Anja Bachmann

How to use smartphones for less obtrusive ambulatory mood assessment and mood recognition

We present MoA2, a context-aware smartphone app for the ambulatory assessment of mood, tiredness and stress level. In principle, it has two features: (1) mood assessment and (2) mood recognition. The mood assessment system combines benefits of state of the art approaches. The mood recognition is concluded by smartphone-based wearable sensing. In a formative study, we evaluated the usability and unobtrusiveness of our mood assessment. A median SUS score of 90 shows a high usability. Subjects reported an easy, fast and intuitive use. The mood recognition was evaluated in terms of classification accuracy. First, we analyzed which features are best for the recognition. Spatio-temporal attributes, i.e. daytime, day of week and location, correlate most with the monitored mood. Based on the identified attributes, we trained personalized classifiers using Naïve Bayes and applied ten-fold-cross validation. The average recognition accuracy was 0.76 which is comparable to related work.

Incremental Frequent Route Based Trajectory Prediction

Recent technological trends enable modern traffic prediction and management systems in which the analysis and prediction of movements of objects is essential. To this extent the present paper proposes IncCCFR---a novel, incremental approach for managing, mining, and predicting the incrementally evolving trajectories of moving objects. In addition to reduced mining and storage costs, a key advantage of the incremental approach is its ability to combine multiple temporally relevant mining results from the past to capture temporal and periodic regularities in movement. The approach and its variants are empirically evaluated on a large real-world data set of moving object trajectories, originating from a fleet of taxis, illustrating that detailed closed frequent routes can be efficiently discovered and used for prediction.

Leveraging smartwatches for unobtrusive mobile ambulatory mood assessment

In ambulatory assessment, subjects are monitored in everyday life. Though, it is difficult to unobtrusively assess information -- e.g. about their context and affective state -- which results in an increased burden for the subjects. This burden is caused by a complex self-report that they need to provide or by additional wearables that need to be carried. Newest technology can solve this issue by assessing a variety of information automatically. We propose to use smartwatches in combination with smartphones to assess physiological and smartphone data from which the affective state of a user can be inferred. We present the principle idea of our app and how we intend on evaluating it. A review of state of the art approaches and available Android Wear smartwatches in terms of sensors is given. We present a number of smartphone sensors and a selection of smartwatches whose combination should be evaluated regarding usefulness for mood assessment and recognition.

Identification of relevant sensor sources for context-aware ESM apps in ambulatory assessment

The experience sampling method (ESM) is applied in ambulatory assessment to prompt subject self-reporting. Existing mobile apps provide time-triggered prompts but lack event-triggers. Hence, the sampling might not occur in moments that are of interest for a psychologist. To identify relevant sensor sources and contexts we conducted an online survey with ambulatory assessment experts. Most relevant for these experts are time, date and user activity, followed by location, notifications and accelerometer. A feasibility test proved that all relevant sources are accessible on Android phones. We also assessed the desired granularity of the data gathered from each sensor source. Our results are a first step towards an ESM platform to create context-aware Android apps for ambulatory assessment.

jActivity: supporting mobile web developers with HTML5/JavaScript based human activity recognition

Human Activity Recognition (HAR) using accelerometers has been studied intensively in the past decade. Recent HTML5 methods allow sampling a mobile phones sensors from within web pages. Our objective is to leverage this for the creation of individual activity recognition modules that can be included into web applications to allow them to gain context-awareness. In this work, jActivity, a first prototype of such a platform-independent HTML5/JavaScript framework is presented, along with experiments to determine the general feasibility and challenges for HAR in web applications. Our results indicate that the realization looks promising, albeit so far limited to certain devices/user agents.

Extracting cross references from life science databases for search result ranking

Scholars in life sciences have to process huge amounts of data in a disciplined and efficient way. These data are spread among thousands of databases which overlap in content but differ substantially with respect to interface, formats and data structure. Search engines have the potential of assisting in data retrieval from these structured sources but fall short of providing a relevance ranking of the results that reflects the needs of life science scholars. One such need is to acquire insights to cross-references among entities in the databases, whereby search hits with many cross-references are expected to be more informative than those with few cross-references. In this work, we investigate to what extend this expectation holds. We propose BioXREF, a method that extracts cross-references from multiple life science databases by combining targeted crawling, pointer chasing, sampling and information extraction. We study the retrieval quality of our method and the relationship between manually crafted relevance ranking and relevance ranking based on cross-references, and report on first, promising results.

Towards smartphone-based sensing of social interaction for ambulatory assessment

In ambulatory assessment, especially when handling subjects with personality disorders, it is important to monitor the subjects social interactions. Smartphones are already applied to assess information via self-reports. Related work also uses them to log context information or prompt event-specific self-reports. We see a high potential for them to be used for monitoring social interactions in in-field studies as they are constant companions in real life and a platform for virtual interactions. Our system will apply pattern recognition and machine learning algorithms to physical sensor measurements such as microphone and radio signals, but also to virtual sensor information such as call and message history and activity of messaging apps. We will evaluate to which extent and how well our system can automatically and unobtrusively find indicators for social interactions and, based on them, identify anomalies in the subjects behavior.