Speaker: Jacques Savoy, University of Neuchâtel, Switzerland
Title: Text Categorization with Style [Slides]
Abstract: Text categorization (determining the predefined label(s) for a given text) is an important task in various NLP applications. Automatic indexing is a well-known example, as well as sentiment analysis. However, other applications are using, so to speak, only functional words (or terms belonging to a stopword list) without any or important meaning. Such words, viewed as noise in an IR system, reflect the style of the author.
Based on such stylistic items, one can detect the true author of a text excerpt (e.g., blog, threating e-mail, malicious code, or “who is Elena Ferrante?”), or to define some information about the profile of the author such as to know whether the document was written by a man or a woman, his/her age, the time period, or some of his/her psychological features. Recently, the detection of fake texts (or news) is receiving more attention. To illustrate some of these aspects, speeches uttered by US presidents or during the US presidential election will be used (with a focus on the last election).
Speaker: Paolo Boldi, University of Milan, Italy
Title: Feature-Rich Networks: Models and Applications [Slides]
Abstract: We shall discuss complex networks whose nodes are endowed with binary features. Such a situation is truly ubiquitous: real-world networks describe connections between real-world objects, that present properties and attributes, and links arise from the properties of the nodes they connect, albeit often in a non-trivial way. We describe a simple but powerful and general framework where this kind of feature-rich networks can be handled. From the general framework, we deduce a generative model that is inspired to Indian Buffet processes with a further fitness parameter that allows more flexibility. Also, we show that the framework allows one to study feature relations, and we show an example using Wikipedia and its categories.
Speaker: Francesco Ricci, Free University of Bozen-Bolzano, Italy
Title: Recommender Systems: Research Challenges [Slides]
Abstract: Recommender systems are information search and filtering tools that provide suggestions for items to be of use to a user. They have become common in a large number of Internet applications (YouTube, Amazon, LinkedIn), helping users to make better choices while browsing and searching for news, music, vacations, or financial investments. Recommender systems exploit data mining and information retrieval techniques to predict to what extent an item suits the user needs and wants and is an optimal choice for the user. In the talk, recommender systems ideas and techniques will be introduced and criticised. We will illustrate the fundamental steps that are required to build and use a recommender system and their underlying assumptions. Most of the talk will be dedicated to discuss some limitations and open challenges for recommender systems, such as preference modelling, choice modelling and the dynamics of the data generated and consumed by recommender systems.