No description available.

No description available.

This study investigates the use of emojis in sentiment analysis for the Swedish language, with the objective to assess if emojis improve the performance of the model. Sentiment analysis is an NLP classification task aimed at extracting people's opinions, sentiments, and attitudes from language. Though sentiment analysis as a research area has made a lot of progress recently, there are still some challenges to overcome. In this work, two of these challenges were considered; the analysis of a non-English language and the impact of emojis. These areas were explored through creating a sentiment annotated dataset of Swedish texts containing emojis, and creating a Swedish sentiment analysis model for evaluation. The sentiment analysis model created, SweVADER, was based on the English Lexicon-based model VADER.  The best performing SweVADER model achieved an accuracy of 0.53 and an F1-score of 0.47. Furthermore, the presence of emojis improved the analysis for most models, but not by much. The results indicate that the use of emojis can improve the sentiment analysis, but there were other features affecting the results as well. The sentiment lexicon used plays a key role, and pre-processing techniques like stemming could affect the performance too. A takeaway from this study is that emojis contain important sentiment information, and should not be disregarded. Furthermore, emojis are useful when analyzing texts, if there is a lack of linguistic resources for the language in question.

Communication is one of the most demanding andimportant parts of effective software development.Furthermore, the effectiveness of software developmentcommunication can be measured with the three collaborativeinterpersonal problem-solving conversation dimensions:Active Discussion, Creative Conflict, and ConversationManagement.Previous work that utilized these dimensions to analyzecommunication relied on manually labeling thecommunication, a process that is time-consuming and notapplicable to real-time use.In this study, natural language processing and supervisedmachine learning were investigated for the automaticclassification and measurement of collaborativeinterpersonal problem-solving conversation dimensions intranscribed software development communication. Thisapproach enables the evaluation of communication andprovides suggestions to improve software developmentefficiency.To determine the optimal classification approach, this workexamined nine different classifiers. It was determined thatthe classifier that scored the highest was Random Forest,followed by Decision Tree and SVM.Random Forest managed to achieve accuracy, precision, andrecall up to 93.66%, 93.76%, and 93.63%, respectively whentrained and tested with stratified 10-fold cross-validation.

Information retrieval systems have come to change how users interact with computerized systems and locate information. A major challenge when designing these systems is how to handle the vocabulary mismatch problem, i.e. that users, when formulating queries, pick different words than those present in the relevant documents that should be retrieved. With recent advances in artificial intelligence and the emergence of transformer-based language models, new methods have been proposed to alleviate this problem. One such method is the usage of document expansion models which append words to each document that are likely to be part of users’ queries. As previous research on document expansion models has been focused on English-language applications, this thesis investigates the effectiveness of one such model for Swedish applications. Although no improvement was found when using this method, the result is likely to be a consequence of dataset quality and domain rather than the method itself. / Informationssökningssystem har förändrat hur användare interagerar med datorsystem och lokaliserar information. En betydande utmaning när dessa system designas är hur det s.k. ”vocabulary mismatch”-problemet ska hanteras, d.v.s. att användare väljer andra söktermer än de som förekommer i de relevanta dokumenten som söksystemet ska hitta. Nya framsteg inom artificiell intelligens och utvecklingen av transformer-baserade språkmodeller har lett till att nya metoder har föreslagits för att mildra det här problemet. En sådan metod är att använda dokumentexpansionsmodeller som lägger till ord till varje dokument som är sannolika att förekomma som söktermer. Då tidigare forskning på dokumentexpansionsmodeller har fokuserat på engelskspråkiga tillämpningar fokuserar det här arbetet i stället på hur väl sådana modeller fungerar för svenskspråkiga tillämpningar. Även om ingen förbättring observerades när denna metod tillämpades är resultatet sannolikt en konsekvens av datamängdens kvalitet och domän snarare än metoden i sig.

Toxic comments are overall likely to make someone feel uncomfortable and leave a discussion and are therefore potentially problematic. Toxic comments occur online on various social media, and depending on the site, get detected manually or via machine learning algorithms (or both), and removed depending on the severity and other factors. The problem is the lack of research on toxic comments on Swedish YouTube channels, meaning that content creators, especially new ones, will be unfamiliar with and unprepared for these toxic comments. We aim to expand research in this area by finding out not only the proportion of comments on Swedish YouTube channels that are toxic, but what type of toxic comments occur, and what types are the most common. A Survey of documents was the chosen research strategy, and mixed methods were used as well, by combining qualitative and quantitative data analysis, with more focus on the quantitative aspect. A random sample of 79 577 YouTube comments was collected as data, and the machine learning program Hatescan was used to generate a toxicity score for each comment, allowing us to sort these comments based on score, and sample to manually analyze the type of toxicity of these comments. The results show that 0.643% of the total comments analyzed were toxic. It was found that most of the toxic comments are directed toward someone from the video. Toxic comments in the form of personal insults, and toxic comments about someone’s intelligence/competence were by far the most common.

In today’s world, it is not enough for companies to consider any one social media channel in isolation. Instead, they must provide their customers with a unified experience across channels and consider interdependencies between channels. Most marketing research that examines user generated content is focused on a single channel and is limited to the English language. This thesis analyses Swedish language content collected from eight different social media platforms: Facebook, YouTube, Instagram, TikTok, Twitter, Tripadvisor, Trustpilot, and Google Reviews. The platforms were compared pairwise by the prevalence of positive, negative, and neutral sentiment in comments and reviews about the theme park Liseberg. The sentiment was predicted using a lexical approach where each word in a wordlist was assigned a weight to denote positive or negative sentiment associated with the word. The study found that there is a statistically significant difference between the positivity, negativity, and neutrality expressed by users on the different social media channels. There was no difference in sentiment between YouTube and Instagram comments, but there were differences in at least one of the three sentiment categories for all other pairwise comparisons of platforms. Having an understanding of the attitudes towards the brand in different channels can support marketers in determining their optimal mix of social media channels. These results are also of interest to researchers who should take the differences between social media platforms into consideration when designing studies around user generated content.

Online sexual solicitors are individuals who deceptively earn the trust of minors online with the goal of eventual sexual gratification. Despite the prevalence of online solicitation, conversations in the domain are difficult to acquire due to the sensitive nature of the data. As a result, researchers studying online solicitors often study conversations between solicitors and decoys which are publicly available online. However, researchers have begun to believe such conversations are not representative of solicitor-victim conversations. Decoys and law enforcement are restricted in that they are unable to initiate contact, suggest meeting, or begin sexual conversations with an offender. Additionally decoys and law enforcement officers both have a goal of gathering evidence which means they often respond positively in contexts which would normally be considered awkward or inappropriate. Multiple researchers have suggested differences may exist between offender-victim and offender-decoy conversations and yet little research has sought to identify the differences and similarities between those talking to solicitors. In this study, the author identifies differences between decoys, officers, and victims within the manipulative process used by online solicitors to entrap victims which is known as grooming. The author looks at differences which occur within grooming stages and strategies within the grooming stages. The research in this study has implications for the data choices of future researchers in this domain. Additionally, this research may be used to inform the training process of officers who will engage in online sex stings.

This dissertation outlines the development of a signal-to-text system that incorporates spatial relationship information to generate scene descriptions. Existing signal-to-text systems generate accurate descriptions in regards to information contained in an image. However, to date, no signalto- text system incorporates spatial relationship information. A survey of related work in the fields of object detection, signal-to-text, and spatial relationships in images is presented first. Three methodologies followed by evaluations were conducted in order to create the signal-to-text system: 1) generation of object localization results from a set of input images, 2) derivation of Level One Summaries from an input image, and 3) inference of Level Two Summaries from the derived Level One Summaries. Validation processes are described for the second and third evaluations, as the first evaluation has been previously validated in the related original works. The goal of this research is to show that a signal-to-text system that incorporates spatial information results in more informative descriptions of the content contained in an image. An additional goal of this research is to demonstrate the signal-to-text system can be easily applied to additional data sets, other than the sets used to train the system, and achieve similar results to the training sets. To achieve this goal, a validation study was conducted and is presented to the reader.

No description available.