Predicting location emotions of users considering multidimensional spatio-temporal dependencies.

Area of Science:

• Computational Social Science
• Geospatial Data Analysis
• Affective Computing

Background:

• Urban residents' emotional states exhibit complex spatio-temporal dynamics.
• Existing emotion prediction models often neglect spatial dimensions, focusing solely on time series.
• Understanding and predicting these spatio-temporal emotion patterns are crucial for urban well-being monitoring.

Purpose of the Study:

• To propose a novel method for predicting user location-based emotions by incorporating multidimensional spatio-temporal dependencies.
• To leverage geotagged social media data for a more comprehensive understanding of urban emotional landscapes.
• To enhance the accuracy of emotion prediction by considering spatial context.

Main Methods:

• Utilized geotagged Weibo data from Shanghai.
• Applied HiSpatialCluster algorithm to identify user stay areas.
• Employed FaceReader algorithm for emotion detection from images and graph embedding for stay area feature extraction.
• Developed an attention-based BiLSTM model to capture spatio-temporal emotion dependencies.

Main Results:

• Achieved a location emotion prediction accuracy of 75% on the Weibo dataset.
• Demonstrated superior performance compared to single LSTM and CNN prediction methods.
• Successfully modeled multidimensional spatio-temporal dependencies in user emotions.

Conclusions:

• The proposed method effectively predicts spatio-temporal emotion changes in urban residents.
• Findings deepen the understanding of emotion variation patterns in urban environments.
• Results offer potential for optimizing location-based recommendation services and urban planning.