Background: Text data includes things like customer reviews and complaints,tweets from social media platforms. When analyzing text-based data, the SentimentModel is used. Understanding news headlines, blogs, the stock market, politicaldebates, and film reviews some of the areas where sentiment analysis is used. Theresults of a sentiment analysis may be used to aid in evaluating whether a reviewis favorable, negative, or neutral. In this thesis we explore the performance of some algorithms.this paper looks at how people feel and think about a furniture store’snew product based on online reviews of it.

Objectives: The problems with natural language processing, on the other hand,make it harder for sentiment analysis to work well and be accurate (NLP). In thepast few years, it has been shown that deep learning models are a promising wayto solve some of NLP’s problems. This paper looks at the most recent studies thatused deep learning to solve problems with sentiment analysis and their performancemetrics

Methods: The literature review is done to figure out which algorithms are best forachieving the above goals. An experiment is done to understand how deep learningworks and what metrics are used to figure out which model is the best for sentimentanalysis. Several datasets have been used to test models that use the term frequencyinversedocument frequency and word embedding.

Results: The experiment indicated that the CNN model strikes the best balancebetween how fast it works and how well it works. When used with word embedding,the RNN model was the most accurate, but it took a long time to process and didn’twork well with TF-IDF. The processing times and results of DNN are about average.

Conclusions: The primary objective of this research is to learn more about thepopular deep learning models and related approaches that have been used for sentimentanalysis of social network data. Before feeding it to deep learning models, wechanged the data using TF-IDF and word embedding. Architectures for DNN, CNN,and RNN were looked into after performing the literature review. The processingtime gap was fixed, and the best combination was found.