Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500
Emotion Recognition from Poems by Maximum Posterior Probability Sreeja. P.S
G.S. Mahalakshmi
Department of Computer Science CEG, Anna University Chennai, India
[email protected]
Department of Computer Science CEG, Anna University Chennai, India
[email protected]
Abstract—Poem is a type of literature designed to convey, ideas, emotions, and experiences in a brilliant way. In this article, we discuss the automatic emotion recognition of poems written in English. This is a pioneering approach in emotion recognition from poems. Emotions from the poems, classified into nine emotions, based on ‘Navarasa’ under ‘Rasa Theory’ which is described in ‘Natyashastra’ written by ‘Bharatha Muni.' The nine basic emotions such as Love, Sadness, Anger, Hatred, Fear, Surprise, Courage, Joy, and Peace, classified as “Navarasa”. As to our knowledge, we are not familiar about a text corpus of poems based on nine emotions, We have manually created an emotion tagged corpus from poems in English. The corpus created is from an exhaustive collection of poems of Indian poets from the period 1850-1950. The poems are mined from the web, and we applied ten cross fold Naïve Bayes classifier to recognize the emotion of a poem by maximum likelihood probability Keywords-Emotion Recognition; Emotion Analysis; Emotion Annotation; Emotion Corpus; Naïve Bayes Classifier
I.
INTRODUCTION
Emotion recognition is one of the dynamic fields in Natural language processing. Recent research into human-machine interaction has a vital role in emotional reactions. Automatic emotion recognition phenomena are studied in opinion mining, market analysis, affective computing, Sentiment Analysis, etc. Humans by nature are emotionally affected by reading poems. Emotion recognition in poems is an important task, as poetry databases or poem websites are growing and number. The retrieval of poem or lyric by emotion has various applications such as verse selection from poetry websites. It can be applied in machine translation of poems, poetic therapy, etc. There are many datasets available for emotion recognition but limited to our knowledge we haven’t find emotion corpus for poems in Nine category such as Love, Sadness, Joy, Fear, Hatred, Courage, Anger, Surprise and Peace. In this paper, we focus on an approach to recognize emotion from English poems based on “Navarasa” described in “Natyashastra.” We introduce a novel corpus of English poems of Indian Poets in 1850-1950. In this article, emotions are categorized into nine
types such as Love, Sadness, Joy, Fear, Hatred, Courage, Anger, Surprise and Peace. ‘Navarasa’ is based on ‘Rasa Theory’ given by ‘Bharatha Muni’ in ‘Natyashastra. ‘Bharata Muni,' is the father of Indian poetics and defines that “rasa” is a thing of relishing or enjoyment of something. Thus Rasa means in poetical viewpoint is a poetical pleasure. Other ways we can say that it is an aesthetic experience of a piece of art. NatyaShastra is an Indian text dated between 2nd century BC and 2nd century AD that analyzes all features of performing art. It is also called the fifth Veda because of its importance [12] Theory of ‘Rasa’ is defined in chapters VI and VII of ‘NatyaShastra.' Bharat Muni explains ‘Rasa’ as “Vibhaavaanubhaav vyabhichaari samyogat ras nishpatti,” The meaning of this verse is, out of the blend (samayoga), of the causes (vibhava), the consequents (anubhava) and the passing mental status (vyabhichari), brings the birth of emotions (rasa). II. RELATED WORK Over the past semi-century, there have been multiple approaches to emotion recognition from the text. There are many emotion recognition types of research based on probabilistic approach. Reference [2], used a simple Natural Language Parser for keyword spotting, phrase length measurement, and emotion identification.[ CHUNG-HSIEN WU et.al] used semantic labels and Separable mixture model to identify emotions.They manually generated the rules for emotion, semantic labels and attitudes. With the help of emotion generation rules, semantic labels and attitudes emotion association rules are automatically derived using priori algorithm. Reference [1] constructed of a large dataset of News headlines are annotated for [7] six basic emotions such as Anger, Disgust, Fear, Joy, Sadness, and Surprise. They proposed LSA and Naïve Bayes Classifier and evaluated several knowledge-based methods for the automatic identification of these emotions in text. Reference [9] constructed a Japanese Emotion Corpus and through the analysis of corpus they have identified the emotions automatically. The advantage is that it can yield high precision. But the main disadvantage is impossible to determine the
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
36
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500
emotion of words which is not in the corpus. Reference[6] has used conditional random field classifier to recognize emotion in a sentence of Bengali blogs. Reference [7] used SVM with tenfold cross-validation to identify emotion in Hierarchical Structure. This method recognizes neutrality, polarity, and emotion of a text hierarchically. According to [8], expert knowledge is mandatory to distinguish emotional keywords for the construction of emotion categories. He used latent semantic analysis to identify emotions. Reference [10] identified emotions in poems using VSM (vector space model) with cosine similarity. The English poems are collected from the web of different poets from worldwide. They have modified traditional VSM with some changes and compared the results. They found that traditional VSM is better than the Modified versions. Reference [11] used a probabilistic approach to identify emotions from the poem. They compared the probabilistic approach with VSM (Vector space model). Again proved that traditional vector space model with cosine similarity give better accuracy than probabilistic approach and other modified VSM methods. Many of the related work mentioned in reference did the labeling of emotions under six [7] or five categories such as (love, sadness, fear, disgust, anger, and surprise). Emotion categorization is different for different domains. Emotions in education field can mainly classify into boredom and interesting. As we have taken poems as our field, we thought to categorize into nine emotions such as (Love, Sadness, Hatred, Anger, Joy, Fear, Courage, Surprise, and Peace) based on Natyshastra [12].
corpus is having 350 poems and 74807 poetic words. The issues in corpus creation are listed below. •
Ambiguity in defining all emotional keywords.
•
Recognizing emotion from sentences with no emotional keywords.
•
Lack of semantic and syntactic information for emotion recognition.
•
Identifying emotion from poems with no emotional keywords.
•
Creative writing such as figurative language, rhyme, poetic form
•
Emotion Intensity.
The corpus has a data size of 350 poems of 8 leading poets of the period 1850-1950. The average words/ poem across the eight poets ranges from 74-284. B. Algorithm 1) Algorithm Search Input: Token t and Corpus C Output: Emotion tags 1. count ← 0 ; // used to keep track of multiple tags 2. For each entity (e ∈ C ) do 2.1.
TABLE I.
‘RASA’ AND THEIR EQUIVALENT ENGLISH EMOTION TYPES NatyaShastra ‘ Rasa’
Emotion
Shringara
Love
Hasya
Joy
Adbhutha
Surprise
Shantha
Peace
Roudra
Anger
Veera
Courage
Karuna
Sad
Bhibatsa
Hate
Bhyanakam
Fear
III.
3
← true ;
2.1.2
t . found
2.1.3
count ← count + 1;
return (t)
2) Algorithm Emotion-Tagging Input: Given poem P, Corpus C Output: Emotion tag et assigned to P 1.
Initialize the token array of P as T ← {t1 , t 2 ,......., t n } ,
2.
n tokens. Initialize the stop word array as S ← {s1 , s 2 ,......., s k } ,
3.
k stop words. Initialize negation word array as N ← n1 , n2 ,......, n j
METHODOLOGY
A. PERC Corpus Details The primary objective of this work is to create a corpus called PERC (Poem Emotion Recognition Corpus).It is classified under nine basic emotions such as Joy, Sad, Love, Hate, Anger, Peace, Courage, Fear, Surprise are described in Navarasa defined in Natyashastra[12] and to recognize emotions from poems. As we are not conscious of any publicly available poetry database, we created our own by mining public websites. The poem database made up poems extracted from the following websites. We collected 350 poems by eight poets. The
if (e . word = t )then t .tag [count ] ← e.tag ; 2.1.1
{
, j negation words. 4. For each ( t ∈ T ) do // Stop word removal from T 4.1 For each (s ∈ S ) do 4.1.1
If (t = s )then
4.1.1.1 5.
T ← T − {t} ;
For each ( t ∈ T ) do t ← search(t )
5.1
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
37
}
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500
If (t. found ← false) then
5.2
5.2.1
For each (n ∈ N ) and do
5.2.1.1
If (n = t ) then
5.2.1.1.1 5.2.1.1.2
t. found ← true ;
q ← search(t i +1 ) // search the
next word emotion in Corpus ti+1 is the next word 5.2.1.1.3 t.tag ← q.tag 5.2.1.1.4 Break; 5.2.2 if (t. found ← false) // the word is not a negation and search in word net 5.2.2.1 t.synonym ← wordnet (t ) 5.2.2.2
q ← search(t.synonym)
5.2.2.3
t.tag ← q.tag
5.2.2.4 6.
insert (C , t )
For each (t.tag [i ]∈ t ) do
⎞ × P(t.tag[i]) P⎛⎜ word ⎟ t tag i . [ ] ⎝ ⎠ t tag i . [ ] ⎤← 6.1 P ⎡ word ⎥⎦ ⎢⎣ P(word )
7. 8. 9.
⎛ ⎤⎞ et ← max⎜ P ⎡t.tag[i] word ⎥⎦ ⎟⎠ ⎝ ⎢⎣ t.emotion ← et.tag[i]
Return (et )
10. For each (t ∈ T ) 10.1 For each ( t .emotion ∈ Emotion ) 10.1.1 t .emotion = e 10.1.2 e.count = e.count + 1
emotion ← max( e .count ) 11. Return (emotion) 10.2
Description of ‘emotion tagging algorithm’ is as follows. This emotion corpus has manually tagged poetic words. Poems of different emotions are collected, and these poems are tokenized and stop words are removed. Stop words are usually the most frequent words, including articles, auxiliary verbs, prepositions, conjunctions, etc. Stop-word list composed of negative verbs such as not, is not, does not, do not, should not, etc. cannot be removed. There are publicly available stop-word lists consisting of most frequent words in a given language. Therefore we created a stop-words list, which does not contribute to emotion recognition The words such as doesn’t, shouldn’t, won’t be converted into does not, should not will not. etc. We have negation word list. Those words are tagged as . We are stemming the tokens as the words in different tense have different emotions. Other tokens are manually tagged using emotion tags such as , , , , , , , , .
Example 1) 2) 3) 4)
I hate chocolates. I hated chocolates. I am a Fearless person I do not love him.
Here the three sentences have different emotions. Sentence (1) tells that still hate chocolates, which means the emotion represented here is hate. However, the second sentence says that once he hated chocolate. So we do not know the emotion at present, which means we have to find out it by contextual information. Therefore, we tagged hate as and hated as . Sentence (3) depicts ‘courage’ only if the word ‘fearless’ is not considered for stemming process. Sentence (4) represents the emotion of hate or anger depends on the context. After the preprocessing stage, the tokens are searched in PER corpus and emotion of each word is identified and tagged using ‘search algorithm.' The emotion tagging of words might result in partial tagging of emotion words. If tokens are not found in the corpus (untagged token), that token is a new word. In that case, that particular word is searched in ‘Wordnet’ to find the synonym of that word. Then that synonym is searched in PERC to assign a tag for that word. That word and its emotion tag are added to the corpus. Sometimes the words can have more than one tag. At this stage, we used ‘Naïve Bayes Learning’ is used for tagging such words. 3) Prior Probability
P (Emotion
i
)=
9
N Emotion i
∑N
Emotion
i =1
(1) i
4) Constructing Unigram Language Model (add one smoothing) P(wordi | Emotioni ) = (Twordi | Emotioni + 1)
n
∑ (T j =1
wordi
| Emotioni + 1) (2)
5) Likelihood probability ( Applying Unigram Language Model) n
P( Poem | Emotioni ) = ∏ ( wordinPoem j | Emotioni ) (3) j =1
6) Posterior Probability P ( Emotion i | Poem ) = P ( Poem | Emotion i ) × P ( Emotion i ) P ( Poem )
(4) 7) Maximum Posterior Probability P ( Emotion i | Poem ) = arg max P ( Poem | Emotion i ) × P ( Emotion i ) i∈(1 .. 9 )
(5) When tags are observed in the tag set of a given poem, bigram model is subjected to handle negation. tagged word and its immediate successor words are observed. If the immediate successor word represents an emotion, then its opposite emotion is swapped Example: Not hate= love
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
38
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500
IV.
RESULT AND DISCUSSION
TABLE II.
CORPUS DETAILS
S.No
Poet
No. of poems
No of lines
No of words
No. of Content words
1
Rabindranath Tagore
186
5556
46099
4501
2
Sarojini Naidu
48
2538
13617
1858
3
Aurobindo
40
817
4809
1423
4
JidduKrishnamurthi
15
498
2496
885
5
Ananda Murthy
14
221
1642
711
6
Lalan
15
330
1116
471
7
Darshan Singh
15
432
2631
1092
8
Nazrul-Islam
17
527
2397
999
Total
350
TABLE III.
STATISTICS DETAILS OF CORPUS
Poet
No of Words
Unique word of Poet
more than once used unique word
Unused words
Average words/ poem
Average of unique word/poem
No. of Named Entities
Rabindranath Tagore
46099
2542
1027
0
249
14
6
Sarojini Naidu
13617
644
114
0
284
13
3
Aurobindo
4809
427
62
0
120
11
1
JidduKrishnamurthi
2496
183
8
6
166
12
1
Ananda Murthy
1642
102
10
5
117
7
1
Lalan
1116
78
6
8
74
5
0
Darshan Singh
2631
243
15
1
175
16
0
Nazrul-Islam
2397
181
16
0
141
11
1
TABLE IV.
RANK OF MAXIMUM USED(WORD-WISE) BY EACH POET
Rank
Rabindranath Tagore
Aurobindo
Sarojini Naidu
Ananda Murthy
Dharshan Singh
Lalan
Jiddu Krishnamurthi
Nazrul- Islam
I
when
Heart
Love
move
eyes
Moon
come
Mother
II
come
Who
Heart
love
heart
Black
clear
Love
III
heart
Life
Life
path
beloved
Flashing
jungle
Come
IV
Like
Light
Song
closer
light
Beauty
liberation
Child
V
Love
Mind
Dream
garland
friend
Eyes
love
Like
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
39
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500 TABLE V.
RANK OF MAXIMUM OCCURRED WORDS IN EACH EMOTION
Rank
Anger
Courage
Fear
Hate
Joy
Love
Peace
Sad
Surprise
I
Anger
Dreams
running
Sorry
laughter
Love
Peace
grief
surprise
II
Love
Believe
stay
Hate
Day
Heart
world
son
morning
III
Pain
Love
fear
Love
Box
loving
human
house
ferocity
IV
Day
Courage
eyes
Thou
Idiot
honey
violence
brought
Rail
V
Inside
Yourself
sorry
Hear
Love
steady
energy
words
Fields
TABLE VI.
DETAILS OF POEM CATEGORIES
Emotions Anger Courage Fear Hate Joy Love Peace Sad Surprise
TABLE VII.
Emotion
Actual
No. of poems 21 15 17 15 55 89 51 70 17 350
EMOTION RECOGNITION RESULTS BY NAÏVE BAYES CLASSIFIER
Correctly Identified Poems
Anger
courage
fear
Hate
Joy
love
peace
Sad
Surprise
Wrongly Identified Poems
Anger
21
15
15
4
0
1
0
0
0
1
0
Courage
15
9
1
9
0
1
0
0
2
2
0
Fear
17
10
3
0
10
0
0
2
0
2
0
Hate
15
10
3
1
0
10
0
0
0
1
0
Joy
55
48
0
2
1
0
48
3
1
0
0
Love
89
73
1
5
0
0
8
73
2
0
0
Peace
51
40
0
3
0
0
3
4
40
1
0
Sad
70
54
4
0
1
0
1
10
0
54
0
Surprise
17
10
0
2
0
0
3
2
0
0
10
350
269
Table II shows the details of corpus such as poet’s name, the number of poems by each poet, the total number of lines, the total number of words and content words. Content words are words which give emotion after removing the stop-words. Observe that the number of content words contributes towards emotion recognition rather than an overall number of words or lines in the poem. In Table III, we analyze the author statistics. The primary attribute of author statistics includes unique words that are words used only by that particular poet which are recorded. For
instance, Tagore used unique words than others as his collection of poems are high. But notice that the average unique word per poem is greater for Darshan Singh. Unused words are nothing but the words that are not used by the particular poet but commonly used by others. Finally, the last attribute represents named entities (NE). Most of the NE includes God’s names such as Krishna, Radha, etc. Table IV gives the details of words specifically used many times by each poet summarized for all poets. For example, the
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
40
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500
Table VII gives the details of emotion recognition result by ten cross folding naïve Bayes classification method. Table column head ‘Actual’ and ‘correctly’ gives the particulars of the correctly identified poem in a total number of poems in each emotion categories. Other column heads say the number of poems misclassified into other categories. The misclassifications of poems are due to ambiguity. Accuracy Measure For Naive Bayes Classifier 1
Precision Comparison Precision measure
word ‘love’ is used by five poets. Four poets use the word 'heart'. The word ‘come’ is used by three poets and the word ‘like’ is used by two poets. Table V gives the information of most occurred used in each emotion, and Table VI depicts the details of the number of poems occur in the corpus.
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
Naïve
corpus
VSM
VSM-I
VSM-II
Anger
0.56
0.37
0.42
0.35
0.42
Courage
0.35
0.22
0.3
0.25
0.26
Fear
0.71
0.3
0.37
0.29
0.3
Hate
0.83
0.45
0.53
0.45
0.45
Joy
0.56
0.64
0.74
0.72
0.77
Love
0.78
0.65
0.76
0.74
0.78
Peace
0.89
0.77
0.87
0.8
0.91
Sad
0.89
0.69
0.76
0.74
0.77
1
0.83
1
1
1
Surprise
0.9
measure
0.8
Methods
0.7 0.6
Anger Hate Peace
0.5
Courage Joy Sad
Fear Love Surprise
0.4 0.3
Fear
Hate
Joy
Love
Peace
Sad
precision
Anger Courage 0.56
0.35
0.71
0.83
0.76
0.78
0.89
0.89
Surprise 1
recall
0.71
0.6
0.59
0.67
0.87
0.82
0.78
0.77
0.59
accuracy
0.95
0.93
0.97
0.98
0.94
0.89
0.95
0.93
0.98
Figure2. Comparison of Precision measures
Emotions recall
Recall Comparison accuracy
Figure1. Precision, Recall and Accuracy measure
Figure 1 depicts the precision-recall and accuracy measure of emotion recognition by Naïve Bayes Classifier. The accuracy of emotion recognition by Naïve Bayes classifier has more than 90% for all the emotion categories. Given to understand that to the best of our knowledge, we do not know of any other work in emotion recognition in poems. Hence our comparisons are of our previous works [10][11]. Emotion recognition by Naïve Bayes Classifier is compared with Probabilistic Corpus-based method [11] and Vector Space Model [10]. In the reference [10] we have changed few steps and modified into two versions as VSM-I and VSM-II. In VSM-I approach, we eliminated the stop-word removal and stemming process of each word in poems. In VSM-II approach, we created our own stop-word list that is a list of emotionless words (words does not contribute to emotion recognition). Usual components like negation words (not, never.., etc.), interrogative words (who, when.., etc.) are eliminated from the emotionless word list, as it contributes to emotion recognition process.
0.9
Recall Measure
precision
0.8 0.7 0.6 0.5 0.4 0.3
Naïve
corpus
VSM
VSM-I
VSM-II
Anger
0.71
0.54
0.63
0.58
0.56
Courage
0.6
0.43
0.53
0.45
0.5
Fear
0.59
0.43
0.53
0.45
0.48
Hate
0.67
0.45
0.56
0.48
0.5
Joy
0.87
0.66
0.76
0.71
0.86
Love
0.82
0.67
0.75
0.72
0.77
Peace
0.78
0.61
0.71
0.66
0.71
Sad
0.77
0.61
0.7
0.66
0.68
Surprise
0.59
0.43
0.5
0.48
0.5
Methods Anger Hate Peace
Courage Joy Sad
Fear Love Surprise
Figure3. Comparison of Recall measures
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
41
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500
except probabilistic corpus-based approach. In this approach ‘Love’ Classification gets a higher value than all approaches. Naïve Bayes classifier gets better recall value for all emotion categories.
Accuracy comparison
Accuracy Measure
Naïve 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
corpus
VSM
VSM-I
VSM-II
Figure 3 and Figure 4 depicts the accuracy and F-measure of all approaches. Naïve Bayes classifier yields greater accuracy and F-measure while comparing with other methods. V. Anger
Coura ge
Fear
Hate
Joy
Love
Peace
Sad
Surpri se
Naïve
0.95
0.93
0.97
0.98
0.94
0.89
0.95
0.93
0.98
corpus
0.89
0.87
0.9
0.93
0.85
0.79
0.89
0.83
0.96
VSM
0.91
0.92
0.93
0.95
0.91
0.87
0.94
0.89
0.97
VSM-I
0.89
0.89
0.9
0.93
0.89
0.84
0.91
0.87
0.97
VSM-II
0.91
0.9
0.9
0.94
0.94
0.88
0.94
0.88
0.97
CONCLUSION AND FUTURE WORK
We have created a corpus PERC based on ‘Navarasa’ described in ‘Natyashastra’. Human experts created the Corpus. The corpus is consisting poems of eight Indian Poets at the period of 1850-1950.We are getting good accuracy measure. We compared with existing works and found that naïve Bayes classifier yields better accuracy measure. Some poems misclassified due to ambiguity. Two ways can resolve it. One is increasing the corpus size. Secondly considering the other features like semantic features and poetic features.
Emotions
Figure4. Comparison of Accuracy measure
REFERENCES
F-measure Comparison
[1]
F-measure
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 Anger
Courag e
Fear
Hate
Joy
Love
Peace
Sad
Surpris e
Naïve
0.63
0.44
0.65
0.74
corpus
0.43
0.29
0.36
0.45
0.81
0.8
0.83
0.82
0.74
0.65
0.66
0.68
0.65
VSM
0.5
0.38
0.43
0.57
0.54
0.75
0.76
0.78
0.73
VSM-I
0.51
0.35
0.67
0.46
0.56
0.72
0.72
0.75
0.72
VSM-II
0.47
0.33
0.65
0.39
0.49
0.7
0.71
0.73
0.69
0.62
Emotions Naïve
corpus
VSM
Figure5. Comparison of F-measure
Figure 2 gives the details of precision measures among all the methods. Other than probabilistic corpus-based method, all other methods obtained a precision value as ‘1’. In the corpusbased method, it checked for the exact word matching and classified. Figure 3 shows the recall measures of all methods. ‘Joy’ classification obtains higher recall value in all the approaches
C. Strapparava and R. Mihalcea, “Learning to identify emotions in text,” In Proceedings of the 2008 ACM Symposium on Applied Computing, pp. 1556-1560, March 2008. [2] C. O. Alm, D.Roth and R. Sproat, “Emotions from text: machine learning for text-based emotion prediction,” In Proceedings of the conference on human language technology and empirical methods in natural language processing. Association for Computational Linguistics, pp. 579-586, October 2005. [3] C. D. Manning and H. Schütze, Foundations of Statistical Natural Language Processing, vol. 999.MIT Press: Cambridge, 1999. [4] C. H. Wu, Z. J. Chuang and Y. C. Lin, “Emotion recognition from text using semantic labels and separable mixture models,” ACM Transactions on Asian Language information Processing (TALIP), vol.5(2), pp.165183, 2006 [5] D. Ghazi, D. Inkpen and S. Szpakowicz, “Hierarchical versus flat classification of emotions in text,” In Proceedings of the NAACL HLT 2010 workshop on computational approaches to analysis and generation of emotion in text, Association for Computational Linguistics. Association for Computational Linguistics, pp. 140-146, June 2010. [6] D. Das and S. Bandyopadhyay, “Word to sentence level emotion tagging for Bengali blogs,” In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, Association for Computational Linguistics, pp. 149-152 20, August 2009. [7] P. Ekman, “An argument for basic emotions. Cognition & emotion,” vol.6(3-4), pp.169-200,1992. [8] J. R. Bellegarda, “Emotion analysis using latent affective folding and embedding,” In Proceedings of the NAACL HLT 2010 workshop on computational approaches to analysis and generation of emotion in text, Association for Computational Linguistics, pp. 1-9, June 2010. [9] J. Minato, D. B. Bracewell, F. Ren and S. Kuroiwa, S, “Japanese emotion corpus analysis and its use for automatic emotion word identification,” Engineering Letters, vol.16(1), pp.172-177, 2008. [10] P. S. Sreeja and G. S. Mahalakshmi, “Applying vector space model for poetic emotion recognition,” Advances in Natural and Applied Sciences, vol. 9(6 SE), pp.486-491, 2015. [11] P.S. Sreeja and G. S. Mahalakshmi, “Comparison of Probabilistic CorpusBased Method and Vector Space Model for Emotion Recognition from Poems,” Asian Journal of Information Technology, vol.15(5), pp.908915, 2016.
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
42
Vol. 14 CIC 2016 Special Issue International Journal of Computer Science and Information Security (IJCSIS) https://sites.google.com/site/ijcsis/ ISSN 1947-5500 [12] M. Ghosh, The Natya Shastra (English Translation), vol.1 (Chapters IXXVII). The Royal Asiatic Society of Bengal: Calcutta, 1950.
AUTHORS PROFILE
P. S. Sreeja is the corresponding author of this paper. She is currently pursuing research in the Department of Computer Science and Engineering at Anna University, Chennai. She Completed her MCA from Bharathidasan University, Trichy, India, and M.Phil. (Computer Science) From the University of Madras, Chennai, India. She is a First rank holder in M.Phil. Her research interests are Artificial Intelligence, Text Mining, and Natural Language Processing.
G. S. Mahalakshmi is an Assistant Professor (Senior Grade) in the Department of Computer Science and Engineering, College of Engineering, Anna University, Chennai. She completed her B.E. (Computer Science and Engineering) from R.V.S. College of Engineering and Technology, Dindigul and M.E. (Computer Science and Engineering) and Ph.D. from College of Engineering, Anna University, Chennai. She has numerous international journal and conference publications to her credit. She is also the author of Tamil Edition of B.E. course – text books - Fundamentals of Computing and Computer Practice of Anna University. She has authored many book chapters and derived 100+ citations to her credit. Her research interests include Reasoning, Knowledge Sharing and representation, Text Mining, Social Network Analysis, bibliometrics, and Natural Language Computing.
International Conference on Advances in Computational Intelligence and Communication (CIC 2016) Pondicherry Engineering College, Puducherry, India October 19 & 20 - 2016
43