding the participant pool to include a more diverse range of individuals.

Algorithmic Bias: The underlying recommendation algorithms used to collect the data may introduce biases that affect the recommendations and user interactions. Future research can mitigate this by employing unbiased algorithms or adjusting for algorithmic biases in the analysis.

Contextual Bias: The dataset's focus on personalized and randomized videos of 30-60 seconds may introduce biases in user behavior and affective engagement. Future research can enhance generalizability by including a wider variety of video lengths and types in the dataset.

To improve the generalizability of the findings, future research can focus on:

Diverse Participant Demographics: Including a more diverse participant pool in terms of age, gender, and background to capture a broader range of user experiences.

Cross-Validation and External Validation: Conducting cross-validation and external validation studies to ensure the robustness and generalizability of the findings across different datasets and scenarios.

Bias Mitigation Strategies: Implementing bias mitigation strategies in data collection, analysis, and algorithm design to reduce biases and ensure fair and unbiased recommendations for all users.

By addressing these limitations and biases, future research can enhance the reliability and applicability of the insights gained from the EEG-SVRec dataset.

Given the temporal dynamics of user behavior and emotions observed in the dataset, how can recommendation algorithms be designed to better capture and respond to these evolving user states over time?

The temporal dynamics of user behavior and emotions observed in the EEG-SVRec dataset present an opportunity to design recommendation algorithms that can adapt to and respond to these evolving user states over time. Here are some strategies to achieve this:

Dynamic User Modeling: Implement dynamic user modeling techniques that continuously update user preferences and emotional states based on real-time EEG and ECG signals. This allows recommendation algorithms to adapt to changes in user behavior and emotions.

Contextual Bandit Algorithms: Utilize contextual bandit algorithms that can optimize recommendations in real-time based on the evolving user states. By incorporating EEG signals as contextual information, algorithms can make more personalized and timely recommendations.

Reinforcement Learning: Employ reinforcement learning techniques to learn and optimize recommendation policies based on user feedback and EEG signals. This enables algorithms to adapt and improve recommendations over time through interaction with users.

Long Short-Term Memory (LSTM) Networks: Use LSTM networks to capture the sequential patterns in user behavior and emotional states over time. By modeling the temporal dependencies in the data, algorithms can make more accurate predictions and recommendations.

Adaptive Learning Rates: Implement adaptive learning rate mechanisms that adjust the model parameters based on the changing user states. This ensures that the recommendation algorithms can quickly adapt to new information and user preferences.

By incorporating these strategies, recommendation algorithms can better capture and respond to the evolving user states over time, leading to more personalized and effective recommendations in short video recommendation systems.