Information Retrieval (IR) is finding content of an unstructured nature with respect to an information need. A retrieval system typically uses a retrieval model to rank the available content by their estimated relevance to an information need. For decades, state-of-the-art retrieval models have used the assumption that terms appear independently in text documents. Chapter 1 of this thesis describes how the relevance likelihood of a document changes by the observed distance between co-occurring query terms in its text. Nowadays, news is abundantly available online, allowing users to discover and follow news events. However, online news is often very redundant; most sources basing their stories on previously published works and add only limited new information. Thus, a user often ends up spending significant amount of effort re-reading the same parts of a story before finding relevant and novel information. In Chapter 2 and Chapter 3, we present a novel approach to construct an online news summary for a given topic. Salient sentences are identified by clustering the sentences in the news stream based on the relative proximity of the sentences and the temporal proximity of their publication times. To improve the coherence of a long summary that describes a news topic, we propose to automatically cluster sentences by subtopics in Chapter 4. In Chapter 5, we show how new topics can be detected in the news stream using the same clustering technique. In real-life decision making, people are often faced with an overload of choices. A recommender system aids the user by reducing the available choices to a shortlist of items that are of interest to the user. In Chapter 6, we learn high-dimensional representations for movies that allow to effectively recommend movies based on a user’s most recently rated movies.@en

First Story Detection (FSD) systems aim to identify those news articles that discuss an event that was not reported before. Recent work on FSD has focussed almost exclusively on efficiently detecting documents that are dissimilar from their nearest neighbor. We propose a novel FSD approach that is more effective, by adapting a recently proposed method for news summarization based on 3-nearest neighbor clustering. We show that this approach is more effective than a baseline that uses dissimilarity of an individual document from its nearest neighbor.@en

Recommender systems leverage both content and user interactions to generate recommendations that fit users' preferences. The recent surge of interest in deep learning presents new opportunities for exploiting these two sources of information. To recommend items we propose to first learn a user-independent high-dimensional semantic space in which items are positioned according to their substitutability, and then learn a user-specific transformation function to transform this space into a ranking according to the user's past preferences. An advantage of the proposed architecture is that it can be used to effectively recommend items using either content that describes the items or user-item ratings. We show that this approach significantly outperforms state-of-the-art recommender systems on the MovieLens 1M dataset.@en