Expressions of Graduality for Sentiments Analysis - A Survey Fabon Dzogang, Marie-Jeanne Lesot, Maria Rifqi and Bernadette Bouchon-Meunier, Senior Member, IEEE Abstract— Given the very ambiguous and imprecise nature of sentiments and of their expressions, this survey focuses on ap- proaches making use of components of graduality in the task of automatic sentiments analysis. To that aim, we review methods taking account of intrinsic psychological models components of graduality as well as extrinsic components issued from computational intelligence approaches. In particular, beyond psychological models of sentiments that define affective states as multidimensional vectors in affective continuous spaces, we identify three components of graduality, namely composition or blending, intensity and inheritance. In our discussion, we review how fuzzy set theory as well as other gradual structures based on a vectorial representation are employed to describe affective states as complex or imprecise entities. Finally, we focus on verbal expressions of sentiments and more specifically, we discuss the use of components of graduality in order to deal with sentiments complex and subtle expressions issued from the expressive power of natural languages. I. I NTRODUCTION Recently, the interest in sentiment analysis has grown very fast. As a matter of fact, an increasing number of scientists are tackling the task of automated sentiments detection and classification, making up a domain called affective computing. On one hand, the identification and characterization of opinions as positive, negative and neutral are tackled in the opinion mining task [1]; on the other hand, the study of sentiments is addressed in the broader sentiment analysis task. These two tasks reunite in the global area of affective computing in which psychological and compu- tational intelligence concepts are, among others, employed in order to identify affective content, classify sentiments or simulate affective agents. Practical applications of this new area of research include for instance, emotion aware robots, automatic movies classifiers, intelligent computer interfaces or avatars, next generation video-games design and automatic marketing surveys. To tackle the automatic sentiments analysis problem, one first needs to define what are sentiments; to that aim authors refer mainly to psychological models. Appraisal event mod- els make use of sentiment eliciting rules for modeling the human affective mechanisms. Even though appraisal based approaches are well suited for the development of artificial affective agents, they provide poor asset in the task of sentiments discrimination and are not covered in this survey. Nevertheless, it must be underlined that some appraisal based approaches make use of graduality through fuzzy inference and fuzzy aggregation for processing affective mechanisms Fabon Dzogang, Marie-Jeanne Lesot, Maria Rifqi and Bernadette Bouchon-Meunier are with Department of machine learning (DAPA), Uni- versit´ e Pierre et Marie Curie - Paris6, CNRS, UMR7606, LIP6. 4 place Jussieu F - 75252 Paris cedex 05. (phone: +33 144 278 887; email: fabon.dzogang@lip6.fr) ambiguity and imprecision, the interested reader might refer to [2], [3], [4]. Beside this appraisal model, two other types of psycho- logical models are respectively categorical models that are based on a set of affective states, and dimensional models that describe sentiments as vectors in a multidimensional contin- uous space. Both possess a valuable descriptive power in the task of sentiments discrimination; as such, they are widely employed in the field of machine learning for tackling the task of sentiment analysis. For instance, they are extensively employed for analyzing sentiments in texts [5], audio [6], video frames [7] or from physiological measures [8]. While the benefits of applying such methods to real life applications are certain, many problems remain unresolved and get in the way of methods for classifying, identifying or discriminating sentiments: for instance, the ambiguous expression of feelings, the influence of the context for their interpretation but also the very personal nature of sentiments which for instance depends on cultures, languages, ages or personal experiences. Adopting the point of view that crisp models and crisp processes might miss the actual point in analyzing sentiments usually expressed with ambiguity and imprecision, this survey focuses on approaches exploiting components of graduality in the task of automatically an- alyzing, classifying or identifying sentiments. In our study the concept of graduality is carried in representations that take account of psychological models intrinsic components. Beyond dimensional models that offer a naturally gradual framework we distinguish between three components of graduality: composition or blending, intensity and inheritance. We also take into consideration extrinsic components of graduality issued from computational intelligence ap- proaches. While the fuzzy set theory offers an extensive set of possibilities to model the ambiguities and imprecision of affective states, we also review methods making use of other gradual structures based on a vectorial representation. This paper is organized as follow. In Section II we dis- cuss the expression of graduality in approaches making use of affective sets representations of sentiments, also known as categorical models. More specifically, we identify three components of graduality: composition or blending, inten- sity and inheritance. In Section III we review the uses of dimensional models. While the task of classification often leads authors to segment the affective space into affective sets, pure dimensional approaches make use of continuity for expressing graduality. Lastly, in Section IV we focus on a particular modality: the expression of sentiments in natural languages, the expressive power of which is particularly well modelled by means of gradual models and processes. The