Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of NursingAnswers.net.
Modelling artificial emotional agent based on combination of classification and fuzzy logic methods
Abstract. Recently, researchers have tried to better understanding human behaviors and interaction to let artificial intelligent agents act like the human. It means a computational agent has its own emotions, attitude and interaction method and let humans interact with computers in novel ways. To achieve this goal, in this study we tried to model and simulate an artificial emotional agent by using classification methods and fuzzy logic. First of all, we defined the six basic Paul Ekman Emotions then in order to classify the sentences, we have tried multi-classification Support vector machine algorithm with polynomial, radial basis function, sigmoid kernels and Naive Bayes classifier to find out the optimized kernel and classifier. The classified sentence is sent to the fuzzy logic system to expand the emotion base on Geneva emotional wheel includes sixteen distinct emotions, each of which we may consider a basic emotion. Besides, in order to obtain the lingual variable of the output emotions, valence and control values, the fuzzy system designed and implemented to achieve the outputs based on psychological theories of human attitudes and emotions. In this study we succeed to simulate humanoid emotions based on sentences and classify them based on the psychological emotional models with 80% accuracy via applying SVM with RBF kernel and get the output emotions in lingual variables by using mamdani fuzzy logic system.
Keywords: Affective computing- Human computer interaction- Emotion recognition
Human in their everyday life make decisions and react often based on reason and emotion. Emotions are playing a key role in human life; they are affected and influenced on how we think, how we make decisions, learning, intentions, goals, motivation, behavior and how we communicate with the other people and creatures. It is clear that without emotions, our life has not enough parameters to be pleasant and people have not sufficient reasons and motivations to carry on. Still, the question is whether intelligent machines with emotions can perform better than pure logical machines or not? And how human and computer can interact like human-human interaction.
Some neurological researchers demonstrate that emotions have fundamental and effective role in human decision making in almost all aspects. Meanwhile, affective computing is the study and development of systems and devices that can recognize, interpret, process, and simulate human affects. It shows signs of future success in emotional computing fields which spans computer science, psychology, and cognitive science. Recognition human’s emotional states can be used in various processes, which is highly relevant and affective in logical processes. Emotion recognition can be implemented in various kinds of media. For instance, Picard proposed the concepts and theories on image and speech recognition . Cowie did research to emotion recognition by signals including emotional keywords, speech and facial emotions . In addition, the other major approaches proposed as the follow: Several pattern recognition methods use front-end signal recognition approaches rather than choosing discourse of information, In the perfect universe of science, emotion is measured in five ways including the change processing of central nerves, the response mood of the nervous systems, the appraisal results of motivational changes and particular action tendencies, movements and facial expressions, and the subjectively experienced feeling state levels .
In this study in order to simulate humanoid emotions, we propose an artificial emotional agent which is capable of recognizing, classifying, expanding and extracting explicit and implicit emotions of corpus in different levels of intensity and control with appropriate linguistic variables. Firs, the dataset is collected from news headlines and social networks and six basic emotions of Paul Ekman emotional model including: Happiness, sadness, fear, anger, surprise and disgust defined for each sentences. In order to classify sentences in appropriate classes that will be coordinated on Geneva emotion wheel, SVM multi classifier with different kernels and Naïve Bayes algorithm are implemented and result are compered. Finally, the Geneva emotion wheel is simulated with mamdani fuzzy logic system. So we have 16 different emotions according to Geneva emotion wheel with appropriate linguistic variable.
In the related researches different methods are used to achieve emotions behind corpus, but all of them only implemented one psychological model and achieved related results. In this study we implemented two different psychological models and two different but depended methods.
The rest of this paper is organized as follow. Section 1 discusses overview of fundamentals and definitions of emotion and affective computing subjects. Section 2 describes the related works of this study. Section 3 introduces the proposed approach. Simulation and results in section 4 and section 5 concludes the study and outlines for future studies.
2. Related works
In recent years, researchers and scientists are getting more interested in intelligent systems to solve problems and be more innovative. Intelligent agents are divide into two type.
a) Logical Intelligent Agents (LIA) b) Emotional Intelligent Agents (EIA). A complete and high performance agent is an agent that uses both types of intelligent agents. However, to achieve the complete agent we need to design a perfect logical and emotional agent. 
The major importance of EIA is covering different kinds of human emotions. Scientists have studied the humans to recognize, define and formulate the emotion states and feelings. In 2010 Calvo studied different methods and proposed emotions as system input regarding to individual modalities or channels such as face, voice and text. In this model different visual information such as moods, colors, objects and everything that is in image and can be extracted are assumed as input. In addition, sound signals are captured via microphone and converted to appropriate signals to be processed and extracting information. Finally, text proposed as the major input factor that can be extracted from visual input data, converting received sound signals to equivalent text and corpus. 
In this section, the literatures and related works based on recognition and extracting emotions through text and corpus according to type and methodology of simulation that were proposed by researchers and scientists are studied.
2.1. Symbolic approach
This method is a rule-based system that uses rules to generate emotion states. In 2006, Abu Maria and Abu Zitar  proposed a conditional and appraisal based system with a rule-based structure. The agent interacts continuously with environment and an internal ‘‘thinking’’ that comes as a result of series of inferences, evolution processes, adaptation, learning, and emotions. They built two models for agent based systems; a) a regular artificial agent b) an emotional artificial agent.
They used both agents to solve a benchmark problem, ‘‘The Orphanage Care Problem’’. The agents are simulated and results compared.
2.2. Neural networks approaches
Artificial neural networks are systems that can be trained to perform particular functions. ANNs with their capacities on infer importance and meaning from imprecise or complex data can be effective to extricate patterns or recognize trends by humans or computational strategies.
In 2014 Akbarzadeh  proposed an inspired model from the limbic system to simulate human emotions. The limbic system is a set of brain structures located on both sides of the thalamus, under the cerebrum. The system underpins an assortment of capacities including feeling, behavior, inspiration, long-term memory, and olfaction.  The limbic system is the host of emotional aspect life of humans and playing a major role with the formation of memories.
The proposed system is able to recognize simple patterns and have long-term memory, olfaction, Inheritance. However, the system cannot recognize and learn advance problems and not determine emotional states and attention process.
Machine learning algorithms are used to recognize emotions in order to learn how to interpret emotion states. Supervised and unsupervised machine learning strategies have been applied to recognize emotions in text. Supervised techniques have the obstacle that large datasets are required for training. Unsupervised techniques are normally preferred in the scope of natural language processing and emotions.
The learning strategies are able to process text and interpret, recognize and state emotions. In 2008 Strapparava and Mihalcea  proposed a supervised (Naïve Bayes) and four unsupervised methods to recognize the six basic emotions. They claimed that each system has its weak points and strengths. Naïve Bayes was the most precise only for Joy emotion. Using the WordNet Affect lexicon had the highest accuracy but a low recall. LSA using all the emotion words had the greatest recall but the accuracy is less.
In 2012, Rafael and colleagues  claimed their approach is evaluated utilizing four data sets of texts reflecting diverse emotional states. An emotional thesaurus and a bag-of-words model are utilized in order to generate vectors for every pseudo-document, then for the categorical models three dimensionality-lessening methods are assessed including Latent Semantic Analysis (LSA), Probabilistic Latent Semantic Analysis (PLSA) and Non-Negative Matrix Factorization (NMF). For the dimensional model, a standard database is utilized to create three-dimensional vectors (valence, arousal, dominance) for every pseudo-report. This three-dimensional model can be applied to produce mentally determined representations. Both models can be used for affect detection based on distances among classes and pseudo-reports. Experiments demonstrate that the categorical model utilizing NMF and the dimensional model have a tendency to perform best.
2.3. Fuzzy logic approach
In 2012 William Frederick Blewitt  studied psychology and cognitive behavior in human and tried to formulate and simulate in computers. He selected two emotional models:
b) Scherer Geneva emotion wheel
He used fuzzy logic adoptive model to simulate the emotional models. Fuzzy logic types-1 and type-2 implemented for Geneva emotional wheel and Millenson’s model of emotion implemented only with fuzzy logic type-1. The models are different in structure, emotions type and number of emotion outputs. Blewitt proposed structure of his model based on fuzzy logic and determined the fuzzy rules and succeeded to simulate the models. He proved that it is possible to implement human psychological structures and emotional models through fuzzy logic. The outputs are numbers and values for each emotion for the models. One of major disadvantage of his work is input data type. He used one X and Y point as input emotion which means limitation in emotion expression because humans have different emotions with different level at the same time.
3. The Proposed Approach
Affective computing is the study of human-computer interaction. In other words, development of systems, which is able to recognize, interpret, process, and simulate human, affects. Affective computing is an interdisciplinary research field that gathers together researchers of different fields, from computer science and artificial intelligence to psychology, cognitive and social sciences. Facial expressions, the way of looking, figures, gestures, words, tone of speech, the force or power of key strokes and the temperature changes of the hand on a mouse can all signify changes in emotional state and detectable by computers. The machine should interpret the feeling and emotional status of individuals and adjust its conduct to them, giving a suitable response for those feelings. 
Emotion is any relatively brief conscious experience described by intense mental action and a high level of pleasure or displeasure.  In Psychology, Emotion frequently is characterized as a perplexing condition of feeling that outcomes in physical and psychological changes that affects thought and conduct. From a philosophical point of view, the nature of their dissimilarity and their hypothetical differences are significant. From a computing sciences point of view, their differences lie altogether in the way of the proposed models.  Emotion is regularly intertwined with mood, temperament, personality, disposition, and motivation. 
The distinctive emotional states classified in three major groups. 
a) Emotions (e.g., fear, anger, sadness, joy, disgust, trust, anticipation, surprise)
b) Moods (e.g., dreamy, cheerful, depressed, stressed, gloomy)
c) Attitudes (e.g., loving, hating, liking, desiring)
Recently, research on emotion is expanded fundamentally in many fields including psychology, neuroscience, endocrinology, medicine, history, sociology, and computer science. Many theories attempt to clarify the source; neurobiology, experience and function of emotions have developed more research on this topic. “Emotions can be defined as a positive or negative experience with different level of intensity which are associated with a specific pattern of physiological activity.” The major role of emotions is motivating adaptive behaviors and responses to internal and external events.  In order to study emotions, it is necessary to formulate emotions thus emotion modeling allows us to consider emotional state as an input or output. In particular, attention is directed towards two concepts that are used in this study. The Ekaman six basic emotion proposed by Paul Ekman  and Geneva emotion wheel proposed by Scherer .
3.1.1. Ekman basic emotion
The six basic emotions are terms that alludes to the hypothesis of American analysts Paul Ekman . Ekman defined six basic emotions based on studying the culture of humans from the Fori tribe in Papua New Guinea in 1972. The tribe individuals could distinguish six emotions on the photos. From that point onward, they took pictures of facial expressions of individuals from the Fori tribe with similar feelings and they displayed these photos to individuals of different races and societies everywhere throughout the world. They likewise translated the emotions on the photos correctly. Following six basic emotions were identified: (1) Anger, (2) Disgust, (3) Fear, (4) Happiness, (5) Sadness, (6) Surprise. Many scientists have affirmed that these emotions are global for all people of the world. 
3.1.2. The Geneva emotion wheels
Scherer studied connection between particular emotions, and relative encounters of valence and control. Through observational examination, educated by broad experiment, Scherer hypothesizes that a structure highlighting sixteen fundamental emotions may be created, with each emotions position and force being dictated by a vector relationship characterized by these two input components . Remarking on Russell’s unique circumflex work , Scherer took a portion of the conclusions drawn and utilized them to tune his model. He additionally made note of comparable results achieved through discrete empirical experiment. The Geneva Emotion Wheel is introduced in the shape appeared in Fig. 1.
Fig. 1. Graphical depiction of the Geneva Emotion Wheel 
The Geneva emotion wheel includes sixteen distinct emotions each of which is considered a basic emotion. Following sixteen basic emotions are identified as following: (1) pride, (2) elation, (3) happiness, (4) satisfaction, (5) relief, (6) hope, (7) interest, (8) surprise, (9) anxiety, (10) sadness, (11) boredom, (12) shame/guilt, (13) disgust, (14) contempt, (15) hostility, (16) anger. 
The wheel represents variable degree of intensity as relative magnitudes of valance and control. The wheel associates an input based on agent perception of the valence and control it feels in a given circumstance. From this, it characterizes an emotional response related to the event, represented and indicated by sixteen basic emotions which can be utilized to characterize adaptive emotional states. 
3.2. Emotional intelligence
Emotional Intelligence (El) officially proposed by Salovey and Mayer in 1990s . However, Danile Goleman defined emotional intelligence in his published book “Emotional intelligence” in 1995 as understanding one’s own feeling, empathy for feelings of others and regulation of emotion in a way that enhance living. Moreover, he claimed that EI might be more important than IQ for success and EI can be improved Unlike IQ.
Concerning ability, EI defined as following. a) Recognize, understand and manage our own emotions. b) Recognize, understand and influence the emotions of others.
EI defined as a portion of human intelligence authentic for the capability to:
- Perceive emotions;
- Unify emotions to simplify thoughts
- Realize emotions
- Regulate emotions
They discussed that recognizing and distinguishing emotions are the first basic phase to achieve full area of emotional intelligence.
The proposed emotional agent as shown in Fig.2 consists of four components, including a dataset, machine learning algorithms, fuzzy logic and output.
Fig. 2. The Proposed Model
Collected database includes sentences and corresponding values of the six basic emotions for each sentence. The dataset is the only input of the system. The output of the system is based on The Geneva emotion wheel and we feed in wheel’s input by a point according to X, Y-axis. Address the sentence’s emotion according to the axis; we only need the prevailing emotion that will be get from the output of the machine learning section. So, in order to classify the dataset in the relevant labels according to the six basic emotions of Ekman based on the prevailing emotion, we tried Support Vector Machine (SVM) and Naive bayes (NB) classifier. For as much as these algorithms intrinsically are binary algorithms, we used some methods and library such as LIBSVM for multi classification classifying. Then, the point is addressed to the most relevant emotion on the Geneva emotion wheel. The wheel implemented by fuzzy logic toolbox. The output of agent contains all the parameters.
The subject of dataset is collected from different news headlines, twitter, comments and other social networks. The major parameter for sentence selection is high emotional theme. 1080 instances are collected and answered by female people. For instance, the sentence” Army photographer killed in Afghanistan” is valued in a range of 0 to 100 for the six basic emotions by 100 females via online questionnaire. The average of all instances of the emotion values are calculated as the emotion valence of an instance for each emotion.
Table 1. Average of the responses for the collected dataset
The value range of valence is from zero to hundred.
The average of each emotion from hundred responses is calculated as the emotion of the sentence and the highest value of the emotions assumed as the prevailing emotion. In order to get the balanced quantity of each emotion as the prevailing emotion for training and test data, the number of each prevailing emotional sentences are according to Table 2.
Table 2. Sentence numbers by the emotion category
|Emotion||Trainings Number||Test number|
As it is clear in Table 2, 80% of instances are used for training data and 20% for test data.
3.4. Machine learning algorithms
Machine learning is the review and development of calculations that can gain from and make expectations on data  – such calculations conquer taking after entirely static program directions by making information for predictions or decisions through building a model from test inputs. Machine learning is utilized in a scope of figuring assignments where outlining and programming unequivocal calculations with great execution is troublesome or infeasible . In this section, we used machine-learning algorithm to categorize and classify our dataset in the prevailing emotion class.
In order to classify our sentences in the dataset into the suitable classes and the labels of the classes are six basic emotions. We implemented two different algorithms with various kernels to assess the performance of the SVM with different kernels and Naïve Bayes to classify with the efficient algorithm and kernel.
3.4.1. Multi-class support vector machines
Support vector machines (SVM) are supervised learning algorithm with associated learning algorithms to analyze data for classification and regression analysis that originally designed for binary classification. How to effectively broaden it for multi-classification is as yet an on-going exploration issue. A few methods have been proposed where ordinary we construct a multi-class classifier by using several binary classifiers. Some researchers also proposed methods that consider all classes at once. Experiments show that the “one-against-one” and DAG methods are more proper for practical use in comparison with other methods. Results demonstrate that for the most substantial issues fewer support vector machines is needed. 
It this study, we use LIBSVM, (SVC: support vector classification (two-class and multi-class)) , a multi-classification library that proposed by Chang and Lin in 2000 still are working on developing the package. LIBSVM applies the “one-against-one” method) for multiclass classification problems.
We implemented different kernels including Polynomial, Radial Basis Function (RBF) and sigmoid to assess the performance of the kernels and achieve the most appropriate result.
The SVM polynomial kernel represents the likeness of vectors (training samples) in a feature space over polynomials of the original variables, permitting learning of non-linear models. The polynomial kernel looks not only at the given features of input samples to determine their similarity, but also blends of these. Concerning regression analysis, such combinations are known as interactive features.
The SVM radial basis function (RBF) kernel, or Gaussian kernel, is a kernel that is in the form of a radial basis function (Gaussian function).
The hyperbolic tangent kernel is also known as the sigmoid kernel and as the Multilayer Perceptron (MLP) kernel. The sigmoid kernel originates from the Neural Networks field, where the bipolar sigmoid function is regularly utilized as an activation function for artificial neurons.
Naive Bayes (NB) classifier is a simple probabilistic classifier in view of applying Bayes hypothesis with strong independency between the features. Assuming an underlying probabilistic model and it allows us to capture uncertainty about the model in a principled way by determining probabilities of the outcomes. It can solve diagnostic and predictive problems. In order to using NB multi classification algorithm, the Naïve Bayes built-in function of Matlab is used for training and test the dataset contents to be classified in the appropriate classes.
3.5. Fuzzy logic
The implementation of Geneva emotion wheel model is complex and need proper tools. The model geometric analyzes and capabilities of Matlab Fuzzy Logic Toolbox should be considered precisely. We used Mamdani inferencing system with the minimum AND operator and centroid defuzzification method in Matlab Fuzzy toolbox. The fuzzy system receives an input vector of J (Figure 3).
Fig. 3. Fuzzy Inferencing System of Geneva emotion wheel
Where x and y are output of the classified sentences by the previous step and used as input of our Fuzzy system and indicate the value of valence and control in range of -1 to 1. It is determined that each the input values have thirty-three fuzzy functions and their associated rules .
The fuzzy logic is used to simulate Geneva emotion wheel to obtain the 16 emotions of the wheel, valance and control parameters. The input variables indicated with XJ and YJ for a given status defined as conceptual sums of experiential perceptions.
Where uiJ is an element of the agent’s environment that impacts on valence of that there are nu for any given xJ. viJ is an element of agent’s environment that impacts on control of that there are nv for any given y. the individual impact is defined by the association function of Fvi; and the sums are normalized between -1 to 1.
Regarding the sixteen output emotions, each output is defined as a vector that indicates the relative magnitude of each sixteen emotions. These emotional states Es are according to table 3.
Table 3.Emotional states Es
In order to implement the wheel, some constrains are assumed for the fuzzy system as follows:
A) The fuzzy functions are trapezoidal.
B) The maxima of the trapezia along either axis coincided with the diameter of each discrete geometrical region indicated by the wheel prototype.
C) Each shoulder of trapezium is equal in width along the axis to width of the maximum.
D) The absolute value of -1 and 1 along each axis is indicated by the edge of the maximum of the first and thirty second membership function for that axis. These limitations are applied and tables 2 shows the vertices of each trapezoidal membership functions of x and y variables according to the value of their first coordinate. 
In our implementation, the nature of the fuzzy functions faced applied constraints. First of all , the fuzzy functions would be trapezoidal and that the maxima of these trapezia would, along either axis, coincide with the diameter of each discrete geometrical region determined by Scherer’s prototype. Then, each shoulder of a trapezium would be equal in width along its axis to the width of its maximum. Finally, that the absolute values of -1 and 1 along each axis would be determined by the edge of the maximum of the first and thirty-second membership functions for that axis (mindful that the thirty-third occurs out of sequence, and functionally occupies the origin of the circumplex).
Following these constraints, and applying geometrical analysis to four significant figures of accuracy, Table 4 was constructed to indicate the vertices of each trapezoidal membership function. As the x and y axes on Scherer’s prototype mirrored each other, this table presents the membership functions of both the x and y variables, numericised according to the value of their first coordinate. It should be noted that coordinates that fall outside of the input value range are included for the sake of completeness; likewise, it should be noted that membership function 33 occurs out of sequence, as it is a special case.
Table 4. Input membership functions
Regarding the prototype of the Geneva emotion wheel, the x and y axis values mirrored each other. The graphical depiction of the MFs for the x value is shown on Fig 4.
Fig 4. Membership functions of input variable x
Each of the sixteen outputs has five membership functions describing its relative magnitude. The functions are defined as Null Intensity, Low Intensity, Middle Intensity, High Intensity and Extreme Intensity. MFs for the eSatisfaction is represented as table 5 and Fig 5.
Table 5. Output Membership functions of Geneva Emotion Wheel
Fig 5. output Membership functions of eSatisfaction
4.0. Simulation and Results
In order to implement the proposed model, we used Matlab compiler on Microsoft windows 10 platform with a 7i Intel CPU, 8 MB Cache and 8GB RAM Memory of a laptop computer.
First, the collected dataset that includes sentences and values are classified in one label of the emotions according to Ekman’s six basic emotions. In order to training and test our multi-classifier we implemented SVM Multi-Classifier with LIBSVM Library with different kernels including RBF, Polynomial and Sigmoid compared with Naïve Bayes classifier algorithm with built in function of Matlab. The results of the implemented classifier algorithms are compered to choose the efficient method. The implementation criteria and parameters are the same for all the classification models. 80% of dataset is used as train data and 20% applied for as test data. Table 6 shows the result of the implementation of the classifiers for the test data.
Table 6. Multi-Classification results with SVM with different kernels and Naïve Bayes algorithms
As it is shown on Fig7 Support vector machine with RBF kernel has more accuracy, less number iteration and less time consumption in comparison with polynomial and sigmoid kernels and Naïve Bayes algorithm.
In order to expand the classifier results base on Geneva emotion wheel to achieve the 16 emotions with valence and control, fuzzy logic with Mamdani fuzzy inferencing and centroid defuzzification is used. The Geneva emotion wheel is divided base on X, Y axis according to Fig1. The input variables are addressed between 1,1 and -1,-1. For measuring the result of the wheel implementation, some criteria for success are defined as: for any input values within the defined input range value, a numerical return is assumed a success. Error results are assumed for all inputs outside the defined operational range of the system. In our test, only one input variable was adjusted for each run of tests, the other being fixed at -1 for all experimental iterations. Table 7 shows the results of the tests.
Table 7. Results of boundary value test
One of most important results of the Geneva implantation in fuzzy logic is the weakness in the case of “extreme” which 4% input states tested and returned error. The list of J values that the errors resulted are shown in table 8. It means there are the error values return false data, in the other words the output is out of range.
Table 8. List of input values of J which error are results
In our examination the figures show membership grades with one fixed input value while the other across the defined input range. Figure 6 shows an instance where the valence value is static and fixed at 0.8 and the control value is dynamic and changes from -1.0 to 1.0. But, in figure 7 the control value fixed at 0.8 and the valence value is dynamic and changes from -1.0 to 1.0. in this figures effect of valence on control on each other is shown.
Fig. 6. The Geneva emotion wheel cross-section vs control, valence fixed value of 0.8
Fig. 7. The Geneva emotion wheel cross-section vs valence, control fixed value of 0.8
Figure 8 shows an instance where the valence value is static and fixed at 0.5 and the control value is dynamic and changes from -1.0 to 1.0. However, in figure 9 the control value fixed at 5.0 and the valence value is dynamic and changes from -1.0 to 1.0
Fig. 8. The Geneva emotion wheel cross-section vs control, valence fixed value of 0.5
Fig. 9. The Geneva emotion wheel cross-section vs valence, control fixed value of 0.5
It is clear and proven that valence and control have direct effect on intensity of each emotion. The figures demonstrate that the Geneva emotion wheel geometry is obeyed within the constraints of the model and the geometry is consistent.
In table 9 based on x and y as the system input in the defined value range, system output is shown. So we have all emotions of the Geneva emotion wheel in output.
Table 9. An instance of system output for each emotion based on the x, y inputs
The output can be used as input of affective computing systems and emotional intelligence applications.
In this paper, we proposed and implement an artificial emotional agent using classification methods and fuzzy logic. The architecture of the proposed model is supposed to resemble some of human feeling. When a computer or machine is enabled with human emotion it can think and react like humans to assist people in many ways. For example, it can act as a counselor, teacher and many agents and systems, which interact with humans. We used a collected sentence dataset with 6 basic Ekman emotions for each sentence and answered by online questionnaire by female. The sentences classified in the relevant labels and addressed on Geneva emotion wheel to extract the 16 emotions of the wheel with valence and control value. The input could be everything, such as a dataset of sentences or a dataset or input of a vision or voice recognition system. However, different dynamic parameters can be taken into account because human are both unique and complicated. Their answer may be also affected by their surrounding environment or affected by their previous experiences or their current mood. In addition, the agent can interact different regarding to its social position and status. The agent can be able to judge or have background view for people, environment and different parameters to act and react more like human. This all can be considering as future studies to develop human interaction computing and make human-human interaction with artificial intelligent agents.
 Cabanac, Michel . 2002. “What is emotion?”. Behavioural Processes 60(2): 69-83.
 Schacter, D. L., Gilbert, D. T., Wegner, D. M., & Hood, B. M. 2011. Psychology (European ed.). Basingstoke: Palgrave Macmillan.
 W. G. Parrott. 2001. Emotions in Social Psychology. Psychology Press
 Jesse Prinz. 2004. Which Emotions Are Basic?. Emotion, Evolution, and Rationality, Oxford University Press
 Russell JA. 1980. A circumplex model of affect. J Pers Soc Psychol. 1980;39:1161–78.
 Scherer KR. 2005. What are emotions? And how can they be measured? Soc Sci Inf. 2005;44(4):695–729.
 Aladdin Ayesh. William Blewitt. 2014. Models for Computational Emotions from Psychological Theories Using Type I Fuzzy Logic. Cogn Comput. DOI 10.1007/s12559-014-9287-7
 Pedrycz W Human. 2010. centricity in computing with fuzzy sets: an interpretability quest for higher order granular constructs. J Ambient Intell Humaniz Comput(2010) 1(1):65–74
 Wu D. 2012. Fuzzy sets and systems in building closed-loop affective computing systems for human–computer interaction: advances and new research directions. In: IEEE international conference on fuzzy systems, (2012) pp 1–8
 Fang-Yie Leu · Jung-chun Liu · Ya-Ting Hsu ·Yi-Li Huang. 2014. The simulation of an emotional robot implemented with fuzzy logic. Soft Comput (2014) 18:1729–1743 DOI 10.1007/s00500-013-1217-1
 Janssen, Joris H.; van den Broek, Egon L. 2016. Tune in to Your Emotions: A Robust Personalized Affective Music Player. User Modeling and User-Adapted Interaction. 22 (3): 255–279. doi:10.1007/s11257-011-9107-7. Retrieved March 2016
 Khulood Abu Maria , Raed Abu Zitar. Emotional agents. 2007 A modeling and an application, Information and Software Technology 49 (2007) 695–716
 Rafael A Calvo and Sunghwan Mac Kim. 2013. Emotions in text: dimensional and categorical
models. Computational Intelligence, 29(3).
 Dickson, Ben. 2017. Exploiting machine learning in cybersecurity”. TechCrunch. Retrieved 2017-05-23.
 Chih-Wei Hsu and Chih-Jen Lin , A Comparison of Methods for Multi-class Support Vector Machines, National Taiwan University
 Koby Crammer and Yoram Singer. 2001. On the Algorithmic Implementation of Multiclass Kernel-based Vector Machines. Journal of Machine Learning Research 2 (2001) 265-292
 Chih-Chung Chang and Chih-Jen Lin. 2011. LIBSVM: a library for support vector machines.
ACM Transactions on Intelligent Systems and Technology, 2(3):27:1–27:27, 2011. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm.
 R. W. Picard. 1995. Aﬀective Computing, M.I.T Media Laboratory Perceptual Computing Section Technical Report No. 321
 Strapparava, C., & Mihalcea, R. 2008. Learning to identify emotions in text. In Proceedings of the 23rd Annual ACM Symposium on Applied Computing, SAC’08 (pp. 1556-1560). DOI: 10.1145/1363686.1364052
 M-R Akbarzadeh-T, Ehsan Lotfi. 2013. Brain emotional learning-based pattern recognizer. Cybernetics and Systems journal. Volume 44. Issue 5. P. 402-421
 Paul Ekman. 1992. Are there basic emotions?. American Psychological Association. Vol 99. Issue 3. 1992.P 550.
Cite This Work
To export a reference to this article please select a referencing stye below:
Related ServicesView all
DMCA / Removal Request
If you are the original writer of this dissertation and no longer wish to have your work published on the UKDiss.com website then please: