KOSI- Key Object Detection for Sentiment Insights.

Abstract

This paper explores an original approach of using computer vision, data mining and an expert system to facilitate marketers and other interested parties to take automated data-driven decisions with the use of actionable insights. The system uses a state-of-the-art algorithm to retrieves all the images of a desired Instagram user profile. Then, the data is passed through a combination of different convolutional neural networks for object detection and a rule-based translation system to determine the interests of this profile user. Further, using a separately trained convolutional neural network with an original dataset developed as part of this study, personality insights are derived. The results from the conducted experiments yield a satisfactory prediction of interests and not very promising results for the personality prediction.