With the rise of machine learning as well as of the social media, Sentiment Analysis has grown rapidly. It focuses on computationally analyzing peoples sentiments, opinions and emotions in regards of products or other topics. There is a large spectrum to sentiment analysis and many ways in machine learning to approach it and its different applications. For our project, we will focus on studying customer reviews, such as for products on Amazon, or commentaries on different topics like those on reddit. This analysis would be to translate or convert the written data into a quantified star like review. Understanding customer satisfaction is crucial for businesses nowadays. Furthermore, it is the basis for political campaigns. Whenever you want to reach people, you have to understand what they think and how they react. However, it is very hard to work with all the tremendous amounts of written data that you get with customer reviews and comments. Therefore, it is very interesting and useful to quantify such data to analyze it fast and easily. If you read a review, you cannot easily say how much a customer liked a product, whereas a number on a scale tells you at the first look to what extent the customer is satisfied - or not. This quantification will speed up and simplify the job of data analysts in marketing: Working with quantitative data is simpler than qualitative one. In our approach, we start with basic classification algorithms seen in class. We will then move to more complex ones using recurrent neural networks and deep learning. By using different algorithms, we are able to see how well they do on the datasets and which ones are more adapted to this problem. Furthermore, as the more complex algorithms do not only take into account individual words but the meaning of whole phrases, we will see the impact of these differences
You will find many ipython notebooks with different approaches in relevant folders.