Do Happy Meals really make us glad? Do salads make us blue? Is cake our comfort? FoodMood is an i... more Do Happy Meals really make us glad? Do salads make us blue? Is cake our comfort? FoodMood is an interactive data visualisation project that gives citizens a rare opportunity to engage and reflect, acknowledge, and understand the connection between emotion and food. The project explores the opportunities presented by the data-sharing world of today by using global English-language tweets about food and sentiment analysis. FoodMood aims to gain a better understanding of global food consumption patterns and its impact on the daily emotional wellbeing of people against the backdrop of country data such as Gross Domestic Product (GDP) and obesity levels. The constantly evolving findings indicate: Meat enjoys a high sentiment rating and is vastly consumed the world over; fast-food companies dominate the food consumption landscapes of most countries, although not all of them enjoy equal sentiment ratings across countries. In some instances countries with low GDP showed higher sentiment towards food than their wealthier neighbours. Ultimately, FoodMood helps reveal a hidden layer of digital, social, and cultural data that pushes the boundaries of awareness and understanding of food.
Today, methods for automatic opinion mining on online data are becoming increasingly relevant. Ov... more Today, methods for automatic opinion mining on online data are becoming increasingly relevant. Over the past few years, methods have been developed that can successfully and with a great degree of accuracy analyze the sentiment in opinions from digital text. These developments enable research into prediction of sentiment. Sentiment prediction has traditionally been used as a tool for stock prediction. In such scenarios, incoming news is analyzed in real-time and the impact of that news on stock prices is estimated, making automatic stock trading possible. Recent developments in sentiment prediction have seen attempts to predict explicit sentiment of the reactions to blogs, before the blogs are even posted. In this thesis, we research the prediction of the general sentiment polarity in reactions to news articles, before the news articles are posted. We use Reddit as our data source for news and comments, and approach the sentiment prediction problem using machine learning methods. To automatically label Reddit comments for sentiment prediction training, we perform automatic domain-knowledge transfer from a classifier trained on Twitter to Reddit. In this thesis, we propose a new machine learning method, a new feature selection method for text and a new machine learning evaluation metric. We provide a thorough analysis of Reddit data, and manually annotate a gold standard from it. Finally, we demonstrate the feasibility of sentiment prediction of the general sentiment polarity in reactions to news articles, before the news articles are posted in limited cases. Ultimately, we provide an analysis of the limitations of our and similar approaches to sentiment prediction, and make recommendations for future research.
Recent advances in computing power have brought the potential for real-time applications of image... more Recent advances in computing power have brought the potential for real-time applications of image and facial recognition from the laboratory to the every-day reality. One of the possible uses of these methods is in the field of Nomadic Information, allowing for generation of media feedback (e.g. displaying advertisements) based on subject's individual or general profiles as observed through an imaging system. While different methods of extracting the features that make up the profiles exist, most of them are extremely specific (e.g. "gender" only) or not very robust. This paper presents a generalized, robust and scalable method of facial feature extraction based on two-dimensional Hidden Markov Models [2], extended by addition of a domain context to the model. To validate this method, classification results are compared with the results obtained by specialized classifiers from literature under controlled and uncontrolled conditions.
Do Happy Meals really make us glad? Do salads make us blue? Is cake our comfort? FoodMood is an i... more Do Happy Meals really make us glad? Do salads make us blue? Is cake our comfort? FoodMood is an interactive data visualisation project that gives citizens a rare opportunity to engage and reflect, acknowledge, and understand the connection between emotion and food. The project explores the opportunities presented by the data-sharing world of today by using global English-language tweets about food and sentiment analysis. FoodMood aims to gain a better understanding of global food consumption patterns and its impact on the daily emotional wellbeing of people against the backdrop of country data such as Gross Domestic Product (GDP) and obesity levels. The constantly evolving findings indicate: Meat enjoys a high sentiment rating and is vastly consumed the world over; fast-food companies dominate the food consumption landscapes of most countries, although not all of them enjoy equal sentiment ratings across countries. In some instances countries with low GDP showed higher sentiment towards food than their wealthier neighbours. Ultimately, FoodMood helps reveal a hidden layer of digital, social, and cultural data that pushes the boundaries of awareness and understanding of food.
Today, methods for automatic opinion mining on online data are becoming increasingly relevant. Ov... more Today, methods for automatic opinion mining on online data are becoming increasingly relevant. Over the past few years, methods have been developed that can successfully and with a great degree of accuracy analyze the sentiment in opinions from digital text. These developments enable research into prediction of sentiment. Sentiment prediction has traditionally been used as a tool for stock prediction. In such scenarios, incoming news is analyzed in real-time and the impact of that news on stock prices is estimated, making automatic stock trading possible. Recent developments in sentiment prediction have seen attempts to predict explicit sentiment of the reactions to blogs, before the blogs are even posted. In this thesis, we research the prediction of the general sentiment polarity in reactions to news articles, before the news articles are posted. We use Reddit as our data source for news and comments, and approach the sentiment prediction problem using machine learning methods. To automatically label Reddit comments for sentiment prediction training, we perform automatic domain-knowledge transfer from a classifier trained on Twitter to Reddit. In this thesis, we propose a new machine learning method, a new feature selection method for text and a new machine learning evaluation metric. We provide a thorough analysis of Reddit data, and manually annotate a gold standard from it. Finally, we demonstrate the feasibility of sentiment prediction of the general sentiment polarity in reactions to news articles, before the news articles are posted in limited cases. Ultimately, we provide an analysis of the limitations of our and similar approaches to sentiment prediction, and make recommendations for future research.
Recent advances in computing power have brought the potential for real-time applications of image... more Recent advances in computing power have brought the potential for real-time applications of image and facial recognition from the laboratory to the every-day reality. One of the possible uses of these methods is in the field of Nomadic Information, allowing for generation of media feedback (e.g. displaying advertisements) based on subject's individual or general profiles as observed through an imaging system. While different methods of extracting the features that make up the profiles exist, most of them are extremely specific (e.g. "gender" only) or not very robust. This paper presents a generalized, robust and scalable method of facial feature extraction based on two-dimensional Hidden Markov Models [2], extended by addition of a domain context to the model. To validate this method, classification results are compared with the results obtained by specialized classifiers from literature under controlled and uncontrolled conditions.
Uploads
Papers by Bruno Jakic