IDENTIFICATION OF CERTAIN EMOTIONS IN TEXT IDENTIFICATION OF CERTAIN EMOTIONS IN TEXT (NATURAL LANGUAGE PROCESSING) Mentor: Prof. Amitabha Mukherjee Pranjal Divyanshu CS365: ARTIFICIAL INTELLIGENCE 5/11/2015 1
IMPORTANCE + SCOPE 5/11/2015 2
METHODOLOGY: OVERVIEW SUPERVISED LEARNING Parsing of text using “Stanford PCFG Parser” Sentiment Categorization using Supervised Learning(Alchemy API) UNSUPERVISED LEARNING Removal of Stop-Words Created term by document matrix Applied PLSA on this matrix 5/11/2015 3
SUPERVISED LEARNING PCFG parser result : “He might have lung cancer. It s just a rumor... but it makes sense. He is very depressed and that s just the beginning of things” negative </score Sentiment Analysis Result: 5/11/2015 4
UNSUPERVISED LEARNING “He might have lung cancer. It s just a rumor... but it makes sense. He is very depressed and that s just the beginning of things” EmotionProbability Sad e+000 Fear e-086 Joy e+000 Guilt e+000 Shame e+000 Anger e-034 Disgust e-132 5/11/2015 5
PLSA 6 P LSA aims to discover something about the meaning behind the words; about the topics in the documents
SAMPLE RESULT: 5/11/2015 7
DIFFERENCE IN OUR APPROACH Use of PLSA instead of LSA which “Carlos Strapparava” and “Rada Mihalcea” have done in their work. Instead of annotated corpus which is used by “Carlos Strapparava” and “Rada Mihalcea” for applying LSA we are updating the term document matrix for later use of PLSA. Removal of certain words that don’t contribute to emotions. 5/11/2015 8