An approach on improving search engines through social content recommendation

Internet growth has provoked that information search had come to have one of the most relevant roles in the industry and to be one of the most current topics in research environments. Internet is the largest information container in history and its facility to generate new information leads to new challenges when talking about retrieving information and discern which one is more relevant than the rest. Parallel to the information growth in quantity, the way information is provided has also changed. One of these changes that has provoked more information traffic has been the emergence of social networks. We have seen how social networks can provoke more traffic than search engines themselves. We can draw conclusions that allow us to take a new approach to the information retrieval problem. Public trusts the most information coming from known contacts. In this document we will explore a possible change in classic search engines to bring them closer to the social side and adquire those social advantages.

Algorithmes d’apprentissage profonds supervisés et non-supervisés: applications et résultats théoriques

La liste des domaines touchés par l’apprentissage machine s’allonge rapidement. Au fur et à mesure que la quantité de données disponibles augmente, le développement d’algorithmes d’apprentissage de plus en plus puissants est crucial. Ce mémoire est constitué de trois parties: d’abord un survol des concepts de bases de l’apprentissage automatique et les détails nécessaires pour l’entraînement de réseaux de neurones, modèles qui se livrent bien à des architectures profondes. Ensuite, le premier article présente une application de l’apprentissage machine aux jeux vidéos, puis une méthode de mesure performance pour ceux-ci en tant que politique de décision. Finalement, le deuxième article présente des résultats théoriques concernant l’entraînement d’architectures profondes nonsupervisées. Les jeux vidéos sont un domaine particulièrement fertile pour l’apprentissage automatique: il estf facile d’accumuler d’importantes quantités de données, et les applications ne manquent pas. La formation d’équipes selon un critère donné est une tˆache commune pour les jeux en lignes. Le premier article compare différents algorithmes d’apprentissage à des réseaux de neurones profonds appliqués à la prédiction de la balance d’un match. Ensuite nous présentons une méthode par simulation pour évaluer les modèles ainsi obtenus utilisés dans le cadre d’une politique de décision en ligne. Dans un deuxième temps nous présentons une nouvelleméthode pour entraîner des modèles génératifs. Des résultats théoriques nous indiquent qu’il est possible d’entraîner par rétropropagation des modèles non-supervisés pouvant générer des échantillons qui suivent la distribution des données. Ceci est un résultat pertinent dans le cadre de la récente littérature scientifique investiguant les propriétés des autoencodeurs comme modèles génératifs. Ces résultats sont supportés avec des expériences qualitatives préliminaires ainsi que quelques résultats quantitatifs.

Processamento e análise de vídeos utilizando Floresta de Caminhos Ótimos

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

Processamento e análise de vídeos utilizando Floresta de Caminhos Ótimos

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

Arquitectura y modelo de referencia para un ecosistema de intercambio de datos personales centrado en el usuario

Los servicios telemáticos han transformando la mayoría de nuestras actividades cotidianas y ofrecen oportunidades sin precedentes con características como, por ejemplo, el acceso ubicuo, la disponibilidad permanente, la independencia del dispositivo utilizado, la multimodalidad o la gratuidad, entre otros. No obstante, los beneficios que destacan en cuanto se reflexiona sobre estos servicios, tienen como contrapartida una serie de riesgos y amenazas no tan obvios, ya que éstos se nutren de y tratan con datos personales, lo cual suscita dudas respecto a la privacidad de las personas. Actualmente, las personas que asumen el rol de usuarios de servicios telemáticos generan constantemente datos digitales en distintos proveedores. Estos datos reflejan parte de su intimidad, de sus características particulares, preferencias, intereses, relaciones sociales, hábitos de consumo, etc. y lo que es más controvertido, toda esta información se encuentra bajo la custodia de distintos proveedores que pueden utilizarla más allá de las necesidades y el control del usuario. Los datos personales y, en particular, el conocimiento sobre los usuarios que se puede extraer a partir de éstos (modelos de usuario) se han convertido en un nuevo activo económico para los proveedores de servicios. De este modo, estos recursos se pueden utilizar para ofrecer servicios centrados en el usuario basados, por ejemplo, en la recomendación de contenidos, la personalización de productos o la predicción de su comportamiento, lo cual permite a los proveedores conectar con los usuarios, mantenerlos, involucrarlos y en definitiva, fidelizarlos para garantizar el éxito de un modelo de negocio. Sin embargo, dichos recursos también pueden utilizarse para establecer otros modelos de negocio que van más allá de su procesamiento y aplicación individual por parte de un proveedor y que se basan en su comercialización y compartición con otras entidades. Bajo esta perspectiva, los usuarios sufren una falta de control sobre los datos que les refieren, ya que esto depende de la voluntad y las condiciones impuestas por los proveedores de servicios, lo cual implica que habitualmente deban enfrentarse ante la disyuntiva de ceder sus datos personales o no acceder a los servicios telemáticos ofrecidos. Desde el sector público se trata de tomar medidas que protejan a los usuarios con iniciativas y legislaciones que velen por su privacidad y que aumenten el control sobre sus datos personales, a la vez que debe favorecer el desarrollo económico propiciado por estos proveedores de servicios. En este contexto, esta tesis doctoral propone una arquitectura y modelo de referencia para un ecosistema de intercambio de datos personales centrado en el usuario que promueve la creación, compartición y utilización de datos personales y modelos de usuario entre distintos proveedores, al mismo tiempo que ofrece a los usuarios las herramientas necesarias para ejercer su control en cuanto a la cesión y uso de sus recursos personales y obtener, en su caso, distintos incentivos o contraprestaciones económicas. Las contribuciones originales de la tesis son la especificación y diseño de una arquitectura que se apoya en un proceso de modelado distribuido que se ha definido en el marco de esta investigación. Éste se basa en el aprovechamiento de recursos que distintas entidades (fuentes de datos) ofrecen para generar modelos de usuario enriquecidos que cubren las necesidades específicas de terceras entidades, considerando la participación del usuario y el control sobre sus recursos personales (datos y modelos de usuario). Lo anterior ha requerido identificar y caracterizar las fuentes de datos con potencial de abastecer al ecosistema, determinar distintos patrones para la generación de modelos de usuario a partir de datos personales distribuidos y heterogéneos y establecer una infraestructura para la gestión de identidad y privacidad que permita a los usuarios expresar sus preferencias e intereses respecto al uso y compartición de sus recursos personales. Además, se ha definido un modelo de negocio de referencia que sustenta las investigaciones realizadas y que ha sido particularizado en dos ámbitos de aplicación principales, en concreto, el sector de publicidad en redes sociales y el sector financiero para la implantación de nuevos servicios. Finalmente, cabe destacar que las contribuciones de esta tesis han sido validadas en el contexto de distintos proyectos de investigación industrial aplicada y también en el marco de proyectos fin de carrera que la autora ha tutelado o en los que ha colaborado. Los resultados obtenidos han originado distintos méritos de investigación como dos patentes en explotación, la publicación de un artículo en una revista con índice de impacto y diversos artículos en congresos internacionales de relevancia. Algunos de éstos han sido galardonados con premios de distintas instituciones, así como en las conferencias donde han sido presentados. ABSTRACT Information society services have changed most of our daily activities, offering unprecedented opportunities with certain characteristics, such as: ubiquitous access, permanent availability, device independence, multimodality and free-of-charge services, among others. However, all the positive aspects that emerge when thinking about these services have as counterpart not-so-obvious threats and risks, because they feed from and use personal data, thus creating concerns about peoples’ privacy. Nowadays, people that play the role of user of services are constantly generating digital data in different service providers. These data reflect part of their intimacy, particular characteristics, preferences, interests, relationships, consumer behavior, etc. Controversy arises because this personal information is stored and kept by the mentioned providers that can use it beyond the user needs and control. Personal data and, in particular, the knowledge about the user that can be obtained from them (user models) have turned into a new economic asset for the service providers. In this way, these data and models can be used to offer user centric services based, for example, in content recommendation, tailored-products or user behavior, all of which allows connecting with the users, keeping them more engaged and involved with the provider, finally reaching customer loyalty in order to guarantee the success of a business model. However, these resources can be used to establish a different kind of business model; one that does not only processes and individually applies personal data, but also shares and trades these data with other entities. From that perspective, the users lack control over their referred data, because it depends from the conditions imposed by the service providers. The consequence is that the users often face the following dilemma: either giving up their personal data or not using the offered services. The Public Sector takes actions in order to protect the users approving, for example, laws and legal initiatives that reinforce privacy and increase control over personal data, while at the same time the authorities are also key players in the economy development that derives from the information society services. In this context, this PhD Dissertation proposes an architecture and reference model to achieve a user-centric personal data ecosystem that promotes the creation, sharing and use of personal data and user models among different providers, while offering users the tools to control who can access which data and why and if applicable, to obtain different incentives. The original contributions obtained are the specification and design of an architecture that supports a distributed user modelling process defined by this research. This process is based on leveraging scattered resources of heterogeneous entities (data sources) to generate on-demand enriched user models that fulfill individual business needs of third entities, considering the involvement of users and the control over their personal resources (data and user models). This has required identifying and characterizing data sources with potential for supplying resources, defining different generation patterns to produce user models from scattered and heterogeneous data, and establishing identity and privacy management infrastructures that allow users to set their privacy preferences regarding the use and sharing of their resources. Moreover, it has also been proposed a reference business model that supports the aforementioned architecture and this has been studied for two application fields: social networks advertising and new financial services. Finally, it has to be emphasized that the contributions obtained in this dissertation have been validated in the context of several national research projects and master thesis that the author has directed or has collaborated with. Furthermore, these contributions have produced different scientific results such as two patents and different publications in relevant international conferences and one magazine. Some of them have been awarded with different prizes.

Social Recommendation Based on Content and Trust

Report published in the Proceedings of the National Conference on "Education and Research in the Information Society", Plovdiv, May, 2014

User behaviour modelling in a multi-dimensional environment for personalization and recommendation

Handling information overload online, from the user's point of view is a big challenge, especially when the number of websites is growing rapidly due to growth in e-commerce and other related activities. Personalization based on user needs is the key to solving the problem of information overload. Personalization methods help in identifying relevant information, which may be liked by a user. User profile and object profile are the important elements of a personalization system. When creating user and object profiles, most of the existing methods adopt two-dimensional similarity methods based on vector or matrix models in order to find inter-user and inter-object similarity. Moreover, for recommending similar objects to users, personalization systems use the users-users, items-items and users-items similarity measures. In most cases similarity measures such as Euclidian, Manhattan, cosine and many others based on vector or matrix methods are used to find the similarities. Web logs are high-dimensional datasets, consisting of multiple users, multiple searches with many attributes to each. Two-dimensional data analysis methods may often overlook latent relationships that may exist between users and items. In contrast to other studies, this thesis utilises tensors, the high-dimensional data models, to build user and object profiles and to find the inter-relationships between users-users and users-items. To create an improved personalized Web system, this thesis proposes to build three types of profiles: individual user, group users and object profiles utilising decomposition factors of tensor data models. A hybrid recommendation approach utilising group profiles (forming the basis of a collaborative filtering method) and object profiles (forming the basis of a content-based method) in conjunction with individual user profiles (forming the basis of a model based approach) is proposed for making effective recommendations. A tensor-based clustering method is proposed that utilises the outcomes of popular tensor decomposition techniques such as PARAFAC, Tucker and HOSVD to group similar instances. An individual user profile, showing the user's highest interest, is represented by the top dimension values, extracted from the component matrix obtained after tensor decomposition. A group profile, showing similar users and their highest interest, is built by clustering similar users based on tensor decomposed values. A group profile is represented by the top association rules (containing various unique object combinations) that are derived from the searches made by the users of the cluster. An object profile is created to represent similar objects clustered on the basis of their similarity of features. Depending on the category of a user (known, anonymous or frequent visitor to the website), any of the profiles or their combinations is used for making personalized recommendations. A ranking algorithm is also proposed that utilizes the personalized information to order and rank the recommendations. The proposed methodology is evaluated on data collected from a real life car website. Empirical analysis confirms the effectiveness of recommendations made by the proposed approach over other collaborative filtering and content-based recommendation approaches based on two-dimensional data analysis methods.

Time-aware topic recommendation based on micro-blogs

Topic recommendation can help users deal with the information overload issue in micro-blogging communities. This paper proposes to use the implicit information network formed by the multiple relationships among users, topics and micro-blogs, and the temporal information of micro-blogs to find semantically and temporally relevant topics of each topic, and to profile users' time-drifting topic interests. The Content based, Nearest Neighborhood based and Matrix Factorization models are used to make personalized recommendations. The effectiveness of the proposed approaches is demonstrated in the experiments conducted on a real world dataset that collected from Twitter.com.

Integrating collaborative filtering and matching-based search for product recommendation

Currently, recommender systems (RS) have been widely applied in many commercial e-commerce sites to help users deal with the information overload problem. Recommender systems provide personalized recommendations to users and thus help them in making good decisions about which product to buy from the vast number of product choices available to them. Many of the current recommender systems are developed for simple and frequently purchased products like books and videos, by using collaborative-filtering and content-based recommender system approaches. These approaches are not suitable for recommending luxurious and infrequently purchased products as they rely on a large amount of ratings data that is not usually available for such products. This research aims to explore novel approaches for recommending infrequently purchased products by exploiting user generated content such as user reviews and product click streams data. From reviews on products given by the previous users, association rules between product attributes are extracted using an association rule mining technique. Furthermore, from product click streams data, user profiles are generated using the proposed user profiling approach. Two recommendation approaches are proposed based on the knowledge extracted from these resources. The first approach is developed by formulating a new query from the initial query given by the target user, by expanding the query with the suitable association rules. In the second approach, a collaborative-filtering recommender system and search-based approaches are integrated within a hybrid system. In this hybrid system, user profiles are used to find the target user’s neighbour and the subsequent products viewed by them are then used to search for other relevant products. Experiments have been conducted on a real world dataset collected from one of the online car sale companies in Australia to evaluate the effectiveness of the proposed recommendation approaches. The experiment results show that user profiles generated from user click stream data and association rules generated from user reviews can improve recommendation accuracy. In addition, the experiment results also prove that the proposed query expansion and the hybrid collaborative filtering and search-based approaches perform better than the baseline approaches. Integrating the collaborative-filtering and search-based approaches has been challenging as this strategy has not been widely explored so far especially for recommending infrequently purchased products. Therefore, this research will provide a theoretical contribution to the recommender system field as a new technique of combining collaborative-filtering and search-based approaches will be developed. This research also contributes to a development of a new query expansion technique for infrequently purchased products recommendation. This research will also provide a practical contribution to the development of a prototype system for recommending cars.

Australian domestic violence protection order legislation : a comparative quantitative content analysis of victim safety provisions

In Australia, protection orders are a key legal response to domestic violence, and are often viewed as a way of providing for victim safety. For instance, recently the joint Australian and New South Wales Law Reform Commissions recommended that a common core purpose of all state and territory domestic violence legislation should be ‘to ensure or maximise the safety and protection of persons who fear or experience family violence’ (2010:Recommendation 7-4). Drawing and building upon prior research in Australia and the United States (‘US’), this paper uses comparative quantitative content analysis to assess the victim safety focus of domestic violence protection order legislation in each Australian state and territory. The findings of this analysis show that the Northern Territory, South Australia and Victoria ‘stand out’ from the other jurisdictions, having the highest victim safety focus in their legislation. However, there remains sizeable scope for improvement in all Australian jurisdictions, in terms of the victim safety focus of their legislative provisions and the considerations of legislative inconsistency between jurisdictions.

Fuzzy Clustering Based Ad Recommendation for TV Programs

Advertisements(Ads) are the main revenue earner for Television (TV) broadcasters. As TV reaches a large audience, it acts as the best media for advertisements of products and services. With the emergence of digital TV, it is important for the broadcasters to provide an intelligent service according to the various dimensions like program features, ad features, viewers’ interest and sponsors’ preference. We present an automatic ad recommendation algorithm that selects a set of ads by considering these dimensions and semantically match them with programs. Features of the ad video are captured interms of annotations and they are grouped into number of predefined semantic categories by using a categorization technique. Fuzzy categorical data clustering technique is applied on categorized data for selecting better suited ads for a particular program. Since the same ad can be recommended for more than one program depending upon multiple parameters, fuzzy clustering acts as the best suited method for ad recommendation. The relative fuzzy score called “degree of membership” calculated for each ad indicates the membership of a particular ad to different program clusters. Subjective evaluation of the algorithm is done by 10 different people and rated with a high success score.

A hybrid recommendation approach for a tourism system

Many current e-commerce systems provide personalization when their content is shown to users. In this sense, recommender systems make personalized suggestions and provide information of items available in the system. Nowadays, there is a vast amount of methods, including data mining techniques that can be employed for personalization in recommender systems. However, these methods are still quite vulnerable to some limitations and shortcomings related to recommender environment. In order to deal with some of them, in this work we implement a recommendation methodology in a recommender system for tourism, where classification based on association is applied. Classification based on association methods, also named associative classification methods, consist of an alternative data mining technique, which combines concepts from classification and association in order to allow association rules to be employed in a prediction context. The proposed methodology was evaluated in some case studies, where we could verify that it is able to shorten limitations presented in recommender systems and to enhance recommendation quality.

An adaptive information grid architecture for recommendation system

This paper presents an adaptive information grid architecture for recommendation systems, which consists of the features of the recommendation rule and a co-citation algorithm. The algorithm addresses some challenges that are essential for further searching and recommendation algorithms. It does not require users to provide a lot of interactive communication. Furthermore, it supports other queries, such as keyword, URL and document investigations. When the structure is compared to other algorithms, the scalability is noticeably better. The high online performance can be obtained as well as the repository computation, which can achieve a high group-forming accuracy using only a fraction of web pages from a cluster.

Preference networks : probabilistic models for recommendation systems

Recommender systems are important to help users select relevant and personalised information over massive amounts of data available. We propose an unified framework called Preference Network (PN) that jointly models various types of domain knowledge for the task of recommendation. The PN is a probabilistic model that systematically combines both content-based filtering and collaborative filtering into a single conditional Markov random field. Once estimated, it serves as a probabilistic database that supports various useful queries such as rating prediction and top-N recommendation. To handle the challenging problem of learning large networks of users and items, we employ a simple but effective pseudo-likelihood with regularisation. Experiments on the movie rating data demonstrate the merits of the PN.

Preference Networks: probabilistic models for recommendation systems

Abstract
Recommender systems are important to help users select relevant and personalised information over massive amounts of data available. We propose an unified framework called Preference Network (PN) that jointly models various types of domain knowledge for the task of recommendation. The PN is a probabilistic model that systematically combines both content-based filtering and collaborative filtering into a single conditional
Markov random field. Once estimated, it serves as a probabilistic database that supports various useful queries such as rating prediction and top-N recommendation. To handle the challenging problem of learning large networks of users and items, we employ a simple but effective pseudo-likelihood with regularisation. Experiments on the movie rating data demonstrate the merits of the PN.