Lifelogging with SAESNEG: a system for the automated extraction of social network event groups

<p> </p> <p>This thesis presents SAESNEG, a System for the Automated Extraction of Social</p> <p>Network Event Groups; a pipeline for the aggregation of the personal social media</p> <p>footprint, and its partitioning into events, the  event clustering  problem. SAESNEG</p> <p>facilitates a reminiscence-friendly user experience, where the user is able to navigate</p> <p>their social media footprint. A range of socio-technical issues are explored: the</p> <p>challenges to reminiscence, lifelogging, ownership, and digital death.</p> <p>Whilst previous systems have focused on the organisation of a single type of data,</p> <p>such as photos or Tweets respectively; SAESNEG handles a variety of types of social</p> <p>network documents found in a typical footprint (e.g. photos, Tweets, check-ins), with</p> <p>a variety of image, text and other metadata   di erently heterogeneous data; adapted</p> <p>to sparse, private events typical of the personal social media footprint.</p> <p>Phase A extracts information, focusing on natural language processing; new techniques</p> <p>are developed; including a novel distributed approach to handling temporal</p> <p>expressions, and a parser for social events (such as birthdays). Information is also</p> <p>extracted from image and metadata, the resultant annotations feeding the subsequent</p> <p>event clustering. Phase B performs event clustering through the application</p> <p>of a number of pairwise similarity strategies   a mixture of new and existing algorithms.</p> <p>Clustering itself is achieved by combining machine-learning with correlation</p> <p>clustering.</p> <p>The main contributions of this thesis are the identi cation of the technical research</p> <p>task (and the associated social need), the development of novel algorithms and approaches,</p> <p>and the integration of these with existing algorithms to form the pipeline.</p> <p>Results demonstrate SAESNEG's capability to perform event clustering on a differently</p> <p>heterogeneous dataset, enabling users to achieve lifelogging in the context of</p> <p>their existing social media networks.</p>