Paper
Recognition of Word Emotion State in Sentences
Changqin Quan
∗,∗∗
Member
Fuji Ren
∗∗
Member
Emotional word spotting has been well used as a basic step in the task of textual emotion recognition
and automatic emotion lexicon construction. To express and recognize emotion of words, especially for the
words bear undirect emotions, emotion ambiguity, or multiple emotions, the notion of “word emotion state”
is proposed, which describes the state of combined basic emotions in a word. Based on Ren-CECps (an an-
notated emotion corpus) and MaxEnt (Maximum entropy) modeling, we explore the effectiveness of several
features and their combinations for word emotion recognition. A compare study on the performances of word
emotion and word emotion state recognition is given. The experimental results showed that a model using
word emotion state can greatly outperform using word emotion.
Keywords: Emotion recognition, word emotion state, Ren-CECps, MaxEnt
1. Introduction
Researchers have found that emotion technology can
be an important component in artificial intelligence
(1)
.
Textual emotion sensing is becoming increasingly im-
portant due to augmented communication via computer
mediated communication (CMC). Possible applications
of textual emotion recognition include online chat sys-
tem. An emotional feedback system can recognize users’
emotion and give appropriate responses. In addition,
Weblog emotion recognition and prediction are valu-
able applications. Blogspace consists of millions of users
who maintain their online diaries, containing frequently-
updated views and personal remarks about a range of
issues. An emotion recognition and prediction system
can understand the public’s reaction to some social is-
sues and predict emotion change. This would be helpful
for solving some psychological problems or giving early
warnings, such as suicide or terrorism.
Textual emotion analysis also can improve the accu-
racy of other nonverbal modalities like speech or facial
emotion recognition, and to improve human computer
interaction systems. However, automatic recognition of
emotional meaning from texts presents a great challenge
because of the manifoldness of expressed meanings in
words. Word emotion analysis is fundamental for tex-
tual emotion analysis.
The approach of emotional word spotting has been
well used as a basic step in the task of textual emo-
tion recognition and automatic emotion lexicon con-
struction. And there are many lexical resources devel-
oped for these tasks, such as GI
(2)
, WordNet-Affect
(3)
,
NTU Sentiment Dictionary
(4)
, Hownet
(5)
, SentiWord-
net
(6)
. In these sentimental or affective lexicons, the
∗
Department of Computer Science, Huazhong Normal Uni-
versity, Wuhan, China, 430079
∗∗
Faculty of Engineering, University of Tokushima,
2-1 Minamijosanjima, Tokushima, 770-8506
words usually bear direct emotions or opinions, such as
happy or sad, good or bad. Although they play a role in
some applications, several problems of emotion expres-
sion in word have been ignored.
Firstly, there are a lot of sentences can evoke emotions
without direct emotional words. For example,
(1) S U 3 ¯ f ú p ! 3 ¯ f %
p"(English: Spring is in children’s eyes, and in their
hearts.)
In sentence (1), we may feel joy, love or expect deliv-
ered by the writer. But there are no direct emotional
words can be found from lexicons. As Ortony indicates,
besides words directly referring to emotional states (e.g.,
“fear”, “cheerful”) and for which an appropriate lexicon
would help, there are words that act only as an indi-
rect reference to emotions depending on the context
(7)
.
Strapparava also addresses this issue
(8)
. The authors
believed that all words can potentially convey affective
meaning, and they distinguished between words directly
referring to emotional states (direct affective words) and
those having only an indirect reference that depends on
the context (indirect affective words).
The second problem is emotion ambiguity of words.
The same word in different contexts may reflect differ-
ent emotions. For example,
(2) ù´8c· U"(English: This is cur-
rently the only thing I can do.)
(3) ¦´·"(English: He is my only one.)
In sentence (2), the word “(English: only)” may
express the emotion of anxiety or expect; while in sen-
tence (3), the word “(English: only)” may express
the emotion of love or expect. The emotion categories
can not be determined without their certain contexts
especially for the words with emotion ambiguity.
In addition, some words can express multiple emo-
tions, such as “U\(English: mingled feelings of joy
and sorrow)”. Statistics on an annotated emotion cor-
IEEJ Trans. XX, Vol.xxx, No.xx, xxxx 1