Departamento de Informática (Department of Informatics)

Textual Entailment consists in determining if an entailment relation exists between two texts. In this paper, we present an Informative Asymmetric Measure called the Asymmetric InfoSimba (AIS ), which we combine with different asym-metric association measures to recognize the specific case of Textual Entailment by Generality. In particular, the AIS proposes na unsupervised, language-independent, threshold free solution. This new measure is tested against the first Recognizing Textual Entailment dataset for an exhaustive number of asymmetric association measures and shows that the combination of the AIS with the Braun-Blanket steadily improves results against competitive measures.

An Intelligent Personal Assistant (IPA) is an agent that has the purpose of helping the user with his daily tasks. This paper is focused on IPAs for Internet of Things (IoT) environments. In this sense, a good IPA has the capability of surveying his user behaviour and suggest tasks or make decisions with the intention of simplifying the user interaction with his surroundings. With this in mind, this paper focuses on studying the accuracy of various classifiers, with the objective of finding the one that suits better the needs of an IPA for IoT. The aim is to test each algorithm with a dataset of events, that relate to past behaviours of the user, and find if there is an opportunity to notify the user that he/she may want to take an action or create an automation based on the learned behaviour.

The shift in human computer interaction from desktop computing to mobile interaction highly influences the needs for new designed interfaces. In this paper, we address the issue of searching for information on mobile devices, an area also known as Mobile Information Retrieval. In particular, we propose to summarize as much as possible the information retrieved by any search engine to allow universal access to information.

In this paper, we consider the problem of building models that have high sentiment classification accuracy across domains. For that purpose, we present and evaluate three new algorithms based on multi-view learning using both high-level and low-level views. Our experimental results present accuracy levels of 80% with two views, combining SVM classifiers over high-level features and unigrams compared to 77.1% for the state-of-the-art SAR algorithm [7].

Textual Entailment consists in determining if an entailment relation exists between two texts. In this paper, we present an Informative Asymmetric Measure called the Simplified Asymmetric InfoSimba (AISs), which we combine with different asymmetric association measures to recognize the specific case of Textual Entailment by Generality. In particular, the AISs proposes an unsupervised, languageindependent, threshold free solution. This new measure is tested against the first Recognizing Textual Entailment data set for an exhaustive number of asymmetric association measures and shows that the combination of the AISs with the Braun-Blanket steadily improves results.

In this paper, we consider the problem of building models that have high sentiment classification accuracy across domains. For that purpose, we present and evaluate three new algorithms based on multi-view learning using both high-level and low-level views, which show improved results compared to the state-of-the-art SAR algorithm [1] over cross-domain text subjectivity classification. Our experimental results present accuracy levels of 80% with two views, combining SVM classifiers over high-level features and unigrams compared to 77.1% for the SAR algorithm.

Textual Entailment consists in determining if an entailment relation exists between two texts. In this paper, we presente an Informative Asymmetric Measure called the Asymmetric InfoSimba (AIS ), which we combine with different asymmetric association measures to recognize the specific case of Textual Entailment by Generality. In particular, the AIS proposes na unsupervised, language-independent, threshold free solution. This new measure is tested against the first Recognizing Textual Entailment dataset for an exhaustive number of asymmetric association measures and shows that the combination of the AIS with the Braun-Blanket steadily improves results against competitive measures.

In the context of Ephemeral Clustering of web Pages, it can be interesting to label each cluster with a small summary instead of just a label. Within this scope, we introduce the paradigm of Textual Entailment by Generality, which can be defined as the entailment from a specific web snippet towards a more general web snippet. The subjacent idea is to find the best web snippet, which summarizes and subsumes all the other web snippets within an ephemeral cluster. To reach this objective, we first propose a new informative asymmetric similarity measure called the Simplified Asymmetric InfoSimba (AISs), which can be combined with different asymmetric association measures. In particular, the AISs proposes na unsupervised language-independent solution to infer Textual Entailment by Generality and as such can help to encounter the web snippet with maximum semantic coverage. This new methodology is tested against the first Recognizing Textual Entailment data set (RTE-1)1 for an exhaustive number of asymmetric association measures with and without the identification of Multiword Units. The comparative experiments with existing state-of-the-art methodologies show promising results.

The adaptation paradigm is a basic defining feature for pervasive computing systems. Adaptation systems must work
efficiently in smart environment while providing suitable information relevant to the user system interaction. The key objective is to deduce the information needed information changes. Therefore relying on fixed operational models would be inappropriate. This paper presents a study on developing a Intelligent Personal Assistant to assist the user in interacting with their Smart Environment. We propose a Unsupervised and Language-Independent Adaptation through Intelligent Speech Interface and a set of methods of Acquiring Knowledge, namely Semantic Similarity and Unsupervised Learning.