Processing Structured Data Streams

Abstract

A large amount of data is daily generated from different sources such as social networks, recommendation systems or geolocation systems. Moreover, this information tends to grow exponentially every year. Companies have discovered that the processing of these data may be important in order to obtain useful conclusions that serve for decision-making or the detection and resolution of problems in a more efficient way, for instance, through the study of trends, habits or customs of the population. The information provided by these sources typically consists of a non-structured and continuous data flow, where the relations among data elements conform graph structures. Inevitably, the processing performance of this information progressively decreases as the size of the data increases. For this reason, non-structured information is usually handled taking into account only the most recent data and discarding the rest, since they are considered not relevant when drawing conclusions. However, this approach is not enough in the case of sources that provide graph-structured data, since it is necessary to consider spatial features as well as temporal features. These spatial features refer to the relationships among the data elements. For example, some cases where it is important to consider spatial aspects are marketing techniques, which require information on the location of users and their possible needs, or the detection of diseases, that use data about genetic relationships among subjects or the geographic scope. It is worth highlighting three main contributions from this dissertation. First, we provide a comparative study of seven of the most common processing platforms to work with huge graphs and the languages that are used to query them. This study measures the performance of the queries in terms of execution time, and the syntax complexity of the languages according to three parameters: number of characters, number of operators and number of internal variables.