Unsupervised neural based hybrid model for sentiment analysis of web/mobile application using public data sources

A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a device to cause the device to:

• extract a text document from a web site;
• tokenize the text document into tokens;
• input the tokens to a word to vector conversion model and running the word to vector conversion model to generate word vectors, a word vector representing a corresponding word in n different dimensions, wherein n is tunable;
• input the word vectors to a term frequency inverse document frequency (TF-IDF) algorithm to generate sentence vectors, a sentence vector of a corresponding sentence generated from word vectors corresponding to words appearing in the corresponding sentence;
• randomly select a subset of the sentence vectors for tagging with sentiment labels;
• train via machine learning a classifier based on the subset of sentence vectors that are tagged;
• run the classifier with the sentence vectors as input to predict a sentiment associated with each of the sentence vectors;
• and
• combine the predicted sentiment associated with said each of the sentence vectors to generate a sentiment associated with the text document.

A machine trained system that determines sentiments in social network communications, comprising:

• a hardware processor;
• and
• a memory device coupled to the hardware processor,
• the hardware processor operable to perform:
• extracting a text document from a web site;
• tokenizing the text document into tokens;
• inputting the tokens to a word to vector conversion model and running the word to vector conversion model to generate word vectors, a word vector representing a corresponding word in n different dimensions, wherein n is tunable;
• inputting the word vectors to a term frequency inverse document frequency (TF-IDF) algorithm to generate sentence vectors, a sentence vector of a corresponding sentence generated from word vectors corresponding to words appearing in the corresponding sentence;
• randomly selecting a subset of the sentence vectors for tagging with sentiment labels;
• training via machine learning a classifier based on the subset of sentence vectors that are tagged;
• running the classifier with the sentence vectors as input to predict a sentiment associated with each of the sentence vectors;
• and
• combining the predicted sentiment associated with said each of the sentence vectors to generate a sentiment associated with the text document.