As of very recently, we have observed a convergence of technologies that have led to the emergence of lifelogging as a technology for personal data application. Lifelogging will become ubiquitous in the near future, not just for memory enhancement and health management, but also in various other domains. While there are many devices available for gathering massive lifelogging data, there are still challenges to modelling large volume of multi-modal lifelog data. In the thesis, we explore and address the problem of how to model lifelog in order to make personal lifelogs more accessible to users from the perspective of collection, organization and visualization. In order to subdivide our research targets, we designed and followed the following steps to solve the problem: 1. Lifelog activity recognition. We use multiple sensor data to analyse various daily life activities. Data ranges from accelerometer data collected by mobile phones to images captured by wearable cameras. We propose a semantic, density-based algorithm to cope with concept selection issues for lifelogging sensory data. 2. Visual discovery of lifelog images. Most of the lifelog information we takeeveryday is in a form of images, so images contain significant information about our lives. Here we conduct some experiments on visual content analysis of lifelog images, which includes both image contents and image meta data. 3. Linkage analysis of lifelogs. By exploring linkage analysis of lifelog data, we can connect all lifelog images using linkage models into a concept called the MemoryMesh. The thesis includes experimental evaluations using real-life data collected from multiple users and shows the performance of our algorithms in detecting semantics of daily-life concepts and their effectiveness in activity recognition and lifelog retrieval.