Sentiment Analysis and Ontology Engineering

The problem of identifying people’s opinions expressed in written language is a relatively new and very active field of research. Having access to huge amount of data due to the ubiquity of Internet, has enabled researchers in different fields—such as natural language processing, machine learning and data mining, text mining, management and marketing and even psychology—to conduct research in order to discover people’s opinions and sentiments from the publicly available data sources. Sentiment analysis and opinion mining are typically done at various level of abstraction: document, sentence and aspect. Recently researchers are also investigating concept-level sentiment analysis, which is a form of aspect-level sentiment analysis in which aspects can be multi terms. Also recently research has started addressing sentiment analysis and opinion mining by using, modifying and extending topic modeling techniques. Topic models are probabilistic techniques for discovering the main themes existing in a collection of unstructured documents. In this book chapter we aim at addressing recent approaches to sentiment analysis, and explain this in the context of wider use. We start the chapter with a brief contextual introduction to the problem of sentiment analysis and opinion mining and extend our introduction with some of its applications in different domains. The main challenges in sentiment analysis and opinion mining are discussed, and different existing approaches to address these challenges are explained. Recent directions with respect to applying sentiment analysis and opinion mining are discussed. We will review these studies towards the end of this chapter, and conclude the chapter with new opportunities for research.

Internet has opened the new doors for information exchange and the growth of social media has created unprecedented opportunities for citizens to publicly raise their opinions, but it has serious bottlenecks when it comes to do analysis of these opinions. Even urgency to gain a real time understanding of citizens concerns has grown very rapidly. Since, the viral nature of social media which is fast and distributed one, some issues get rapidly distributed and unpredictably become important through this word of mouth opinions expressed online which in turn has known as sentiments of the users. The decision makers and people do not yet realized to make sense of this mass communication and interact sensibly with thousands of others with the help of sentiment analysis. To understand thoroughly use of sentiment analysis in today’s business world, this chapter covers the brief about sentiment analysis including introduction of sentiment analysis, early history of sentiment analysis, problems of sentiment analysis, basic concepts of sentiment analysis with mathematical treatment, sentiment and subjectivity classification comprises of opinion mining and summarization, past scenarios of opinion or sentiment collection and their analysis. Methodologies like Sentiment Analysis as Text Classification Problem, Sentiment analysis as Feature Classification with mathematical treatment are explored. Also, Economic consequences of sentiment analysis on individual, society and organization with the help of social media sentiment analysis are provided as supporting component.

Due to the growing volume of available textual information, there is a great demand for Natural Language Processing (NLP) techniques that can automatically process and manage texts, supporting the information retrieval and communication in core areas of society (e.g. healthcare, business, and science). NLP techniques have to tackle the often ambiguous and linguistic structures that people use in everyday speech. As such, there are many issues that have to be considered, for instance slang, grammatical errors, regional dialects, figurative language, etc. Figurative Language (FL), such as irony, sarcasm, simile, and metaphor, poses a serious challenge to NLP systems. FL is a frequent phenomenon within human communication, occurring both in spoken and written discourse including books, websites, fora, chats, social network posts, news articles and product reviews. Indeed, knowing what people think can help companies, political parties, and other public entities in strategizing and decision-making polices. When people are engaged in an informal conversation, they almost inevitably use irony (or sarcasm) to express something else or different than stated by the literal sentence meaning. Sentiment analysis methods can be easily misled by the presence of words that have a strong polarity but are used sarcastically, which means that the opposite polarity was intended. Several efforts have been recently devoted to detect and tackle FL phenomena in social media. Many of applications rely on task-specific lexicons (e.g. dictionaries, word classifications) or Machine Learning algorithms. Increasingly, numerous companies have begun to leverage automated methods for inferring consumer sentiment from online reviews and other sources. A system capable of interpreting FL would be extremely beneficial to a wide range of practical NLP applications. In this sense, this chapter aims at evaluating how two specific domains of FL, sarcasm and irony, affect Sentiment Analysis (SA) tools. The study’s ultimate goal is to find out if FL hinders the performance (polarity detection) of SA systems due to the presence of ironic context. Our results indicate that computational intelligence approaches are more suitable in presence of irony and sarcasm in Twitter classification.

People’s opinion has always driven human choices and behaviors, even before the diffusion of Information and Communication Technologies. Thanks to the World Wide Web and the widespread of On-Line collaborative tools such as blogs, focus groups, review web sitesorums, social networks, millions of messages appear on the web, which is becoming a rich source of opinioned data. Sentiment analysis refers to the use of natural language processing, text analysis and computational linguistics to identify and extract subjective information in documents, comments and posts. The aim of this work is to show how the adoption of a probabilistic approach based on the Latent Dirichlet Allocation (LDA) as Sentiment Grabber can be an effective Sentiment Analyzer. Through this approach, for a set of documents belonging to a same knowledge domain, a graph, the Mixed Graph of Terms, can be automatically extracted. This graph, which contains a set of Mixed Graph of Terms, can be transformed in a Sentiment Oriented Terminological Ontology thanks to a methodology that involves the introduction of annotated lexicon as Wordnet. The chapter shows how the obtained ontology can be discriminative for sentiment classification. The proposed method has been tested in different contexts: standard datasets and comments extracted from social networks. The experimental evaluation shows how the proposed approach is effective and the results are quite satisfactory.

Description Logic (DL) Class Expression Learning (CEL) is a recent research topic of interest in the field of machine learning. Given a set of positive and negative examples of individuals in an ontology, the learning problem consists of finding a new class expression or concept such that most of the positive examples are instances of that concept, whereas the negatives examples are not. Therefore, the class expression learning can be seen as a search process in the space of concepts. In this chapter, the use of CEL algorithms is proposed as a tool to find the class expression that describes as much of the instances of positive documents as possible, being the main novelty of the proposal that the ontology is focused on inferring knowledge at syntactic level to determine the orientation of opinion. Furthermore, the use of CEL algorithms can be an alternative to complement other types of classifiers for sentiment analysis, incorporating such description classes as relevant new features into the knowledge base. To do so, an ontology-based text model for the representation of text documents is presented. The process for the ontology population and the use of the class expression learning of sentiment concepts are also described. To show the usefulness and effectiveness of our proposal, we use a set of documents about positive feedback focused on films to learn the positive sentiment concept and to classify the documents, comparing the results obtained against the result obtained by a C4.5 decision tree classifier, using the standard bag of words structure. Finally, we describe the problems that have arisen and solutions that have been adopted in our proposal.

Distilling sentiments and moods hidden in the written (natural) language is a challenging issue which attracts research and commercial interests, aimed at studying the users behavior on the Web and evaluating the public attitudes towards brands, social events, political actions. The understanding of the written language is a very complicated task: sentiments and opinions are concealed in the sentences, typically associated to adjectives and verbs; then the intrinsic meaning of some textual expressions is not amenable to rigid linguistic patterns. This work presents a framework for detecting sentiment and emotion from text. It exploits an affective model known as Hourglass of Emotions, a variant of Plutchik’s wheel of emotions. The model defines four affective dimensions, each one with some activation levels, called ‘sentic levels’ that represent an emotional state of mind and can be more or less intense, depending on where they are placed in the corresponding dimension. Our approach draws from the Computational Intelligence area to provide a conceptual setting to sentiment and emotion detection and processing. The novelty is the fuzzy linguistic modeling of the Hourglass of Emotions: dimensions are modeled as fuzzy linguistic variables, whose linguistic terms are the sentic levels (emotions). This linguistic modeling naturally enables the use of fuzzy linguistic aggregation operators (from Computing with Words paradigm), such as LOWA (Linguistic Ordered Weighted Averaging) that inherently accomplishes an aggregation of the emotions in order to get an emotional expression that synthesizes a set of emotions associated with different sentic levels and activation intensities. The whole process for the emotion detection and synthesis is described through its main tasks, from the text parsing up to emotions extraction, returning a predominant emotion, associated with each dimension of the Hourglass of Emotions. An ad-hoc ontology has been designed to integrate lexical information and relations, along with the Hourglass model.

Due to the rapid growth of structured/unstructured and user-generated data (e.g., social media sites) volume of data is becoming too big or it moves too fast or it exceeds current processing capacity and so traditional data processing applications are inadequate. Computational Intelligence with Concept-based approaches can detect sentiments analyzing the concept based on text expressions without analyzing the singlef words as in the purely syntactical techniques. On human-centric intelligent systems Semantic networks can simulate the human complex frames in a reasoning process providing efficient association and inference mechanisms, while ontology can be used to fill the gap between human and Computational Intelligence for a task domain. For an evolving environment it is necessary to understand what knowledge is required for a task domain with an adaptive ontology matching. To reflect the evolving knowledge this paper considers ontologies based on folksonomies according to a new concept structure called “Folksodriven” to represent folksonomies. To solve the problems inherent an uncontrolled vocabulary of the folksonomy it is presented a Folksodriven Structure Network (FSN): a folksonomy tags suggestions built from the relations among the Folksodriven tags (FD tags). It was observed that the properties of the FSN depend mainly on the nature, distribution, size and the quality of the reinforcing FD tags. So, the studies on the transformational regulation of the FD tags are regarded to be important for an adaptive folksonomies classifications in an evolving environment used by Intelligent Systems to represent the knowledge sharing. The chapter starts from the discussion on the deformation exhibiting linear behavior on FSN based on folksonomy tags chosen by different user on web site resources. Then it’s formulated a constitutive law on FSN investigating towards a systematic mathematical analysis on stress analysis and equations of motion for an evolving ontology matching on an environment defined by the users’ folksonomy choice. The adaptive ontology matching and the elastodynamics are merged to obtain what we can call the elasto-adaptive-dynamics methodology of the FSN. Furthermore it is shown the last development defining a hyperelastic dynamic considering the internal folksonomy behavior of the stress and strain from original to deformed configuration.

Sentiment analysis research has acquired a growing importance due to its applications in several different fields. A large number of companies have included the analysis of opinions and sentiments of costumers as a part of their mission. Therefore, the analysis and automatic classification of large corpora of documents in natural language, based on the conveyed feelings and emotions, has become a crucial issue for text mining purposes. This chapter aims to relate the sentiment-based characterization inferred from books with the distribution of emotions within the same texts. The main result consists in a method to compare and classify texts based on the feelings expressed within the narrative trend.

Sentiment analysis, which is also known as opinion mining, has been an increasingly popular research area focusing on sentiment classification/regression. In many studies, computational models have been considered as effective and efficient tools for sentiment analysis. Computational models could be built by using expert knowledge or learning from data. From this viewpoint, the design of computational models could be categorized into expert based design and data based design. Due to the vast and rapid increase in data, the latter approach of design has become increasingly more popular for building computational models. A data based design typically follows machine learning approaches, each of which involves a particular strategy of learning. Therefore, the resulting computational models are usually represented in different forms. For example, neural network learning results in models in the form of multi-layer perceptron network whereas decision tree learning results in a rule set in the form of decision tree. On the basis of above description, interpretability has become a main problem that arises with computational models. This chapter explores the significance of interpretability for computational models as well as analyzes the factors that impact on interpretability. This chapter also introduces several ways to evaluate and improve the interpretability for computational models which are used as sentiment analysis systems. In particular, rule based systems, a special type of computational models, are used as an example for illustration with respects to evaluation and improvements through the use of computational intelligence methodologies.

These years, micro-blog emotion mining becomes one of the research hotspots in social network data mining. Different from state of the art study, this paper presents a novel method for emotion classification, which is SVM\(^{\textit{perf}}\) based method combined with syntactic structure of Chinese micro-blogs. The classified emotion type includes Happiness, Anger, Disgust, Fear, Sadness and Surprise. For the proposed method, an emotional lexicon is constructed and linguistic features are extracted from micro-blog corpus firstly. Secondly, for the current feature space dimension is higher, Chi-square test is used to extract the high-frequency and high-class relevance keywords. At the same time, Pointwise Mutual Information (PMI) is used to pick the effective low frequency words in feature dimension reduction, which can reduce the computational complexity. Finally, SVM\(^{\textit{perf}}\) is applied for the emotion classification. In order to illustrate the effectiveness of the algorithm, LIBSVM and SVM-Light are used as the baseline. The data from Sina Micro-blog (weibo.com) have been used as the experiment data. The experiment results demonstrate that all the above features contribute to emotion classification in micro-blogs, and the results validate the feasibility of the proposed approach. It also shows that SVM\(^{\textit{perf}}\) is an appropriate choice of classifier for emotion classification.

The motivation of this chapter hinges on the growing popularity in the use of news and social media information and their increasing influence on the financial investment community. This chapter investigates the interplay between news/social sentiment and financial market movement in the form of empirical impact. The underlying belief is that news and social media influence investor sentiment, which in turn drives financial decisions and predicates the upward or downward movement of the financial markets. This book chapter contributes to the existing literature of sentiment analysis in the following three areas: (a) It provides a review of existing findings about influence of social media and news sentiment to asset prices and documents the persistent correlation between media sentiment and market movement. (b) It shows that abnormal news sentiment can be a predictive proxy for financial market returns and volatility, based on the intuition that extreme investor sentiment changes tend to have long and last effects to market movement. (c) It presents a number of approaches to formulate investment strategies based on the sentiment trend, shocks and feedback strength. The results show that the sentiment-based strategies yield superior risk-adjusted returns over other benchmark strategies. Altogether, this chapter provides a framework of existing empirical knowledge on the impact of sentiment on financial markets and further prescribes advanced investment strategies based on sentiment analytics.

Customer experience management (CEM) denotes a set of practices, processes, and tools, that aim at personalizing customer’s interactions with a company according to customer’s needs and desires (Weijters et al., J Serv Res 10(1):3–21, 2007 [29]). E-business specialists have long realized the potential of ubiquitous computing to develop context-aware CEM applications (CA-CEM), and have been imagining CA-CEM scenarios that exploit a rich combination of sensor data, customer profile data, and historical data about the customer’s interactions with his environment. However, to realize this potential, e-commerce tool vendors need to figure out which software functionalities to incorporate into their products that their customers (e.g. retailers) could use/configure to build CA-CEM solutions. We propose to provide such functionalities in the form of an application framework within which CA-CEM functionalities can be specified, designed, and implemented. Our framework relies on, (1) a cognitive modeling of the purchasing process, identifying potential touchpoints between sellers and buyers, and relevant influence factors, (2) an ontology to represent relevant information about consumer categories, property types, products, and promotional material, (3) computational intelligence techniques to compute consumer- or category-specific property values, and (4) approximate reasoning algorithms to implement some of the CEM functionalities. In this paper, we present the principles underlying our framework, and outline steps for using the framework for particular purchase scenarios. We conclude by discussing directions for future research.

Sentiment analysis refers to the class of computational and natural language processing based techniques used to identify, extract or characterize subjective information, such as opinions, expressed in a given piece of text. The main purpose of sentiment analysis is to classify a writer’s attitude towards various topics into positive, negative or neutral categories. Sentiment analysis has many applications in different domains including, but not limited to, business intelligence, politics, sociology, etc. Recent years, on the other hand, have witnessed the advent of social networking websites, microblogs, wikis and Web applications and consequently, an unprecedented growth in user-generated data is poised for sentiment mining. Data such as web-postings, Tweets, videos, etc., all express opinions on various topics and events, offer immense opportunities to study and analyze human opinions and sentiment. In this chapter, we study the information published by individuals in social media in cases of natural disasters and emergencies and investigate if such information could be used by first responders to improve situational awareness and crisis management. In particular, we explore applications of sentiment analysis and demonstrate how sentiment mining in social media can be exploited to determine how local crowds react during a disaster, and how such information can be used to improve disaster management. Such information can also be used to help assess the extent of the devastation and find people who are in specific need during an emergency situation. We first provide the formal definition of sentiment analysis in social media and cover traditional and the state-of-the-art approaches while highlighting contributions, shortcomings, and pitfalls due to the composition of online media streams. Next we discuss the relationship among social media, disaster relief and situational awareness and explain how social media is used in these contexts with the focus on sentiment analysis. In order to enable quick analysis of real-time geo-distributed data, we will detail applications of visual analytics with an emphasis on sentiment visualization. Finally, we conclude the chapter with a discussion of research challenges in sentiment analysis and its application in disaster relief.

The rapid growth of the World Wide Web and social media allows users playing an active role in the contents’ creation process. Users can evaluate the brands’ reputation and quality exploiting the information provided by new marketing channels, such as social media, social networks, and electronic commerce (or e-commerce). Consequently, enterprises need to spot and analyze these digital data in order to improve their reputation among the consumers. The aim of this chapter is to highlight the common approaches of sentiment analysis in social media streams and the related issues with the cloud computing, providing the readers with a deep understanding of the state of the art solutions.

Consumers often provide on-line reviews on products or services they have purchased, and frequently seek on-line reviews about a product or service before deciding whether to make a purchase. Organisations seek consumer opinions about their products, since this invaluable information allows them to improve future product versions, and to predict sales. The vast amount of on-line customer reviews has attracted research into approaches for intelligently mining these reviews to support decision-making processes. This chapter provides an overview of recent fuzzy-based approaches to sentiment analysis of customer reviews. It also presents a framework which can be utilised for sentiment analysis and review rating prediction tasks. The framework includes methods for preparing the dataset; extracting the best features for prediction via Singular Value Decomposition and a Genetic Algorithm; and constructing a classifier for performing the review rating predictions.

Since the inception of the Web 2.0, World Wide Web is widely being used as a platform by customers and manufactures to share experiences and opinions regarding products, services, marketing campaigns, social events, etc. As a result, there is enormous growth in user-generated contents (e.g. customer reviews), providing an opportunity for data analysts to computationally evaluate users’ sentiments and emotions for developing real-life applications for business intelligence, product recommendation, enhanced customer services, and target marketing. Since users’ feedbaks (aka reviews) are very useful for products development and marketing, large business houses and corporates are taking interest in opinion mining and sentiment analysis systems. In this chapter, we propose the design of an Ontology Learning and Sentiment Analysis (OntoLSA) system for ontology learning and sentiment analysis using rule-based and machine learningapproaches. The rule-based approach aims to identify candidate concepts, which are analyzed using a customized HITS algorithm to compile a list of feasible concepts. Feasible concepts and their relationships (both structural and non-structural) are used to generate a domain ontology, which is later on used for opinion mining and sentiment analysis. The proposed system is also integrated with a visualization module to facilitate users to navigate through review documents at different levels of granularity using a graphical user interface.

Disease monitoring and tracking is of tremendous value, not only for containing the spread of contagious diseases but also for avoiding unnecessary public concerns and even panic. In this chapter, we present a near real-time sentiment analysis service of public health-related tweets. Traditionally, it is impossible for humans to effectively measure the degree of public health concerns due to limited resources and significant time delays. To solve this problem, we have developed a computational intelligence approach for Epidemic Sentiment Monitoring System (ESMOS) to automatically analyze the disease sentiments and gauge the Measure of Concern (MOC) expressed by Twitter users. More specifically, we present a knowledge-based approach that employs a disease ontology to detect the outbreak of diseases and to analyze the linguistic expressions that convey subjective expressions and sentiment polarity of emotions, feelings, opinions, personal attitudes, etc. with a sentiment classifier. The two-step sentiment classification method utilizes the subjective vocabulary corpus (MPQA), sentiment strength corpus (AFINN), as well as emoticons and profanity words that are often used in social media postings. It first automatically classifies the tweets into personal and non-personal classes, eliminating many tweets such as non-personal “retweets” of news articles from further consideration. In the second stage, the personal tweets are classified into Negative and non-Negative sentiments. In addition, we present a model to quantify the public’s Measure of Concern (MOC) about a disease, based on sentiment classification results. The trends of the public MOC are visualized on a timeline. Correlation analyses between MOC timeline and disease-related sentiment category timelines show that the peaks of the MOC are weakly correlated with the peaks of the News timeline without any appreciable time delay or lead. Our sentiment analysis method and the MOC trend analyses can be generalized to other topical domains, such as mental health monitoring and crisis management. We present the ESMOS prototype for public health-related disease monitoring, for public concern trending and for mapping analyses.