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

## TL;DR

This paper introduces a method to predict users' emotions at specific locations by analyzing both spatial and temporal patterns using Weibo image data.

## Contribution

A novel method for predicting location-based emotions by incorporating multidimensional spatio-temporal dependencies using graph embedding and attention-based BiLSTM.

## Key findings

- The proposed method achieved 75% prediction accuracy for location emotions.
- It outperformed traditional LSTM and CNN methods in emotion prediction.
- The approach enhances understanding of spatio-temporal emotional patterns in urban areas.

## Abstract

Emotion has significant spatio-temporal characteristics, and predicting the spatio-temporal changes in emotion is an important premise for monitoring the emotional state of urban residents. Most prediction methods focus on the prediction of emotion in time series without considering the spatial properties of emotion. Based on geotagged image data on the Weibo platform from Shanghai, a user location emotion prediction method that considers multidimensional spatio-temporal dependencies between different emotional states is proposed in this paper. The method introduces the HiSpatialCluster algorithm to identify the users’ stay area. Then, the FaceReader algorithm is applied to determine the emotional quadrant of users from image data, and a graph embedding algorithm is employed to obtain the feature vector representing each stay area. Finally, an attention-based BiLSTM method is applied to construct the multidimensional spatio-temporal dependencies of emotion for prediction. Experiments on the Weibo dataset show that the prediction accuracy of location emotion reaches 75%, which is better than that of the single LSTM and CNN method. The results of this paper can not only deepen the understanding of the spatio-temporal variation patterns of emotion but also optimize location-based recommendation services.