Thesis (Ed.D.)--Boston University

La thèse concerne la transformation de données textuelles non structurées en données structurées et exploitables par des systèmes de recommandation. Deux grandes catégories d'informations sont utilisées dans le domaine des moteurs de recommandation : les données descriptives de contenus comme les méta-données ou les tags (filtrage thématique), et les données d'usages qui peuvent être des notes ou encore des pages Web visitées par exemple (filtrage collaboratif). D'autres données sont présentes sur le Web et ne sont pas encore réellement exploitées. Avec l'émergence du Web 2.0, les internautes sont de plus en plus amenés à partager leurs sentiments, opinions, expériences sur des produits, personnalités, films, musiques, etc. Les données textuelles produites par les utilisateurs représentent potentiellement des sources riches d'informations qui peuvent être complémentaires des données exploitées actuellement par les moteurs de recommandation et peuvent donc ouvrir de nouvelles voies d'études dans ce domaine en plein essor. Notre objectif dans le cadre de la thèse est de produire, à partir de commentaires issus de sites communautaires (blogs ou forums), des matrices d'entrées pertinentes pour les systèmes de recommandation. L'idée sous-jacente est de pouvoir enrichir un système pour un service débutant, qui possède encore peu d'utilisateurs propres, et donc peu de données d'usages, par des données issues d'autres utilisateurs. Nous faisons tout d'abord un état de l'art de la recommandation automatique. Nous présentons ensuite le moteur ainsi que les données utilisées pour les expérimentations. Le chapitre suivant décrit les premières expérimentations en mode thématique. Nous faisons ensuite un nouvel état de l'art sur la classification d'opinion. Pour finir, nous décrivons les expérimentations menées pour l'approche collaborative à l'aide de la classification d'opinion. / The thesis is about the transformation of unstructured textual data in structured data in order to be used by a recommender system. Recommender systems can operate on two main types of data: content descriptors as metadata or tags (content-based filtering), and usage data as rates or visited Web pages for example (collaborative filtering). Other data exist on the Web which are not used yet. With the emergence of the Web 2.0, users share their feelings, opinions, experiences on products, personalities, movies, music, etc. (through comments for example). This textual data generated by users potentially represent rich sources of information which can supplement data exploited by recommender systems. The exploitation of this kind of data could open new paths in this burgeoning field. Our objective in this thesis is to generate matrices relevant for recommender systems. The underlying idea is to enrich a system for a beginner service, which has still few own users, then too little usage data, by information on other users on the Web. The thesis begins with a state of the art of automatic recommendation. Then, we present the recommender systems and the textual corpus used for experiments. The next chapter presents first experiments with the content-based filtering approach. The next part contains the state of the art of opinion mining. Finally, we describe experiments done with collaborative filtering approach using opinion classification.

No description available.

This work is concerned with the automatic understanding of evaluative text. We investigate sentence level opinion polarity prediction by assigning lexical polarities and deriving sentence polarity from these with the use of contextual valence shifters. A methodology for iterative failure analysis is developed and used to refine our lexicon and identify new contextual shifters. Algorithms are presented that employ these new shifters to improve sentence polarity prediction accuracy beyond that of a state-of-the-art existing algorithm in the domain of consumer product reviews. We then apply the best configuration of our algorithm to the domain of movie reviews. / Science, Faculty of / Computer Science, Department of / Graduate

Opinion dynamics is extensively used in studying large-scale social, economical, political and natural phenomena that involve many interacting agents. It also can be used to model the evolution of teams of autonomous vehicles operating in a coordinated fashion with civilian and military applications, when arbitration among individual goals needs to be negotiated. Recently, research was conducted on how opinion dynamics can be the core of collective decision-making mechanisms for swarm robotics. Opinion dynamics with a time varying opinion space, which is the set of all possible opinions an agent may have, is a relatively recent research topic. In this work, the Deffuant-Weisbuch model (DW model), which allows to model opinion dynamics in shrinking opinion spaces, was applied. In simulating this class of systems and in extracting information from them it is crucial to establish reliable algorithms and criteria for counting the numbers of clusters, as this ultimately affects the determination of the steady state of the system. A method was applied to combine Fuzzy c-means clustering and subtractive clustering to check convergence of the system and avoid negative influence of outliers. Different scenarios are simulated to study the influence of characteristic parameters on the formation of opinions, which is quantified by the formation of clusters in the opinion space. Additionally, we simulate the scenario of a two dimensional opinion space in which one side shrinks, and evaluate how the rate of shrinking influences the steady state opinion space. This is a simplified model to gain some insight on the effect of extreme changes of opinions in multi-dimensional opinion space.

Indiana University-Purdue University Indianapolis (IUPUI) / What do Americans think about the United Nations? Social scientists have put forward an array of viewpoints on the subject, focusing on such things as partisan differences in the attitudes of Americans towards the work of the UN to public skepticism of the organization’s objectives and effectiveness. I argue, in this thesis, that public opinion is a causal factor because of its potential to influence political outcomes. For example, public opinion can influence the effectiveness of the UN’s work in three main ways: 1) It can serve as an instrument for establishing the credibility of the international body’s work and/ or in discrediting the system as a whole; 2) it can serve as a link between the US and the UN in encouraging support for the United Nations in one of its most important members; and 3) Public opinion can stress the importance of a particular issue and pressure influential actors to take action. I choose to focus solely on the United States in this thesis despite the UN’s 192 other member states for the reason that overwhelmingly negative assessments have been offered of the organization since the Iraq War. The research depicting this idea points to a decline in American popular support for the UN in the past decade. By investigating six different hypotheses which seek to explain this possible decline, I conclude that American public support for the international body after the Iraq war has declined and can best be explained by hypothesis 3 on inadequate coverage of UN matters in the media and hypothesis 5 on the thought that the UN is “ineffective;” although this presumed decline is not steady due to opinion level variations in the recent decade.

No description available.

No description available.

Call number: LD2668 .T4 1960 E48

Call number: LD2668 .T4 1978 M85 / Master of Science