Sentiment analysis is a subset of NLP and has encountered an outstanding change with the introduction of new approaches to increase its precision and performance. Neural networks and transformers in deep learning are very widely used techniques in sentiment analysis. This paper suggests the most recent research techniques that are currently used in sentiment analysis and also examines the difficulty and limitations of recently used techniques in sentiment analysis. This paper also provides directions for future studies. The increasing effectiveness of deep learning algorithms in sentiment analysis is an important finding from this survey. To analyze text data and determine the underlying sentiment, neural networks, and transformers have shown extraordinary potential. They are especially helpful in traversing the nuances of human language, which present difficulties for traditional machine learning techniques. Despite these developments, there are still difficulties with reliably identifying irony and sarcasm in sentiment analysis tools. To mitigate these issues, future research should concentrate on devising more robust sentiment analysis strategies capable of accurately interpreting sarcasm and irony. Additionally, efforts should be directed toward developing diverse datasets to ensure sentiment analysis techniques do not favor certain demographics or viewpoints.