Large-Scale Estimation and Analysis of Web Users' Mood from Web Search Query and Mobile Sensor Data

Area of Science:

• Computational Social Science
• Human-Computer Interaction
• Affective Computing

Background:

• Estimating web user mood states is crucial for personalized services but faces challenges in data collection and ground truth.
• Existing methods struggle with non-invasive and accurate mood state determination.

Purpose of the Study:

• To develop a robust and non-invasive method for estimating web user mood states.
• To create a nationwide mood score for analyzing population-level mood dynamics.
• To investigate the relationship between public mood, major events (like COVID-19), and user behavior.

Main Methods:

• A two-step model was built: first, estimating mood from search queries, then supplementing with mobile sensor data for ground truth.
• Large-scale data analysis involving over 11 million users was conducted on a commercial platform.
• Development of a nationwide mood score aggregating individual mood values.

Main Results:

• The two-step model significantly improved the accuracy of mood state estimation.
• The nationwide mood score revealed daily and weekly mood rhythms.
• Mood fluctuations were observed during the COVID-19 pandemic, inversely correlating with new case numbers.
• The system detected real-time mood shifts in response to major news events and identified mood patterns associated with advertisement engagement.

Conclusions:

• The proposed method offers an effective and scalable approach to estimate web user mood states.
• The nationwide mood score provides valuable insights into population-level psychological responses to societal events.
• This research demonstrates the potential of leveraging digital footprints for understanding and tracking collective human emotion.