Environmental, Social and Governance (ESG) integration in financial activities is a crucial topic that is gaining importance in financial markets. During the years, many studies have been conducted about Initial Public Offering (IPO) and underpricing since they are fundamental aspects of firms’ lifecycle. Nevertheless, none of these studies have appropriately related firms’ ESG characteristics to IPO underpricing. In order to fill this knowledge gap, this thesis’s purpose is to investigate whether the ESG factors of a firm have effects on its IPO underpricing in the U.S stock market. The U.S has been chosen as it is the biggest stock market in the world and because of the quality and reliability of the data available for this country.  A quantitative study is applied to investigate the relationship between ESG characteristics of the firms and the level of underpricing. First, to obtain the measurement of the ESG level of the pre-IPO firms, we have conducted two textual analysis of IPO prospectus, namely, term frequency and sentiment analysis. These indicators aim to show the disclosure level of ESG factors and whenever ESG is perceived negatively or positively by the market. Successively, the multiple regression is performed for each ESG indicator to find which measures have the analytical abilities to explain IPO underpricing. Based on the multiple regression results, we can conclude that the frequency of environmental & governance terms occurred in IPO prospectus, the negative tone, and the overall sentiment of the environmental context are significantly explaining IPO underpricing. These results have given meaningful answers for our research. The market does not perceive the social factors of a firm in the IPO context. On the other hand, environmental and governance aspects still attract the market’s attention in different ways. The market is concerned about the disclosure level of the governance activities and whether these activities are sufficiently mentioned in the prospectus. Meanwhile, the market takes into serious consideration the environmental activities of a firm by assessing the qualities of these activities. Moreover, the market is more sensitive to the negative information about environmental content than positive information in the IPO context. The textual analysis methods applied in this thesis have some limitations. However, this study has the reliability to confirm that some companies’ ESG factors affect IPO underpricing. As a consequence, it is possible to state that the market cares about  ESG issues.

Denna studie undersöker möjligheten att med hjälp av sentimentanalys av golfspelares twitterkonton kunna förutsäga deras kommande resultat. Studien baserades på två dataset: 155 professionella golfares resultat och 112 101 tweets insamlade från två säsonger på PGA­touren. Vår studie kan vara av intresse för till exempel spelbolag, spelare, tränare och fans. Det känslor golfspelarna uttryckt i sina tweets kvantifierades till ett siffervärde med hjälp av den lexikala sentimentsanalysmetoden AFINN. Resultaten av vår studie visar på mycket låg korrelation mellan de insamlade dataseten och att sentimentvärdena innehar en låg grad av prediktiv förmåga. Dessa resultat står i kontrast mot liknande forskning utförd på annan sport. Vår rekommendation för framtida studier är att basera modellen på fler variabler utöver sentimentvärde för att tydligare klargöra hur de känslor golfspelare uttrycker på twitter kan användas för att förutspå deras kommande resultat. / In this study we examine the relationship between the sentiment value of golf players’ tweets and their sports results to evaluate the predictive power of the their twitter accounts. Findings on this topic may be of value to bookmakers, gamblers, coaches and fans of sport. Our study is based on two datasets: PGA­tour golf statistics and 112 101 tweets made by 155 profesional golfers over the course of two seasons. The golf players’ sentiment was quantified using the lexical sentiment analysis method AFINN. In contrast to other research with similiar methods, our findings suggest that there is low correlation betweet the datasets and that the methods used in our study have low predictive power. Our recommendation is that future studies use additional prediction variables besides sentiment score to better evaluate the predictive power of golf players’ tweets.

Sentiment classification is the process of categorizing data into categories based on its polarity with a wide array of applications across several industries. This report examines a combination of two prominent approaches to sentiment classification using a lexicon of weighted words and machine learning respectively. These approaches are compared with the combined hybrid approach in order to give an account of their relative strengths and weaknesses. When run on a set of IMDb movie reviews the results indicate that the hybrid model performs better than the lexicon-based approach, in turn being outperformed by the learning-based approach. However, the gain in convenience brought on by eliminating the need for training data makes the hybrid model an appealing alternative to the other approaches with a slight trade-off in performance. / Att klassificera text i kategorier baserat på känslan de uttrycker är ett aktuellt område idag och kan tillämpas inom många industrier. Rapporten undersöker en kombination av de två framstående tillvägagångssätten till denna typ av klassificering baserade på ett lexikon med definerade ordvikter respektive maskininlärning. Denna hybridlösning jämförs mot de två andra tillvägagångssätten för att framlägga deras relativa styrkor och svagheter. På ett dataset med filmrecensioner från IMDb får maskininlärningsklassificeraren bäst resultat, följt av hybridlösningen och sist den lexikonbaserade lösningen. Trots det kan hybridlösningen vara att föredra i situationer där det är ogenomförbart eller oskäligt att förbereda träningsdata för maskininlärningsklassificeraren, dock med ett visst avkall på prestanda.

Indiana University-Purdue University Indianapolis (IUPUI) / The number of smart mobile devices has grown at a significant rate in recent years. This growth has resulted in an exponential number of publicly available mobile Apps. To help the selection of suitable Apps, from various offered choices, the App distribution platforms generally rank/recommend Apps based on average star ratings, the number of installs, and associated reviews ― all the external factors of an App. However, these ranking schemes typically tend to ignore critical internal factors (e.g., bugs, security vulnerabilities, and data leaks) of the Apps. The AppStores need to incorporate a holistic methodology that includes internal and external factors to assign a level of trust to Apps. The inclusion of the internal factors will describe associated potential security risks. This issue is even more crucial with newly available Apps, for which either user reviews are sparse, or the number of installs is still insignificant. In such a scenario, users may fail to estimate the potential risks associated with installing Apps that exist in an AppStore. This dissertation proposes a security-related and evidence-based ranking framework, called SERS (Security-related and Evidence-based Ranking Scheme) to compare similar Apps. The trust associated with an App is calculated using both internal and external factors (i.e., security flaws and user reviews) following an evidence-based approach and applying subjective logic principles. The SERS is formalized and further enhanced in the second part of this dissertation, resulting in its enhanced version, called as E-SERS (Enhanced SERS). These enhancements include an ability to integrate any number of sources that can generate evidence for an App and consider the temporal aspect and reputation of evidence sources. Both SERS and E-SERS are evaluated using publicly accessible Apps from the Google PlayStore and the rankings generated by them are compared with prevalent ranking techniques such as the average star ratings and the Google PlayStore Rankings. The experimental results indicate that E-SERS provides a comprehensive and holistic view of an App when compared with prevalent alternatives. E-SERS is also successful in identifying malicious Apps where other ranking schemes failed to address such vulnerabilities. In the third part of this dissertation, the E-SERS framework is used to propose a trust-aware composition model at two different granularities. This model uses the trust score computed by E-SERS, along with the probability of an App belonging to the malicious category, as the desired attributes for selecting a composition as the two granularities. Finally, the trust-aware composition model is evaluated with the average star rating parameter and the trust score. A holistic approach, as proposed by E-SERS, to computer a trust score will benefit all kinds of Apps including newly published Apps that follow proper security measures but initially struggle in the AppStore rankings due to a lack of a large number of reviews and ratings. Hence, E-SERS will be helpful both to the developers and users. In addition, the composition model that uses such a holistic trust score will enable system integrators to create trust-aware distributed systems for their specific needs.

Web surveys in higher education are particularly important for assessing the quality of academic teaching and learning. Traditionally, mainly quantitative data is used for quality assessment. Increasingly, questions are being raised about the impact of attitudes of the individuals involved. Therefore, especially the analysis of open-ended text responses in web surveys offers the potential for impact evaluation. Despite the fact that qualitative text mining and sentiment analysis are being introduced in other research areas, these instruments are still slowly gaining access to evaluation research. On the one hand, there is a lack of methodological expertise to deal with large numbers of text responses (e.g. via semantic analysis, linguistically supported coding, etc.). On the other hand, deficiencies in interdisciplinary expertise are identified in order to be able to contextualize the results. The following contribution aims to address these issues. The presentation will contribute to the field of impact evaluation and reveals methodological implications for the development of text mining and sentiment analysis in evaluation processes.

Dans cette thèse, nous étudions comment les méthodes d'apprentissage de représentations peuvent être appliquées à des données générées par l'utilisateur. Nos contributions couvrent trois applications différentes, mais partagent un dénominateur commun: l'extraction des représentations d'utilisateurs concernés. Notre première application est la tâche de recommandation de produits, où les systèmes existant créent des profils utilisateurs et objets qui reflètent les préférences des premiers et les caractéristiques des derniers, en utilisant l'historique. De nos jours, un texte accompagne souvent cette note et nous proposons de l'utiliser pour enrichir les profils extraits. Notre espoir est d'en extraire une connaissance plus fine des goûts des utilisateurs. Nous pouvons, en utilisant ces modèles, prédire le texte qu'un utilisateur va écrire sur un objet. Notre deuxième application est l'analyse des sentiments et, en particulier, la classification de polarité. Notre idée est que les systèmes de recommandation peuvent être utilisés pour une telle tâche. Les systèmes de recommandation et classificateurs de polarité traditionnels fonctionnent sur différentes échelles de temps. Nous proposons deux hybridations de ces modèles: la première a de meilleures performances en classification, la seconde exhibe un vocabulaire de surprise. La troisième et dernière application que nous considérons est la mobilité urbaine. Elle a lieu au-delà des frontières d'Internet, dans le monde physique. Nous utilisons les journaux d'authentification des usagers du métro, enregistrant l'heure et la station d'origine des trajets, pour caractériser les utilisateurs par ses usages et habitudes temporelles. / In this thesis, we study how representation learning methods can be applied to user-generated data. Our contributions cover three different applications but share a common denominator: the extraction of relevant user representations. Our first application is the item recommendation task, where recommender systems build user and item profiles out of past ratings reflecting user preferences and item characteristics. Nowadays, textual information is often together with ratings available and we propose to use it to enrich the profiles extracted from the ratings. Our hope is to extract from the textual content shared opinions and preferences. The models we propose provide another opportunity: predicting the text a user would write on an item. Our second application is sentiment analysis and, in particular, polarity classification. Our idea is that recommender systems can be used for such a task. Recommender systems and traditional polarity classifiers operate on different time scales. We propose two hybridizations of these models: the former has better classification performance, the latter highlights a vocabulary of surprise in the texts of the reviews. The third and final application we consider is urban mobility. It takes place beyond the frontiers of the Internet, in the physical world. Using authentication logs of the subway users, logging the time and station at which users take the subway, we show that it is possible to extract robust temporal profiles.

Twelve years after Satoshi Nakamoto published the paper describing the functioning mechanism and principals of cryptocurrency that maintains secure and anonymous digital transactions beyond any banks, cryptocurrencies have become a multi-billion-dollar industry comprising millions of investors, miners, developers and profiteers. However, the actual price determinants and ways to forecast future price changes remain an open question yet to discover the answer for. This study attempts to figure out whether media hype exerts that much influence upon cryptocurrencies price movements and whether it can be used as the basis for future movements prediction. Two cryptocurrencies, Bitcoin and Tezos, and 7 mass and social media factors for each of them were considered on daily basis from 08-01-2018 to 10-31-2020. To explore the interdependence between media drivers and cryptocurrencies' prices in short, medium and long timespan, this study deploys wavelet coherence approach. There was found, that price changes turn to be the supreme prior to hype, even though the growing ado may push the prices even higher. Thus, hype is failing to prove itself as a reliable cryptocurrency price predictor. Crypto investors, though, should anyways take the news background into account while building trading strategies,...

OpenAI recently introduced ChatGPT, a chatbot powered by the GPT-3 family of deep learninglanguage models (LLMs). With the aid of machine learning techniques, ChatGPT has been fine-tuned to improve its capacity to respond to a diverse range of queries, and it has been describedas one of the most advanced machine learning technologies currently available. While AI israpidly advancing and being integrated into society, the comprehension of people's attitudestowards these novel technologies is not progressing at the same rate. Prior research studies andliterature have highlighted the importance of assessing user sentiment towards newly launchedAI services. Evaluating the expressed attitudes towards the recently introduced ChatGPT canprovide valuable insights into the product's potential, as well as highlighting any challenges orproblems encountered by users. This paper presents a study that examines the attitudesexpressed on the social media platforms Twitter and Reddit. For data collection, this studyutilized social media data in the form of free text obtained through the APIs of Twitter andReddit. A qualitative analysis is carried out with the aid of a sentiment analysis tool to assesslanguage and categorize text data based on their expressed attitudes. This data is presented in aquantitative summary. The findings indicate a favorable disposition among users towardsChatGPT in general but that there are areas of concern where users have conveyed sentimentsof feeling intimidated or having a negative resonance with ChatGPT's capabilities andachievements. This study contributes to the existing understanding of user attitudes towardsChatGPT and highlights the necessity for further research to delve deeper into this area.

The abundance, accessibility, and scale of data have engendered an era where machine learning can quickly and accurately solve complex problems, identify complicated patterns, and uncover intricate trends. One research area where many have applied these techniques is the stock market. Yet, financial domains are influenced by many factors and are notoriously difficult to predict due to their volatile and multivariate behavior. However, the literature indicates that public sentiment data may exhibit significant predictive qualities and improve a model’s ability to predict intricate trends. In this study, momentum SVM classification accuracy was compared between datasets that did and did not contain sentiment analysis-related features. The results indicated that sentiment containing datasets were typically better predictors, with improved model accuracy. However, the results did not reflect the improvements shown by similar research and will require further research to determine the nature of the relationship between sentiment and higher model performance.

Natural Language Processing (NLP) is a branch in data science that marries artificial intelligence with linguistics. Essentially, it tries to program computers to understand human language, both spoken and written. Over the past decade, researchers have applied novel algorithms to gain a better understanding of human sentiment. While no easy feat, incredible improvements have allowed organizations, politicians, governments, and other institutions to capture the attitudes and opinions of the public. It has been particularly constructive for companies who want to check the pulse of a new product or see what the positive or negative sentiments are for their services. NLP has even become useful in boosting sales performance and improving training. Over the years, there have been countless studies on sales performance, both from a psychological perspective, where characteristics of salespersons are explored, and from a data science/AI (Artificial Intelligence) perspective, where text is analyzed to predict sales forecasting (Pai & Liu, 2018) and coach sales agents using AI trainers (Luo et al., 2021). However, few studies have discussed how NLP models can help characterize sales performance using actual sales transcripts. Thus, there is a need to explore to what extent NLP models can inform B2B businesses of the characteristics embodied within their salesforce. This study aims to fill that literature gap. Through a partnership with a medium-sized tech company based out of California, USA, this study conducted an experiment to try and answer to what extent can we characterize sales performance based on real-life sales communication? And in what ways can conversational data inform the sales team at a California-based mid-sized tech company about how top performers communicate with customers? In total, over 5000 sentences containing over 110 000 words were collected and analyzed using two separate rule-based sentiment analysis techniques: TextBlob developed by Steven Loria (2013) and Valence Aware Dictionary and sEntiment Reasoner (VADER) developed by CJ Hutto and Eric Gilbert (2014). A Naïve Bayes classifier was then adopted to test and train each sentiment output from the two rule-based techniques. While both models obtained high accuracy, above 90%, it was concluded that an oversampled VADER approach yields the highest results. Additionally, VADER also tends to classify positive and negative sentences more correctly than TextBlob, when manually reviewing the output, hence making it a better model for the used dataset.