MIRACLE’s Ad-Hoc and Geographical IR Approaches for CLEF 2006

This paper presents the 2006 Miracle team’s approaches to the Ad-Hoc and Geographical Information Retrieval tasks. A first set of runs was obtained using a set of basic components. Then, by putting together special combinations of these runs, an extended set was obtained. With respect to previous campaigns some improvements have been introduced in our system: an entity recognition prototype is integrated in our tokenization scheme, and the performance of our indexing and retrieval engine has been improved. For GeoCLEF, we tested retrieving using geo-entity and textual references separately, and then combining them with different approaches.

A New Method to Retrieve, Cluster And Annotate Clinical Literature Related To Electronic Health Records

The access to medical literature collections such as PubMed, MedScape or Cochrane has been increased notably in the last years by the web-based tools that provide instant access to the information. However, more sophisticated methodologies are needed to exploit efficiently all that information. The lack of advanced search methods in clinical domain produce that even using well-defined questions for a particular disease, clinicians receive too many results. Since no information analysis is applied afterwards, some relevant results which are not presented in the top of the resultant collection could be ignored by the expert causing an important loose of information. In this work we present a new method to improve scientific article search using patient information for query generation. Using federated search strategy, it is able to simultaneously search in different resources and present a unique relevant literature collection. And applying NLP techniques it presents semantically similar publications together, facilitating the identification of relevant information to clinicians. This method aims to be the foundation of a collaborative environment for sharing clinical knowledge related to patients and scientific publications.

Estudio de las redes sociales para mejorar los servicios de valor añadido en las empresas de telecomunicaciones. Propuesta de un sistema de recuperación de información de las redes sociales.

Las redes sociales en la actualidad son muy relevantes, no solo ocupan mucho tiempo en la vida diaria de las personas si no que también sirve a millones de empresas para publicitarse entre otras cosas. Al fenómeno de las redes sociales se le ha unido la faceta empresarial. La liberación de las APIs de algunas redes sociales ha permitido el desarrollo de aplicaciones de todo tipo y que puedan tener diferentes objetivos como por ejemplo este proyecto. Este proyecto comenzó desde el interés por Ericsson del estudio del API de Google+ y sugerencias para dar valores añadidos a las empresas de telecomunicaciones. También ha complementando la referencia disponible en Ericsson y de los otros dos proyectos de recuperación de información de las redes sociales, añadiendo una serie de opciones para el usuario en la aplicación. Para ello, se ha analizado y realizado un ejemplo, de lo que podemos obtener de las redes sociales, principalmente Twitter y Google+. Lo primero en lo que se ha basado el proyecto ha sido en realizar un estudio teórico sobre el inicio de las redes sociales, el desarrollo y el estado en el que se encuentran, analizando así las principales redes sociales que existen y aportando una visión general sobre todas ellas. También se ha realizado un estado de arte sobre una serie de webs que se dedican al uso de esa información disponible en Internet. Posteriormente, de todas las redes sociales con APIs disponibles se realizó la elección de Google+ porque es una red social nueva aun por explorar y mejorar. Y la elección de Twitter por la serie de opciones y datos que se puede obtener de ella. De ambas se han estudiado sus APIs, para posteriormente con la información obtenida, realizar una aplicación prototipo que recogiera una serie de funciones útiles a partir de los datos de sus redes sociales. Por último se ha realizado una simple interfaz en la cual se puede acceder a los datos de la cuenta como si se estuviera en Twitter o Google+, además con los datos de Twitter se puede realizar una búsqueda avanzada con alertas, un análisis de sentimiento, ver tus mayores retweets de los que te siguen y por último realizar un seguimiento comparando lo que se comenta sobre dos temas determinados. Con este proyecto se ha pretendido proporcionar una idea general de todo lo relacionado con las redes sociales, las aplicaciones disponibles para trabajar con ellas, la información del API de Twitter y Google+ y un concepto de lo que se puede obtener. Today social networks are very relevant, they not only take a long time in daily life of people but also serve millions of businesses to advertise and other things. The phenomenon of social networks has been joined the business side. The release of the APIs of some social networks has allowed the development of applications of all types and different objectives such as this project. This project started from an interest in the study of Ericsson about Google+ API and suggestions to add value to telecommunications companies. This project has complementing the reference available in Ericsson and the other two projects of information retrieval of social networks, adding a number of options for the user in the application. To do this, we have analyzed and made an example of what we can get it from social networks, mainly Twitter and Google+. The first thing that has done in the project was to make a theoretical study on the initiation of social networks, the development and the state in which they are found, and analyze the major social networks that exist. There has also been made a state of art on a number of websites that are dedicated to the use of this information available online. Subsequently, about all the social networks APIs available, Google+ was choice because it is a new social network even to explore and improve. And the choice of Twitter for the number of options and data that can be obtained from it. In both APIs have been studied, and later with the information obtained, make a prototype application to collect a number of useful features from data of social networks. Finally there has been a simple interface, in which you can access the account as if you were on Twitter or Google+. With Twitter data can perform an advanced search with alerts, sentiment analysis, see retweets of who follow you and make comparing between two particular topics. This project is intended to provide an overview of everything related to social networks, applications available to work with them, information about API of Google+ and Twitter, and a concept of what you can get.

Challenges for the Multilingual Web of Data

The Web has witnessed an enormous growth in the amount of semantic information published in recent years. This growth has been stimulated to a large extent by the emergence of Linked Data. Although this brings us a big step closer to the vision of a Semantic Web, it also raises new issues such as the need for dealing with information expressed in different natural languages. Indeed, although the Web of Data can contain any kind of information in any language, it still lacks explicit mechanisms to automatically reconcile such information when it is expressed in different languages. This leads to situations in which data expressed in a certain language is not easily accessible to speakers of other languages. The Web of Data shows the potential for being extended to a truly multilingual web as vocabularies and data can be published in a language-independent fashion, while associated language-dependent (linguistic) information supporting the access across languages can be stored separately. In this sense, the multilingual Web of Data can be realized in our view as a layer of services and resources on top of the existing Linked Data infrastructure adding i) linguistic information for data and vocabularies in different languages, ii) mappings between data with labels in different languages, and iii) services to dynamically access and traverse Linked Data across different languages. In this article we present this vision of a multilingual Web of Data. We discuss challenges that need to be addressed to make this vision come true and discuss the role that techniques such as ontology localization, ontology mapping, and cross-lingual ontology-based information access and presentation will play in achieving this. Further, we propose an initial architecture and describe a roadmap that can provide a basis for the implementation of this vision.

A Balanced Memory-Based Collaborative Filtering Similarity Measure.

Collaborative filtering recommender systems contribute to alleviating the problem of information overload that exists on the Internet as a result of the mass use of Web 2.0 applications. The use of an adequate similarity measure becomes a determining factor in the quality of the prediction and recommendation results of the recommender system, as well as in its performance. In this paper, we present a memory-based collaborative filtering similarity measure that provides extremely high-quality and balanced results; these results are complemented with a low processing time (high performance), similar to the one required to execute traditional similarity metrics. The experiments have been carried out on the MovieLens and Netflix databases, using a representative set of information retrieval quality measures.

Sample Size vs. Bias in Defect Prediction

Most empirical disciplines promote the reuse and sharing of datasets, as it leads to greater possibility of replication. While this is increasingly the case in Empirical Software Engineering, some of the most popular bug-fix datasets are now known to be biased. This raises two significants concerns: first, that sample bias may lead to underperforming prediction models, and second, that the external validity of the studies based on biased datasets may be suspect. This issue has raised considerable consternation in the ESE literature in recent years. However, there is a confounding factor of these datasets that has not been examined carefully: size. Biased datasets are sampling only some of the data that could be sampled, and doing so in a biased fashion; but biased samples could be smaller, or larger. Smaller data sets in general provide less reliable bases for estimating models, and thus could lead to inferior model performance. In this setting, we ask the question, what affects performance more? bias, or size? We conduct a detailed, large-scale meta-analysis, using simulated datasets sampled with bias from a high-quality dataset which is relatively free of bias. Our results suggest that size always matters just as much bias direction, and in fact much more than bias direction when considering information-retrieval measures such as AUC and F-score. This indicates that at least for prediction models, even when dealing with sampling bias, simply finding larger samples can sometimes be sufficient. Our analysis also exposes the complexity of the bias issue, and raises further issues to be explored in the future.

Medida de similitud basada en saliencia

La proliferación en todos los ámbitos de la producción multimedia está dando lugar a la aparición de nuevos paradigmas de recuperación de información visual. Dentro de éstos, uno de los más significativos es el de los sistemas de recuperación de información visual, VIRS (Visual Information Retrieval Systems), en los que una de las tareas más representativas es la ordenación de una población de imágenes según su similitud con un ejemplo dado. En este trabajo se presenta una propuesta original para la evaluación de la similitud entre dos imágenes, basándose en la extensión del concepto de saliencia desde el espacio de imágenes al de características para establecer la relevancia de cada componente de dicho vector. Para ello se introducen metodologías para la cuantificación de la saliencia de valores individuales de características, para la combinación de estas cuantificaciones en procesos de comparación entre dos imágenes, y para, finalmente, establecer la mencionada ponderación de cada característica en atención a esta combinación. Se presentan igualmente los resultados de evaluar esta propuesta en una tarea de recuperación de imágenes por contenido en comparación con los obtenidos con la distancia euclídea. Esta comparación se realiza mediante la evaluación de ambos resultados por voluntarios.

Modelado y Comparación de Colecciones en la Web Semántica

En esta tesis se estudia la representación, modelado y comparación de colecciones mediante el uso de ontologías en el ámbito de la Web Semántica. Las colecciones, entendidas como agrupaciones de objetos o elementos con entidad propia, son construcciones que aparecen frecuentemente en prácticamente todos los dominios del mundo real, y por tanto, es imprescindible disponer de conceptualizaciones de estas estructuras abstractas y de representaciones de estas conceptualizaciones en los sistemas informáticos, que definan adecuadamente su semántica. Mientras que en muchos ámbitos de la Informática y la Inteligencia Artificial, como por ejemplo la programación, las bases de datos o la recuperación de información, las colecciones han sido ampliamente estudiadas y se han desarrollado representaciones que responden a multitud de conceptualizaciones, en el ámbito de la Web Semántica, sin embargo, su estudio ha sido bastante limitado. De hecho hasta la fecha existen pocas propuestas de representación de colecciones mediante ontologías, y las que hay sólo cubren algunos tipos de colecciones y presentan importantes limitaciones. Esto impide la representación adecuada de colecciones y dificulta otras tareas comunes como la comparación de colecciones, algo crítico en operaciones habituales como las búsquedas semánticas o el enlazado de datos en la Web Semántica. Para solventar este problema esta tesis hace una propuesta de modelización de colecciones basada en una nueva clasificación de colecciones de acuerdo a sus características estructurales (homogeneidad, unicidad, orden y cardinalidad). Esta clasificación permite definir una taxonomía con hasta 16 tipos de colecciones distintas. Entre otras ventajas, esta nueva clasificación permite aprovechar la semántica de las propiedades estructurales de cada tipo de colección para realizar comparaciones utilizando las funciones de similitud y disimilitud más apropiadas. De este modo, la tesis desarrolla además un nuevo catálogo de funciones de similitud para las distintas colecciones, donde se han recogido las funciones de (di)similitud más conocidas y también algunas nuevas. Esta propuesta se ha implementado mediante dos ontologías paralelas, la ontología E-Collections, que representa los distintos tipos de colecciones de la taxonomía y su axiomática, y la ontología SIMEON (Similarity Measures Ontology) que representa los tipos de funciones de (di)similitud para cada tipo de colección. Gracias a estas ontologías, para comparar dos colecciones, una vez representadas como instancias de la clase más apropiada de la ontología E-Collections, automáticamente se sabe qué funciones de (di)similitud de la ontología SIMEON pueden utilizarse para su comparación. Abstract This thesis studies the representation, modeling and comparison of collections in the Semantic Web using ontologies. Collections, understood as groups of objects or elements with their own identities, are constructions that appear frequently in almost all areas of the real world. Therefore, it is essential to have conceptualizations of these abstract structures and representations of these conceptualizations in computer systems, that define their semantic properly. While in many areas of Computer Science and Artificial Intelligence, such as Programming, Databases or Information Retrieval, the collections have been extensively studied and there are representations that match many conceptualizations, in the field Semantic Web, however, their study has been quite limited. In fact, there are few representations of collections using ontologies so far, and they only cover some types of collections and have important limitations. This hinders a proper representation of collections and other common tasks like comparing collections, something critical in usual operations such as semantic search or linking data on the Semantic Web. To solve this problem this thesis makes a proposal for modelling collections based on a new classification of collections according to their structural characteristics (homogeneity, uniqueness, order and cardinality). This classification allows to define a taxonomy with up to 16 different types of collections. Among other advantages, this new classification can leverage the semantics of the structural properties of each type of collection to make comparisons using the most appropriate (dis)similarity functions. Thus, the thesis also develops a new catalog of similarity functions for the different types of collections. This catalog contains the most common (dis)similarity functions as well as new ones. This proposal is implemented through two parallel ontologies, the E-Collections ontology that represents the different types of collections in the taxonomy and their axiomatic, and the SIMEON ontology (Similarity Measures Ontology) that represents the types of (dis)similarity functions for each type of collection. Thanks to these ontologies, to compare two collections, once represented as instances of the appropriate class of E-Collections ontology, we can know automatically which (dis)similarity functions of the SIMEON ontology are suitable for the comparison. Finally, the feasibility and usefulness of this modeling and comparison of collections proposal is proved in the field of oenology, applying both E-Collections and SIMEON ontologies to the representation and comparison of wines with the E-Baco ontology.

Improving Speech Interaction in Vehicles Using Context-Aware Information through A SCXML Framework

Speech Technologies can provide important benefits for the development of more usable and safe in-vehicle human-machine interactive systems (HMIs). However mainly due robustness issues, the use of spoken interaction can entail important distractions to the driver. In this challenging scenario, while speech technologies are evolving, further research is necessary to explore how they can be complemented with both other modalities (multimodality) and information from the increasing number of available sensors (context-awareness). The perceived quality of speech technologies can significantly be increased by implementing such policies, which simply try to make the best use of all the available resources; and the in vehicle scenario is an excellent test-bed for this kind of initiatives. In this contribution we propose an event-based HMI design framework which combines context modelling and multimodal interaction using a W3C XML language known as SCXML. SCXML provides a general process control mechanism that is being considered by W3C to improve both voice interaction (VoiceXML) and multimodal interaction (MMI). In our approach we try to anticipate and extend these initiatives presenting a flexible SCXML-based approach for the design of a wide range of multimodal context-aware HMI in-vehicle interfaces. The proposed framework for HMI design and specification has been implemented in an automotive OSGi service platform, and it is being used and tested in the Spanish research project MARTA for the development of several in-vehicle interactive applications.

Evaluation of a User-Adapted Spoken Language Dialogue System: Measuring the Relevance of the Contextual Information Sources

We present an evaluation of a spoken language dialogue system with a module for the management of userrelated information, stored as user preferences and privileges. The flexibility of our dialogue management approach, based on Bayesian Networks (BN), together with a contextual information module, which performs different strategies for handling such information, allows us to include user information as a new level into the Context Manager hierarchy. We propose a set of objective and subjective metrics to measure the relevance of the different contextual information sources. The analysis of our evaluation scenarios shows that the relevance of the short-term information (i.e. the system status) remains pretty stable throughout the dialogue, whereas the dialogue history and the user profile (i.e. the middle-term and the long-term information, respectively) play a complementary role, evolving their usefulness as the dialogue evolves.

Mutual Information and Perplexity based Clustering of Dialogue Information for Dynamic Adaptation of Language Models

We present two approaches to cluster dialogue-based information obtained by the speech understanding module and the dialogue manager of a spoken dialogue system. The purpose is to estimate a language model related to each cluster, and use them to dynamically modify the model of the speech recognizer at each dialogue turn. In the first approach we build the cluster tree using local decisions based on a Maximum Normalized Mutual Information criterion. In the second one we take global decisions, based on the optimization of the global perplexity of the combination of the cluster-related LMs. Our experiments show a relative reduction of the word error rate of 15.17%, which helps to improve the performance of the understanding and the dialogue manager modules.

Enabling query technologies for the semantic sensor web

Sensor networks are increasingly being deployed in the environment for many different purposes. The observations that they produce are made available with heterogeneous schemas, vocabularies and data formats, making it difficult to share and reuse this data, for other purposes than those for which they were originally set up. The authors propose an ontology-based approach for providing data access and query capabilities to streaming data sources, allowing users to express their needs at a conceptual level, independent of implementation and language-specific details. In this article, the authors describe the theoretical foundations and technologies that enable exposing semantically enriched sensor metadata, and querying sensor observations through SPARQL extensions, using query rewriting and data translation techniques according to mapping languages, and managing both pull and push delivery modes.

Mutual Information and Perplexity based Clustering of Dialogue Information for Dynamic Adaptation of Language Models

We present two approaches to cluster dialogue-based information obtained by the speech understanding module and the dialogue manager of a spoken dialogue system. The purpose is to estimate a language model related to each cluster, and use them to dynamically modify the model of the speech recognizer at each dialogue turn. In the first approach we build the cluster tree using local decisions based on a Maximum Normalized Mutual Information criterion. In the second one we take global decisions, based on the optimization of the global perplexity of the combination of the cluster-related LMs. Our experiments show a relative reduction of the word error rate of 15.17%, which helps to improve the performance of the understanding and the dialogue manager modules.

On the dynamic adaptation of language models based on dialogue information

We present an approach to adapt dynamically the language models (LMs) used by a speech recognizer that is part of a spoken dialogue system. We have developed a grammar generation strategy that automatically adapts the LMs using the semantic information that the user provides (represented as dialogue concepts), together with the information regarding the intentions of the speaker (inferred by the dialogue manager, and represented as dialogue goals). We carry out the adaptation as a linear interpolation between a background LM, and one or more of the LMs associated to the dialogue elements (concepts or goals) addressed by the user. The interpolation weights between those models are automatically estimated on each dialogue turn, using measures such as the posterior probabilities of concepts and goals, estimated as part of the inference procedure to determine the actions to be carried out. We propose two approaches to handle the LMs related to concepts and goals. Whereas in the first one we estimate a LM for each one of them, in the second one we apply several clustering strategies to group together those elements that share some common properties, and estimate a LM for each cluster. Our evaluation shows how the system can estimate a dynamic model adapted to each dialogue turn, which helps to improve the performance of the speech recognition (up to a 14.82% of relative improvement), which leads to an improvement in both the language understanding and the dialogue management tasks.