Bob Price

Activity-based serendipitous recommendations with the Magitti mobile leisure guide

This paper presents a context-aware mobile recommender system, codenamed Magitti. Magitti is unique in that it infers user activity from context and patterns of user behavior and, without its user having to issue a query, automatically generates recommendations for content matching. Extensive field studies of leisure time practices in an urban setting (Tokyo) motivated the idea, shaped the details of its design and provided data describing typical behavior patterns. The paper describes the fieldwork, user interface, system components and functionality, and an evaluation of the Magitti prototype.

Proactive Insider Threat Detection through Graph Learning and Psychological Context

The annual incidence of insider attacks continues to grow, and there are indications this trend will continue. While there are a number of existing tools that can accurately identify known attacks, these are reactive (as opposed to proactive) in their enforcement, and may be eluded by previously unseen, adversarial behaviors. This paper proposes an approach that combines Structural Anomaly Detection (SA) from social and information networks and Psychological Profiling (PP) of individuals. SA uses technologies including graph analysis, dynamic tracking, and machine learning to detect structural anomalies in large-scale information network data, while PP constructs dynamic psychological profiles from behavioral patterns. Threats are finally identified through a fusion and ranking of outcomes from SA and PP. The proposed approach is illustrated by applying it to a large data set from a massively multi-player online game, World of War craft (WoW). The data set contains behavior traces from over 350,000 characters observed over a period of 6 months. SA is used to predict if and when characters quit their guild (a player association with similarities to a club or workgroup in non-gaming contexts), possibly causing damage to these social groups. PP serves to estimate the five-factor personality model for all characters. Both threads show good results on the gaming data set and thus validate the proposed approach.

Collaborative Filtering Is Not Enough? Experiments with a Mixed-Model Recommender for Leisure Activities

Collaborative filtering (CF) is at the heart of most successful recommender systems nowadays. While this technique often provides useful recommendations, conventional systems also ignore data that could potentially be used to refine and adjust recommendations based on a users context and preferences. The problem is particularly acute with mobile systems where information delivery often needs to be contextualized. Past research has also shown that combining CF with other techniques often improves the quality of recommendations. In this paper, we present results from an experiment assessing user satisfaction with recommendations for leisure activities that are obtained from different combinations of these techniques. We show that the most effective mix is highly dependent on a users familiarity with a geographical area and discuss the implications of our findings for future research.

Enhancing Mobile Recommender Systems with Activity Inference

Todays mobile leisure guide systems give their users unprecedented help in finding places of interest. However, the process still requires significant user interaction, for example to specify preferences and navigate lists. While interaction is effective for obtaining desired results, learning the interaction pattern can be an obstacle for new users, and performing it can slow down experienced users. This paper describes how to infer a users high-level activity automatically to improve recommendations. Activity is determined by interpreting a combination of current sensor data, models generated from historical sensor data, and priors from a large time-use study. We present an initial user study that shows an increase in prediction accuracy from 62% to over 77%, and discuss the challenges of integrating activity representations into a user model.

Scalable architecture for context-aware activity-detecting mobile recommendation systems

One of the main challenges in building multi-user mobile information systems for real-world deployment lies in the development of scalable systems. Recent work on scaling infrastructure for conventional web services using distributed approaches can be applied to the mobile space, but limitations inherent to mobile devices (computational power, battery life) and their communication infrastructure (availability and quality of network connectivity) challenge system designers to carefully design and optimize their software architectures. Additionally, notions of mobility and position in space, unique to mobile systems, provide interesting directions for the segmentation and scalability of mobile information systems. In this paper we describe the implementation of a mobile recommender system for leisure activities, codenamed Magitti, which was built for commercial deployment under stringent scalability requirements. We present concrete solutions addressing these scalability challenges, with the goal of informing the design of future mobile multi-user systems.

A unified information criterion for evaluating probe and test selection

Diagnostic tasks often need to make the decision of what measurement to make or what action to take in order to resolve ambiguities in diagnosis. Intuitively one would like to seek the most ldquoinformativerdquo choice. In the paper, we formalize this intuition and propose an information criterion for evaluating and comparing measurement/action choices based on their information contribution. The criterion is mutual information, an information-theoretic concept measuring statistical dependence. The information criterion gives a precise quantitative metric to differentiate the quality of measurement/action choices. We use a few concrete example in two separate paradigms, probe selection in circuit diagnosis and test generation in production plants, to illustrate the mutual information criterion. Despite the apparent differences of the two paradigms, the information criterion works coherently. We demonstrate how different probing actions or test plans vary in their information values.

Activity duration analysis for context-aware services using foursquare check-ins

Location-based Social Networks (LBSN) such as Foursquare are becoming an increasingly popular social media, where users share their location and activities with other users mainly using smartphones and Internet of Things. Data logged by LBSNs, such as Foursquare user check-in events, can be used to derive user models and improve the context-awareness and efficiency of various applications like recommender systems. In particular, activity duration is one important aspect of user behavior that we can derive from LBSNs to improve the timing of recommendation sent to users. These durations are not otherwise directly available from LBSNs, partial GPS-enabled tracking or explicit recording due to various practical constraints. From a Four-square dataset with about 3.7 million users and 300 million check-ins, we observed patterns which inspired us to design methods to determine the duration of user activities. We discuss preliminary results and outline plans for thorough evaluations and future research.

Deriving and improving CMA-ES with information geometric trust regions

CMA-ES is one of the most popular stochastic search algorithms. It performs favourably in many tasks without the need of extensive parameter tuning. The algorithm has many beneficial properties, including automatic step-size adaptation, efficient covariance updates that incorporates the current samples as well as the evolution path and its invariance properties. Its update rules are composed of well established heuristics where the theoretical foundations of some of these rules are also well understood. In this paper we will fully derive all CMA-ES update rules within the framework of expectation-maximisation-based stochastic search algorithms using information-geometric trust regions. We show that the use of the trust region results in similar updates to CMA-ES for the mean and the covariance matrix while it allows for the derivation of an improved update rule for the step-size. Our new algorithm, Trust-Region Co-variance Matrix Adaptation Evolution Strategy (TR-CMA-ES) is fully derived from first order optimization principles and performs favourably in compare to standard CMA-ES algorithm.

Pervasive Diagnosis

In model-based production, a planner uses a system description to create plans that achieve production goals. The same description can be used by model-based diagnosis to infer the condition of components from sensor data. When production is realized by a sequence of plans, prior work has demonstrated that diagnosis can be used to adapt the plans to compensate for component degradation. However, the sources of diagnostic information are severely limited. Diagnosis must either make inferences from observations during production over which it has no control (passive diagnosis), or production must be halted to introduce diagnostic-specific plans (explicit diagnosis). We observe that the declarative nature of the model-based approach allows the planner to achieve production goals in multiple ways. This flexibility is exploited by a novel paradigm, i.e., pervasive (active) diagnosis, which constructs informative production plans that simultaneously achieve production goals while uncovering additional diagnostic information about the condition of components. We present an efficient heuristic search for these informative production plans and show through experiments on a model of an industrial digital printing press that the theoretical increase in long-run productivity can be realized on practical real-time systems. We obtain higher long-run productivity than a decoupled combination of planning and diagnosis.