Search

Sentiment analysis (SA) plays an important role in inferring sentiment or emotion from text and visual contents, such as images and videos to determine the overall contextual polarity of a document. Today, image recognition and classification are rapidly growing fields in the area of machine learning (ML). This paper presents a review of open-source machine learning algorithms, built using neural network-based frameworks such as TensorFlow and Keras, to serve as a benchmark for bespoke SA algorithms. This research also advocates open-source scikit-learn models for text tweets and image classification.Two prominent, publicly available and manually annotated benchmark text and image datasets were used to enable and assist in the correlation of this work with existing, present and future avant-garde and innovative methods. Quantitative results across four statistical criteria, including precision, recall, F1-score and accuracy compare favourably to the often complicated and tailored state-of-the-art methodologies developed. For SA, empirical results suggest deep-learning model frameworks to outperform scikit-learn algorithms. All experiments were conducted on computer hardware comprising 64GB of RAM and a NVIDEA GeForce RTX 2080 Ti GPU.