Search

RESUMEN: Este artículo muestra resultados de una investigación interdisciplinar aplicada a la capacidad predictiva de las redes socia les, específicamente Twitter, en las pri marias legales en Chile de 2017. Mediante la incorporación de inteligencia compu tacional, se monitoreó la interacción de todos los usuarios chilenos que men cionaron al menos una vez a algunos de los cinco candidatos en competencia para diseñar un modelo de pronóstico que consideró el contexto propio de la comunicación política. Como resultado, se lograron modelos cuyos pronósticos estuvieron bajo 2% en el error absoluto medio (MAE), es decir, con mayor preci sión que las encuestas electorales. ABSTRACT: This article shows the results of an interdisciplinary research applied to the predictive capacity of social networks, specifically Twitter, in the legal primaries in Chile in 2017. Through the incorporation of computational intelligence, we monitored the interaction of all Chilean users who mentioned at least once some of the five candidates competing to design a forecast model that considered the context of political communication, which delivered results under 2% in the mean absolute error (MAE), with more precision than the electoral polls. RESUMO: O seguinte artigo fornece os resultados de uma pesquisa interdisciplinar apli cada à capacidade preditiva das redes sociais, especificamente o Twitter, nas primárias legais constituídas no Chile durante o ano de 2017. Por meio da incor poração de inteligência computacional, monitorou-se a interação dos internautas chilenos que mencionaram pelo menos uma vez algum dos cinco candidatos na competição eleitoral para desenhar um modelo de previsão de resultados que considerasse contexto da comuni cação política. Esse modelo apresentou resultados de menos de 2% na mediada do erro absoluto (MAE), gerando maior precisão do que as pesquisas eleitorais.

National audience; Recent years have witnessed the growth of a new class of data-intensive applications that do not fit the DBMS data model and querying paradigm. Instead, the data arrive at high speeds taking the form of an unbounded sequence of values (data streams) and queries run continuously returning new results as new data arrive. In these applications, data streams from external sources flow into a Data Stream Management System (DSMS) where they are processed by different operators. Many applications share the same need for processing data streams in a continuous fashion. For most distributed streaming applications, the centralized processing of continuous queries over distributed data is simply not viable. This paper addresses the problem of computing continuous join queries over distributed data streams. We present a new method, called DHTJoin that exploits the power of a Distributed Hash Table (DHT) combining hash-based placement of tuples and dissemination of queries by exploiting the embedded trees in the underlying DHT, thereby incuring little overhead. Unlike state of the art solutions that index all data, DHTJoin identifies, using query predicates, a subset of tuples in order to index the data required by the user's queries, thus reducing network traffic. DHTJoin tackles the dynamic behavior of DHT networks during query execution and dissemination of queries. We provide a performance evaluation of DHTJoin which shows that it can achieve significant performance gains in terms of network traffic.

Heurističke metode nizanja vrijednosti i varijabli su ključni element u ograničenom programiranju. Poznate su kao strategija nabrajanja i mogu značajno utjecati na postupak rješavanja problema. Međutim, prilično je teško izabrati odgovarajući heuristički postupak jer je komplicirano predvidjeti njihovo ponašanje. U zadnje je vrijeme za tu svrhu predloženo samostalno (autonomno) pretraživanje. Ideja je da se strategije koje su se pokazale lošima tijekom postupka rješavanja dinamički zamijene onima koje više obećavaju. Ta se zamjena izvodi korištenjem funkcije izbora, koja u zadanom vremenu procijenjuje ponuđenu strategiju preko indikatora kvalitete. Važnu ulogu u tom procesu ima optimizator kojemu je cilj fino podešavanje funkcije izbora kako bi se garantirala precizna procjena strategija. U ovom radu evaluiramo karakteristike dviju jakih funkcija izbora: prvu podržava genetski algoritam, a drugu optimizator roja čestica. Dajemo interesantne rezultate i demonstriramo mogućnost korištenja tih metoda optimiziranja za samostalno pretraživanje u kontekstu ograničenog programiranja., The variable and value ordering heuristics are a key element in Constraint Programming. Known together as the enumeration strategy they may have important consequences on the solving process. However, a suitable selection of heuristics is quite hard as their behaviour is complicated to predict. Autonomous search has been recently proposed to handle this concern. The idea is to dynamically replace strategies that exhibit poor performances by more promising ones during the solving process. This replacement is carried out by a choice function, which evaluates a given strategy in a given amount of time via quality indicators. An important phase of this process is performed by an optimizer, which aims at finely tuning the choice function in order to guarantee a precise evaluation of strategies. In this paper we evaluate the performance of two powerful choice functions: the first one supported by a genetic algorithm and the second one by a particle swarm optimizer. We present interesting results and we demonstrate the feasibility of using those optimization techniques for Autonomous Search in a Constraint Programming context.

International audience

Recent years have witnessed the growth of a new class of data-intensive applications that do not fit the DBMS data model and querying paradigm. Instead, the data arrive at High speeds taking the form of an unbounded sequence of values (data streams) and queries run continuously returning new results as new data arrive. In these applications, data streams from external sources flow into a Data Stream Management System (DSMS) where they are processed by different operators. Many applications share the same need for processing data streams in a continuous fashion. For most distributed streaming applications, the centralized processing of continuous queries over distributed data is simply not viable. This research addresses the problem of computing continuous join queries over distributed data streams.We present a new method, called DHTJoin that exploits the power of a Distributed Hash Table (DHT) combining hash-based placement of tuples and dissemination of queries by exploiting the embedded trees in the underlying DHT, thereby incuring little overhead. Unlike state of the art solutions that index all data, DHTJoin identifies, using query predicates, a subset of tuples in order to index the data required by the user's queries, thus reducing network traffic. DHTJoin tackles the dynamic behavior of DHT networks during query execution and dissemination of queries. We provide a performance evaluation of DHTJoin which shows that it can achieve significant performance gains in terms of network traffic.; De nombreuses applications distribuées partagent la même nécessité de traiter des flux de données de façon continue, par ex. la surveillance de réseau ou la gestion de réseaux de capteurs. Dans ce contexte, un problème important et difficile concerne le traitement de requêtes continues de jointure qui nécessite de maintenir une fenêtre glissante sur les données la plus grande possible, afin de produire le plus possible de résultats probants. Dans cette thèse, nous proposons une nouvelle méthode pair-à-pair, DHTJoin, qui tire parti d'une Table de Hachage Distribuée (DHT) pour augmenter la taille de la fenêtre glissante en partitionnant les flux sur un grand nombre de nœuds. Contrairement aux solutions concurrentes qui indexent tout les tuples des flux, DHTJoin n'indexe que les tuples requis pour les requêtes et exploite, de façon complémentaire, la dissémination de requêtes. DHTJoin traite aussi le problème de la dynamicité des nœuds, qui peuvent quitter le système ou tomber en panne pendant l'exécution. Notre évaluation de performances montre que DHTJoin apporte une réduction importante du trafic réseau, par rapport aux méthodes concurrentes.

Recent years have witnessed the growth of a new class of data-intensive applications that do not fit the DBMS data model and querying paradigm. Instead, the data arrive at High speeds taking the form of an unbounded sequence of values (data streams) and queries run continuously returning new results as new data arrive. In these applications, data streams from external sources flow into a Data Stream Management System (DSMS) where they are processed by different operators. Many applications share the same need for processing data streams in a continuous fashion. For most distributed streaming applications, the centralized processing of continuous queries over distributed data is simply not viable. This research addresses the problem of computing continuous join queries over distributed data streams.We present a new method, called DHTJoin that exploits the power of a Distributed Hash Table (DHT) combining hash-based placement of tuples and dissemination of queries by exploiting the embedded trees in the underlying DHT, thereby incuring little overhead. Unlike state of the art solutions that index all data, DHTJoin identifies, using query predicates, a subset of tuples in order to index the data required by the user's queries, thus reducing network traffic. DHTJoin tackles the dynamic behavior of DHT networks during query execution and dissemination of queries. We provide a performance evaluation of DHTJoin which shows that it can achieve significant performance gains in terms of network traffic.; De nombreuses applications distribuées partagent la même nécessité de traiter des flux de données de façon continue, par ex. la surveillance de réseau ou la gestion de réseaux de capteurs. Dans ce contexte, un problème important et difficile concerne le traitement de requêtes continues de jointure qui nécessite de maintenir une fenêtre glissante sur les données la plus grande possible, afin de produire le plus possible de résultats probants. Dans cette thèse, nous proposons une nouvelle méthode pair-à-pair, DHTJoin, qui tire parti d'une Table de Hachage Distribuée (DHT) pour augmenter la taille de la fenêtre glissante en partitionnant les flux sur un grand nombre de nœuds. Contrairement aux solutions concurrentes qui indexent tout les tuples des flux, DHTJoin n'indexe que les tuples requis pour les requêtes et exploite, de façon complémentaire, la dissémination de requêtes. DHTJoin traite aussi le problème de la dynamicité des nœuds, qui peuvent quitter le système ou tomber en panne pendant l'exécution. Notre évaluation de performances montre que DHTJoin apporte une réduction importante du trafic réseau, par rapport aux méthodes concurrentes.

International audience; This paper addresses the problem of computing approximate answers to continuous join queries. We present a new method, called DHTJoin, which combines hash-based placement of tuples in a Distributed Hash Table (DHT) and dissemination of queries exploiting the trees formed by the underlying DHT links. DHTJoin distributes the query workload across multiple DHT nodes and provides a mechanism that avoids indexing tuples that cannot contribute to join results. We provide a performance evaluation which shows that DHTJoin can achieve significant performance gains in terms of network traffic.

National audience