Michal Kompan

• PhD.
• Researcher at Kempelen Institute of Intelligent Technologies

We study estimator selection and hyper-parameter tuning in off-policy evaluation. Although cross-validation is the most popular method for model selection in supervised learning, off-policy evaluation relies mostly on theory, which provides only limited guidance to practitioners. We show how to use cross-validation for off-policy evaluation. This c...

Off-policy learning is a framework for optimizing policies without deploying them, using data collected by another policy. In recommender systems, this is especially challenging due to the imbalance in logged data: some items are recommended and thus logged more frequently than others. This is further perpetuated when recommending a list of items,...

In this paper, we study the problem of estimator selection and hyper-parameter tuning in off-policy evaluation. Although cross-validation is the most popular method for model selection in supervised learning, off-policy evaluation relies mostly on theory-based approaches, which provide only limited guidance to practitioners. We show how to use cros...

In this paper, we present results of an auditing study performed over YouTube aimed at investigating how fast a user can get into a misinformation filter bubble, but also what it takes to “burst the bubble”, i.e., revert the bubble enclosure. We employ a sock puppet audit methodology, in which pre-programmed agents (acting as YouTube users) delve i...

In this paper, we present results of an auditing study performed over YouTube aimed at investigating how fast a user can get into a misinformation filter bubble, but also what it takes to "burst the bubble", i.e., revert the bubble enclosure. We employ a sock puppet audit methodology, in which pre-programmed agents (acting as YouTube users) delve i...

The widespread use of temporal aspects in user modeling indicates their importance, and their consideration showed to be highly effective in various domains related to user modeling, especially in recommender systems. Still, past and ongoing research, spread over several decades, provided multiple ad-hoc solutions, but no common understanding of th...

In this paper, we describe a black-box sockpuppeting audit which we carried out to investigate the creation and bursting dynamics of misinformation filter bubbles on YouTube. Pre-programmed agents acting as YouTube users stimulated YouTube's recommender systems: they first watched a series of misinformation promoting videos (bubble creation) and th...

Off-policy learning is a framework for optimizing policies without deploying them, using data collected by another policy. In recommender systems, this is especially challenging due to the imbalance in logged data: some items are recommended and thus logged much more frequently than others. This is further perpetuated when recommending a list of it...

The negative effects of misinformation filter bubbles in adaptive systems have been known to researchers for some time. Several studies investigated, most prominently on YouTube, how fast a user can get into a misinformation filter bubble simply by selecting wrong choices from the items offered. Yet, no studies so far have investigated what it take...

Most of the research in the recommender systems domain is focused on the optimization of the metrics based on historical data such as Mean Average Precision (MAP) or Recall. However, there is a gap between the research and industry since the leading Key Performance Indicators (KPIs) for businesses are revenue and profit. In this paper, we explore t...

Most of the research in the recommender systems domain is focused on the optimization of the metrics based on historical data such as Mean Average Precision (MAP) or Recall. However, there is a gap between the research and industry since the leading Key Performance Indicators (KPIs) for businesses are revenue and profit. In this paper, we explore t...

Cold-start problem, which arises upon the new users arrival, is one of the fundamental problems in today's recommender approaches. Moreover, in some domains as TV or multime-dia-items take long time to experience by users, thus users usually do not provide rich preference information. In this paper we analyze the minimal amount of ratings needs to...

While digital space is a place where users communicate increasingly, the recent threat of COVID-19 infection even more emphasised the necessity of effective and well-organised online environment. Therefore, it is nowadays, more whenever in the past, important to deal with various unhealthy phenomena, that prohibit effective communication and knowle...

One of the most critical problems in e-commerce domain is the information overload problem. Usually, an enormous number of products is offered to a user. The characteristics of this domain force researchers to opt for session-based recommendation methods, from which nearest-neighbors-based (SkNN) approaches have been shown to be competitive with an...

The popularity of e-commerce is increasing day-by-day. In order to provide a seamless experience and tailored offer for the customers, the knowledge of their preferences and behavior is required. The demography of customers is one of the important information used for, e.g., segmentation. To provide optimal service, machine learning is often used f...

Users preferences evolve over the time. This socalled dynamics is a serious challenge which is widely researched in several domains. In these domains, users are usually active for a long period of time and they tend to interact with a wide range of items. To make it more complicated, users preferences are likely to evolve only in some aspects while...

In the domain of e-commerce, acquiring a new customer is generally more expensive than keeping the existing ones. A successful prediction of churn of a specific customer provides an opportunity to change his/her decision to leave. In this paper we propose a novel complex user model focused on the user churn intent prediction. The idea of our model...

One of the important purposes of data mining on the web is to reveal hidden characteristics of users including their behavior. These characteristics are often used to analyze previous user actions, his/her preferences, and also to predict the future behavior. An average user session consists of only few actions, which brings several complications f...

Nowadays, personalized recommendations are widely used and popular. There are a lot of systems in various fields, which use recommendations for different purposes. One of the basic problems is the distrust of users of recommended systems. Users often consider the recommendations as an intrusion of their privacy. Therefore, it is important to make r...

Recommender systems generate items that should beinteresting for the customers. However, recommenders usually fail in the cold-start scenario - when a new item or a newcustomer appears. In our work, we study the cold-start problemfor a new customer. For a cold-start customer we find the most similar customers and use a “their” pre-trained collabora...

When analyzing user implicit feedback in recommender systems, several biases need to be taken into account. A user is influenced by the position (i.e., position bias) or by the appeal of the items (i.e., visual bias). Since images have become an essential part of the Web, the study of their impact on user behavior during the decision-making tasks i...

Identification of typical user behaviour within a web application is a crucial assumption for revealing user characteristics, preferences and habits. Typical and repeating features of user behaviour during his/her interaction with web application can be generalized through behavioural patterns. In this paper, we propose HyBPMine—a novel method, for...

User behaviour in data intensive applications such as the Web-based applications represents a complex set of actions influenced by plenty of factors. Thanks to this complexity, it is extremely hard for human to be able to understand all its aspects. Despite of this, by observing user actions from multiple views, we are able to extract and to model...

The context of a user is a notoriously researched topic in the recom-mender systems community. It greatly influences user preferences and respectively his/her behaviour. The research focuses on the actual influence affecting user and temporal preferences of users. These tell us what the user likes, but fail at describing his/her behavior. We believ...

The personalised recommendations are used routinely in today's e-learning systems especially in computer science and engineering domains. Students' personal characteristics that influence learning styles and collaboration, well accepted in education domain are generally omitted in the domain of recommendation. We propose a methodology for enhancing...

As the Web becomes more and more dynamic, it is interesting to explore the short-term modelling of its user behaviour. Nowadays, it is important to have an information about user’s preferences and needs online. It allows us, in addition to other advantages, also to predict user’s future actions. In this paper we describe the doctoral research focus...

Nowadays, the increasing demand for group recommendations can be observed.In this paper we address the problem of recommendation performance for groupsof users(group recommendation). We focus onthe performance of very Top-N recommendations, which are important when recommending the long lasting items (only a few such items are consumed per session,...

The behavior of users over the web is one of the most relevant and research topic nowadays. Not only mining the user׳s behavior in order to provide better content is popular, but the prediction of the user׳s behavior is interesting and can increase user experience. Moreover, the business clearly desires such information to improve their services. I...

In this paper we propose a novel user model for personalized recommendation in domains, where items are described by multiplecharacteristics (e.g., metadata attributes) and users’ preferences are expressed on the level of items by some kind of explicit feedback (e.g., rating) or derived from implicit user feedback (e.g., time spent on items). The p...

Cold-start problem, which arises upon the new users arrival, is one of the fundamental problems in today's recommender approaches. Moreover, in some domains as TV or multime-dia-items take long time to experience by users, thus users usually do not provide rich preference information. In this paper we analyze the minimal amount of ratings needs to...

It has been shown that social information as group structure or personality characteristics improve the group recommendation. Sometimes no such information is available, specifically when ad-hoc groups are constructed. Moreover, often the items' content is not available (or users' preferences are unknown). In this paper we explore the usage of voti...

The social aspects of the group members are usually omitted in today's group recommenders. In this paper we propose novel approach for the intergroup processes modeling, while the friendship type, user's personality and the group context is considered in order to reflect the group member influence. Moreover, the bi-directional emotional contagion i...

The popularity of group recommender systems has increased in the last years. More and more social activity is generated by users over the Web and thus not only domains as TV, music or holiday resorts are used and researched anymore for group recommendations, but also collaborative learning support, digital libraries and other domains seem to be pro...

Approaches for the personalized recommendations focus mainly on the user's activity over various portals. User's preferences are not dependent on the long term users' history only, but actual user's situation plays crucial role in the user's preferences adjustment and formation. An item liked by the user in some context, can be disliked by the same...

The amount of information available on the web is increasing day by day. Users are overloaded and cannot access desired information in acceptable time. Plenty of approaches for the web personalization, which tries to solve information overload, have been proposed in the literature, but often there are various limitations of such approaches, e.g. co...

With amount of information on the web, users often require functionality able to filter the content according to their preferences. To solve the problem of overwhelmed users we propose a content-based recommender. Our method for the personalized recommendation is dedicated to the domain of news on the Web. We propose an effective representation of...

The intensive research in the personalized recommendation area results into the need for the automatizing routine processes within the recommenders' design or evaluation. In this paper we propose a novel framework for evaluation and experimentation with recommenders. Proposed approach supports basic recommenders' types – content based and collabora...

In this paper we present a proposal including collocations into the preprocessing of the text mining, which we use for the fast news article recommendation and experiments based on real data from the biggest Slovak newspaper. The news article section can be predicted based on several article’s characteristics as article name, content, keywords etc....

The information overloading is one of the serious problems nowadays. We can see it in various domains including business. Especially news represent area where information overload currently prevents effective information gathering on daily basis. This is more significant in connection to the web and news web-based portals, where the quality of the...

Fakulta informatiky a informačných technológií, Slovenská technická univerzita v Bratislave, Ilkovičova 3, 842 16 Bratislava Abstrakt Spotreba energie si neustále vyžaduje našu pozornosť. Zdroje nie sú nevyčerpateľné a tak riešením nemôže byť len výstavba nových elektrární. Práve inteligentné domácnosti majú potenciál podieľať sa na optimalizácii s...