Komal Kapoor

Just in Time Recommendations: Modeling the Dynamics of Boredom in Activity Streams

Recommendation methods have mainly dealt with the problem of recommending new items to the user while user visitation behavior to the familiar items (items which have been consumed before) are little understood. In this paper, we analyze user activity streams and show that users temporal consumption of familiar items is driven by boredom. Specifically, users move on to a different item when bored and return to the same item when their interest is restored. To model this behavior we include two latent psychological states of preference for items - sensitization and boredom. In the sensitization state the user is highly engaged with the item, while in the boredom state the user is disinterested. We model this behavior using a Hidden Semi-Markov Model for the gaps between user consumption activities. We show that our model performs much better than the state-of-the-art temporal recommendation models at predicting the revisit time to the item. Moreover, we attribute two main reasons for this: (1) recommending items that are not in the bored state for the user, (2) recommending items where user has restored her interests.

Correlations between Community Structure and Link Formation in Complex Networks

Background Links in complex networks commonly represent specific ties between pairs of nodes, such as protein-protein interactions in biological networks or friendships in social networks. However, understanding the mechanism of link formation in complex networks is a long standing challenge for network analysis and data mining. Methodology/Principal Findings Links in complex networks have a tendency to cluster locally and form so-called communities. This widely existed phenomenon reflects some underlying mechanism of link formation. To study the correlations between community structure and link formation, we present a general computational framework including a theory for network partitioning and link probability estimation. Our approach enables us to accurately identify missing links in partially observed networks in an efficient way. The links having high connection likelihoods in the communities reveal that links are formed preferentially to create cliques and accordingly promote the clustering level of the communities. The experimental results verify that such a mechanism can be well captured by our approach. Conclusions/Significance Our findings provide a new insight into understanding how links are created in the communities. The computational framework opens a wide range of possibilities to develop new approaches and applications, such as community detection and missing link prediction.

A hazard based approach to user return time prediction

In the competitive environment of the internet, retaining and growing ones user base is of major concern to most web services. Furthermore, the economic model of many web services is allowing free access to most content, and generating revenue through advertising. This unique model requires securing user time on a site rather than the purchase of good which makes it crucially important to create new kinds of metrics and solutions for growth and retention efforts for web services. In this work, we address this problem by proposing a new retention metric for web services by concentrating on the rate of user return. We further apply predictive analysis to the proposed retention metric on a service, as a means for characterizing lost customers. Finally, we set up a simple yet effective framework to evaluate a multitude of factors that contribute to user return. Specifically, we define the problem of return time prediction for free web services. Our solution is based on the Coxs proportional hazard model from survival analysis. The hazard based approach offers several benefits including the ability to work with censored data, to model the dynamics in user return rates, and to easily incorporate different types of covariates in the model. We compare the performance of our hazard based model in predicting the user return time and in categorizing users into buckets based on their predicted return time, against several baseline regression and classification methods and find the hazard based approach to be superior.

Measuring spontaneous devaluations in user preferences

Spontaneous devaluation in preferences is ubiquitous, where yesterdays hit is todays affliction. Despite technological advances facilitating access to a wide range of media commodities, finding engaging content is a major enterprise with few principled solutions. Systems tracking spontaneous devaluation in user preferences can allow prediction of the onset of boredom in users potentially catering to their changed needs. In this work, we study the music listening histories of Last.fm users focusing on the changes in their preferences based on their choices for different artists at different points in time. A hazard function, commonly used in statistics for survival analysis, is used to capture the rate at which a user returns to an artist as a function of exposure to the artist. The analysis provides the first evidence of spontaneous devaluation in preferences of music listeners. Better understanding of the temporal dynamics of this phenomenon can inform solutions to the similarity-diversity dilemma of recommender systems.

Predicting MMO Player Gender from In-Game Attributes Using Machine Learning Models

What in-game attributes predict players’ offline gender? Our research addresses this question using behavioral logs of over 4,000 EverQuest II players. The analysis compares four variable sets with multiple combinations of character types (avatar characteristics or gameplay behaviors; primary or nonprimary character), three server types within the game (roleplaying, player-vs-player, and player-vs-environment), and three types of predictive machine learning models (JRip, J48, and Random Tree). Overall, the most highly predictive, interpretable model has an f-measure of 0.94 and suggests the primary character gender and number of male and female characters a player has provide the most prediction value, with players choosing characters to match their own gender. The results also suggest that female players craft, scribe recipes, and harvest items more than male players. While the strength of these findings varies by server type, they are generally consistent with previous research and suggest that players tend to play in ways that are consistent with their offline identities.

Improving classification accuracy of youtube videos by exploiting focal points in social tags

Past literature [1] has shown that problems involving tacit communication among humans and agents are better solved by identifying communication “focal” points based on domain specific human biases. Cast differently, classification of user-generated content into generalized categories is the equivalent of automated programs trying to match human adjudged labels. It seems logical to suspect that identification and incorporation of features generally found salient by humans or “focal points”, can allow an automated agent to better match human adjudged labels in classification tasks. In this paper, we leverage this correspondence, by using domain-specific focal points to further inform the classification algorithms of the inherent human biases. We empirically evaluate our method, by classifying YouTube videos using user-annotated tags. Improvements in classification accuracy over the state-of-the-art classification techniques on using our transformed (using focal points) and highly reduced feature space reveals the value of the approach in subjective classification tasks.

A cognitive basis for theories of intrinsic motivation

Since intelligent agents make choices based on both external rewards and intrinsic motivations, the structure of a realistic decision theory should also present as an indirect model of intrinsic motivation. We have recently proposed a model of sequential choice-making that is grounded in well-articulated cognitive principles. In this paper, we show how our model of choice selection predicts behavior that matches the predictions of state-of-the-art intrinsic motivation models, providing both a clear causal mechanism for explaining its effects and testable predictions for situations where its predictions differ from those of existing models. Our results provide a unified cognitively grounded explanation for phenomena that are currently explained using different theories of motivation, creativity and attention.