Representation and exploitation of event sequences

Abstract

The Ten Commandments, the thirty best smartphones in the market and the five most wanted people by the FBI. Our life is ruled by sequences: thought sequences, number sequences, event sequences. . . a history book is nothing more than a compilation of events and our favorite film is just a sequence of scenes. All of them have something in common, it is possible to acquire relevant information from them. Frequently, by accumulating some data from the elements of each sequence we may access hidden information (e.g. the passengers transported by a bus on a journey is the sum of the passengers who got on in the sequence of stops made); other times, reordering the elements by any of their characteristics facilitates the access to the elements of interest (e.g. the publication of books in 2019 can be ordered chronologically, by author, by literary genre or even by a combination of characteristics); but it will always be sought to store them in the smallest space possible. Thus, this thesis proposes technological solutions for the storage and subsequent processing of events, focusing specifically on three fundamental aspects that can be found in any application that needs to manage them: compressed and dynamic storage, aggregation or accumulation of elements of the sequence and element sequence reordering by their different characteristics or dimensions. The first contribution of this work is a compact structure for the dynamic compression of event sequences. This structure allows any sequence to be compressed in a single pass, that is, it is capable of compressing in real time as elements arrive. This contribution is a milestone in the world of compression since, to date, this is the first proposal for a variable-to-variable dynamic compressor for general purpose. Regarding aggregation, a data warehouse-like proposal is presented capable of storing information on any characteristic of the events in a sequence in an aggregated, compact and accessible way. Following the philosophy of current data warehouses, we avoid repeating cumulative operations and speed up aggregate queries by preprocessing the information and keeping it in this separate structure. Finally, this thesis addresses the problem of indexing event sequences considering their different characteristics and possible reorderings. A new approach for simultaneously keeping the elements of a sequence ordered by different characteristics is presented through compact structures. Thus, it is possible to consult the information and perform operations on the elements of the sequence using any possible rearrangement in a simple and efficient way.