Ante Odić

Predicting and Detecting the Relevant Contextual Information in a Movie-Recommender System

Context-aware recommender system (CARS) is a highly researched and implemented way of providing a personalized service that helps users to find their desired content. One of the remaining issues is how to decide which contextual information to acquire and how to incorporate it into CARS. While the relevant contextual information will improve the recommendations, the irrelevant contextual information could have a negative impact on the recommendation accuracy. By testing the independence between the contextual variable on the users’ ratings for items, we can detect its relevanceandimpactonthefeedbackfortheitemconsumedinthatspecificcontext.Inthisarticle,we propose several new theoretical concepts that should help deciding which information to use, as well as a methodology for detecting which contextual information contributes to explaining the variance in the ratings, based on statistical testing. The experiment was conducted on the real movie dataset that contains 12 different pieces of contextual information. We used two statistical tests with power analysis for the detection, and three contextualized matrix-factorization algorithms with slightly different reasoning for the prediction of ratings. The results showed a significant difference in the prediction of ratings in the context that was detected as relevant by our method, and the one that was detected as irrelevant, pointing to the importance of the power analysis and the benefits of the proposed method in the case of a small dataset.

The impact of weak ground truth and facial expressiveness on affect detection accuracy from time-continuous videos of facial expressions

In this paper we address two issues concerning real-world time-continuous emotion detection from videos of users’ faces: (i) the impact of weak ground truth on the emotion detection accuracy and (ii) the impact of the users’ facial expressiveness on the emotion detection accuracy. We implemented an appearance-based emotion detection algorithm that uses Gabor features and a k nearest neighbors classifier. We tested the performance of this algorithm on two datasets with different ground truth strengths (a firm ground truth dataset and a weak ground truth dataset). Then we split the dataset into three subsets reflecting different levels of users’ facial expressiveness (low, mid and high) and performed separate emotion detection.

Emotion-Aware Recommender Systems – A Framework and a Case Study

Recent work has shown an increase of accuracy in recommender systems that use emotive labels. In this paper we propose a framework for emotion-aware recommender systems and present a survey of the results in such recommender systems. We present a consumption-chain-based framework and we compare three labeling methods within a recommender system for images: (i) generic labeling, (ii) explicit affective labeling and (iii) implicit affective labeling.

How to improve the statistical power of the 10-fold cross validation scheme in recommender systems

At this stage development of recommender systems (RS), an evaluation of competing approaches (methods) yielding similar performances in terms of experiment reproduction is of crucial importance in order to direct the further development toward the most promising direction. These comparisons are usually based on the 10-fold cross validation scheme. Since the compared performances are often similar to each other, the application of statistical significance testing is inevitable in order to not to get misled by randomly caused differences of achieved performances. For the same reason, to reproduce experiments on a different set of experimental data, the most powerful significance testing should be applied. In this work we provide guidelines on how to achieve the highest power in the comparison of RS and we demonstrate them on a comparison of RS performances when different variables are contextualized.

Emotions and Personality in Personalized Services

Personalized systems traditionally used the traces of user interactions to learn the user model, which was used by sophisticated algorithms to choose the appropriate content for the user and the situation. Recently, new types of user models started to emerge, which take into account more user-centric information, such as emotions and personality. Initially, these models were conceptually interesting but of little practical value as emotions and personality were difficult to acquire. However, with the recent advancement in unobtrusive technologies for the detection of emotions and personality these models are becoming interesting both for researchers and practitioners in the domain of personalized systems. This chapter introduces the book, which aims at covering the whole spectrum of knowledge needed to research and develop emotionand personality-aware systems. The chapters cover (i) psychological theories, (ii) computational methods for the unobtrusive acquisition of emotions and personality, (iii) applications of personalized systems in recommender systems, conversational systems, music information retrieval, and e-learning, (iv) evaluation methods, and (v) privacy issues. M. Tkalčič (B) Department of Computational Perception, Johannes Kepler University, Altenbergerstrasse 69, Linz, Austria e-mail: marko.tkalcic@jku.at B. De Carolis ·M. de Gemmis Department of Computer Science, University of Bari Aldo Moro, Via E.Orabona, 4, 70125 Bari, Italy e-mail: berardina.decarolis@uniba.it M. de Gemmis e-mail: marco.degemmis@uniba.it A. Odić Outfit7 (Slovenian Subsidiary Ekipa2 D.o.o.), Ljubljana, Slovenia e-mail: ante.odic@outfit7.com A. Košir Faculty of Electrical Engineering, University of Ljubljana, Tržaška 25, 1000 Ljubljana, Slovenia e-mail: andrej.kosir@fe.uni-lj.si

Introduction to Emotions and Personality in Personalized Systems

Personalized systems traditionally used the traces of user interactions to learn the user model, which was used by sophisticated algorithms to choose the appropriate content for the user and the situation. Recently, new types of user models started to emerge, which take into account more user-centric information, such as emotions and personality. Initially, these models were conceptually interesting but of little practical value as emotions and personality were difficult to acquire. However, with the recent advancement in unobtrusive technologies for the detection of emotions and personality these models are becoming interesting both for researchers and practitioners in the domain of personalized systems. This chapter introduces the book, which aims at covering the whole spectrum of knowledge needed to research and develop emotion- and personality-aware systems. The chapters cover (i) psychological theories, (ii) computational methods for the unobtrusive acquisition of emotions and personality, (iii) applications of personalized systems in recommender systems, conversational systems, music information retrieval, and e-learning, (iv) evaluation methods, and (v) privacy issues.

Impact of the Context Relevancy on Ratings Prediction in a Movie-Recommender System

Recommender systems are a popular and a highly researched way of helping users get to their desired content in the huge amount of available data, and services online. Understanding the situation in which users consume the items was shown to improve the recommendation process. For that reason, context-aware recommender system (CARS) employs contextual information in order to enhance the users model and to improve the recommendations. An issue that is still open is how to decide which pieces of contextual information to acquire and how to incorporate them into CARS, since using irrelevant piece of contextual information could have a negative impact on the recommendations. We propose a methodology for detecting which pieces of contextual information contribute to explaining the variance in the ratings, based on statistical testing. We also inspect the impact of the detected relevant pieces of contextual information on the ratings prediction based on the matrix-factorization algorithm. The experiment was conducted on the MovieAT database. The results showed a significant difference in the ratings prediction using the relevant and the irrelevant pieces of contextual information. We also confirmed the positive impact of the relevant, and negative impact of the irrelevant pieces of contextual information with respect to the uncontextualized model.

Affective and Personality Corpora

In this chapter we describe publicly available datasets with personality and affective parameters relevant to the research questions covered by this book. We briefly describe the available data, acquisition procedure, and other relevant details of these datasets. There are three datasets acquired through the users’ natural interaction with different services: LDOS CoMoDa, LJ2M and myPersonality. Two datasets were acquired in controlled, laboratory settings: LDOS PerAff-1 and DEAP. Finally, we also mention four stimuli datasets from the Media Core project: ANET, IADS, ANEW, IAPS, as well as the 1000 songs dataset. We summarise this information for a quick reference to researchers interested in using these datasets or preparing the acquisition procedure of their own.

EMPIRE 2015: Workshop on Emotions and Personality in Personalized Systems

The EMPIRE workshop focuses on recommender systems (and other personalized systems) that take advantage of user-centric properties, such as emotions and personality. The workshop is organized as a focused mini-conference with technical and position papers. The goal is to gather the scattered work under a common umbrella and take advantage of the discussion time to draw future research opportunities.

Exploiting implicit affective labeling for image recommendations

Recent work has shown an increase of accuracy in recommender systems that use affective labels. In this paper we compare three labeling methods within a recommender system for images: (i) generic labeling, (ii) explicit affective labeling and (iii) implicit affective labeling. The results show that the recommender system performs best when explicit labels are used. However, implicitly acquired labels yield a significantly better performance of the CBR than generic metadata while being an unobtrusive feedback tool.