Simon Dooms

Context-aware recommendations through context and activity recognition in a mobile environment

The mobile Internet introduces new opportunities to gain insight in the user’s environment, behavior, and activity. This contextual information can be used as an additional information source to improve traditional recommendation algorithms. This paper describes a framework to detect the current context and activity of the user by analyzing data retrieved from different sensors available on mobile devices. The framework can easily be extended to detect custom activities and is built in a generic way to ensure easy integration with other applications. On top of this framework, a recommender system is built to provide users a personalized content offer, consisting of relevant information such as points-of-interest, train schedules, and touristic info, based on the user’s current context. An evaluation of the recommender system and the underlying context recognition framework shows that power consumption and data traffic is still within an acceptable range. Users who tested the recommender system via the mobile application confirmed the usability and liked to use it. The recommendations are assessed as effective and help them to discover new places and interesting information.

Comparison of group recommendation algorithms

In recent years recommender systems have become the common tool to handle the information overload problem of educational and informative web sites, content delivery systems, and online shops. Although most recommender systems make suggestions for individual users, in many circumstances the selected items (e.g., movies) are not intended for personal usage but rather for consumption in groups. This paper investigates how effective group recommendations for movies can be generated by combining the group members’ preferences (as expressed by ratings) or by combining the group members’ recommendations. These two grouping strategies, which convert traditional recommendation algorithms into group recommendation algorithms, are combined with five commonly used recommendation algorithms to calculate group recommendations for different group compositions. The group recommendations are not only assessed in terms of accuracy, but also in terms of other qualitative aspects that are important for users such as diversity, coverage, and serendipity. In addition, the paper discusses the influence of the size and composition of the group on the quality of the recommendations. The results show that the grouping strategy which produces the most accurate results depends on the algorithm that is used for generating individual recommendations. Therefore, the paper proposes a combination of grouping strategies which outperforms each individual strategy in terms of accuracy. Besides, the results show that the accuracy of the group recommendations increases as the similarity between members of the group increases. Also the diversity, coverage, and serendipity of the group recommendations are to a large extent dependent on the used grouping strategy and recommendation algorithm. Consequently for (commercial) group recommender systems, the grouping strategy and algorithm have to be chosen carefully in order to optimize the desired quality metrics of the group recommendations. The conclusions of this paper can be used as guidelines for this selection process.

Dynamic generation of personalized hybrid recommender systems

The problem of information overload has been a relevant and active research topic for the past twenty years. Since then, numerous algorithms and recommendation approaches have been proposed, which gives rise to a new type of problem: recommendation algorithm overload. Although hybrid recommendation techniques, which combine the strengths of individual recommenders, have become well-accepted, the procedure of building and tuning a hybrid recommender is still a tedious and time-consuming process. In our work, we focus on dynamically building personalized hybrid recommender systems on an individual user basis. By means of a dynamic online learning strategy we combine the most appropriate recommendation algorithms for a user based on realtime relevance feedback. Learning effectiveness of genetic algorithms, machine learning techniques and other optimization approaches will be studied in both an offline and online setting.

Design and evaluation of a group recommender system

Though most recommender systems make suggestions for individual users, in many circumstances the selected items (e.g., movies) are not for personal usage but rather for consumption in group. In this paper, we present a recommender system for audio-visual content that generates suggestions for groups of people (such as families or friends) in the home environment. In this context, different group recommendation strategies are evaluated for various algorithms and sizes of the group. An offline evaluation proves the assumption that for randomly composed groups the accuracy of all recommendation algorithms decreases if the group size grows. Besides, the results show that the group recommendation strategy which produces the most accurate results is depending on the algorithm that is used for generating individual recommendations. Consequently, if an existing recommender system for individuals is extended to a recommender system for groups, the group recommendation strategy has to be chosen based on the utilized recommendation algorithm in order to maximize the efficiency of the group recommendations.

Time dependency of data quality for collaborative filtering algorithms

The efficiency of personal suggestions generated by collaborative filtering techniques is highly dependent on the quality and quantity of the available consumption data. Extending data sets with additional consumption data (from the past) might enrich the user profiles and generally leads to more accurate recommendations. Although if a considerable amount of profile information is already available and detailed personal preferences can be derived, supplementary consumption data may not have any (or a very limited) added value for the recommendation algorithm. These additional consumption data increase the required storage capacity and the computational load to generate the personal recommendations. Moreover, since personal preferences and the relevance of content items may vary over time, older consumption data might be outdated and lead to inaccurate recommendations. Therefore, we investigate which consumption data are (the most) relevant to feed the conventional collaborative filtering algorithms. For provider-generated content systems, we demonstrate that the accuracy of collaborative filtering algorithms increases by extending user profiles with additional older consumption data. In contrast, we witness the opposite effect for user-generated content systems: involving older consumption data has a negative influence on the recommender accuracy. These results are important for website owners who intend to employ a recommendation system at a minimum storage and computation cost.

A Framework for Dataset Benchmarking and Its Application to a New Movie Rating Dataset

Rating datasets are of paramount importance in recommender systems research. They serve as input for recommendation algorithms, as simulation data, or for evaluation purposes. In the past, public accessible rating datasets were not abundantly available, leaving researchers no choice but to work with old and static datasets like MovieLens and Netflix. More recently, however, emerging trends as social media and smartphones are found to provide rich data sources which can be turned into valuable research datasets. While dataset availability is growing, a structured way for introducing and comparing new datasets is currently still lacking. In this work, we propose a five-step framework to introduce and benchmark new datasets in the recommender systems domain. We illustrate our framework on a new movie rating dataset—called MovieTweetings—collected from Twitter. Following our framework, we detail the origin of the dataset, provide basic descriptive statistics, investigate external validity, report the results of a number of reproducible benchmarks, and conclude by discussing some interesting advantages and appropriate research use cases.

Offline optimization for user-specific hybrid recommender systems

User-specific hybrid recommender systems aim at harnessing the power of multiple recommendation algorithms in a user-specific hybrid scenario. While research has previously focused on self-learning hybrid configurations, such systems are often too complex to take out of the lab and are seldom tested against real-world requirements. In this work, we describe a self-learning user-specific hybrid recommender system and assess its ability towards meeting a set of pre-defined requirements relevant to online recommendation scenarios: responsiveness, scalability, system transparency and user control. By integrating a client-server architectural design, the system was able to scale across multiple computing nodes in a very flexible way. A specific user-interface for a movie recommendation scenario is proposed to illustrate system transparency and user control possibilities, which integrate directly in the hybrid recommendation process. Finally, experiments were performed focusing both on weak and strong scaling scenarios on a high performance computing environment. Results showed performance to be limited only by the slowest integrated recommendation algorithm with very limited hybrid optimization overhead.

Mining cross-domain rating datasets from structured data on twitter

While rating data is essential for all recommender systems research, there are only a few public rating datasets available, most of them years old and limited to the movie domain. With this work, we aim to end the lack of rating data by illustrating how vast amounts of ratings can be unambiguously collected from Twitter. We validate our approach by mining ratings from four major online websites focusing on movies, books, music and video clips. In a short mining period of 2 weeks, close to 3 million ratings were collected. Since some users turned up in more than one dataset, we believe this work to be amongst the first to provide a true cross-domain rating dataset.

OMUS: an optimized multimedia service for the home environment

Media content in home environments is often scattered across multiple devices in the home network. As both the available multimedia devices in the home (e.g., smartphones, tablets, laptops, game consoles, etc.) and the available content (video and audio) is increasing, interconnecting desired content with available devices is becoming harder and home users are experiencing difficulties in selecting interesting content for their current context. In this paper, we start with an analysis of the home environment by means of a user study. Information handling problems are identified and requirements for a home information system formulated. To meet these requirements we propose the OMUS home information system which includes an optimized content aggregation framework, a hybrid group-based contextual recommender system, and an overall web-based user interface making both content and recommendations available for all devices across the home network. For the group recommendations we introduced distinct weights for each user and showed that by varying the weights, the coverage (i.e., items that can be returned by the recommender) considerably increases. Also the addition of genre filter functionality was proven to further boost the coverage. The OMUS system was evaluated by means of focus groups and by qualitative and quantitative performance assessment of individual parts of the system. The modularity of internal components and limited imposed hardware requirements implies flexibility as to how the OMUS system can be deployed (ranging from e.g., embedded in hardware devices or more software services based).

A File-Based Approach for Recommender Systems in High-Performance Computing Environments

Since recommendation systems tackle the problem of information overload, the processing of huge datasets can not be avoided. When these datasets no longer fit into the RAM memory of a computing node, a scalable data storage approach is required. While database systems are frequently used for this goal, they have their disadvantages and when not properly designed may slow down the recommendation process. In this paper we propose an alternative file-based data storage approach that is particularly well suited for a high-performance computing environment where the usage of databases may not always be an option. By breaking down the recommendation process in separate phases and carefully structuring the input and output of each phase, we have build a file-based recommendation system that scales proportional with the number of computing nodes and processor cores available in each node.