Product Sentiment Analysis with TuriCreate. Sentient Analysis with TuriCreate and Python was one of the most interestiing projects as I got an indepth first-hand experience to how a machine processes and understands human sentiment.
Using TuriCreate's .count_words() function in the .text_analytics package, we can count the number of words each sentence has. In this case each person has a review on a particular product. Using the above functions and packages we realize the actual number of words from which a true meaning of the review can be deduced. For example, if a review has the word amazing used 5 times and the word bad used twice, we can understand that the person is happy about the recieved product and thus the review could be deemed as positive.
Using an ROC_curve (which is like a Sigmoid Function) we can get a probability of the review which is then stored as a sentiment. On co-tabulating with the reviews we could see that those reviews which are a 5-star or 4-star have high values while the other reviews have decreasing values.
We can also use some specific words, and based on these words a model could be trained. Values for each of these special words can be tabulated and based on these words as features, a Machine Learning model can be created.
The model used to classify a positive or negative review is a Logistic Regression Classifier.