Exploring the Suitability of Geospatial Visualization and Temporal Filtering Techniques for Lifelog Data
Summary
Lifelogging is the collection of data through sensors carried by a user which record the user's life experiences. Examples of sensors that are used to collect lifelog data are (wearable) cameras, GPS, and sensors collecting biometric data. This activity results in lifelogs, that is, multimodal databases containing large amounts of data that is challenging to navigate. Related research mostly focuses on lifelog retrieval; building a system which can retrieve specific data of the lifelog. These systems are made to enable the user to retrieve certain memories. Yet, they lack in supporting exploration of the data for leisure browsing or situations with a less clear information need, for example, when people only vaguely remember a certain situation. A good visualization of a lifelog's content and the ability to filter it as needed could encourage and support such exploration, but related studies are sparse. Therefore, in this thesis we aim to determine if geospatial visualization and temporal filtering techniques can be used to make lifelogs accessible and easy to search and explore. To achieve this, we implemented and evaluated a system, the Lifelog Browser, which visualizes the LSC'20 test collection, a benchmark lifelog dataset, and makes it accessible. The system features three different geospatial visualizations (Dot map, Cluster map and Heat map) and three different temporal filtering techniques (color-coded Week matrix, color-coded Month matrix and Custom drop-down), which were evaluated for suitability for different search goals in the context of lifelog data. This was done through a user study which included both a lifelog exploration and retrieval challenge, inspired by the Lifelog Search Challenge. Our results show that this type of system is suitable for lifelog access, search and exploration and that out of the chosen approaches, the Cluster geospatial visualization and Custom drop-down temporal filtering technique are considered most useful for lifelog exploration and search.