Résumé: In this paper we present preliminary work studying the interactions of a community of focussed forum users and their discussions around several television series. We use k-means clustering and a number of novel citation-analysis inspired measures to perform bottom-up role detection on this community of TV fans, and show that these emergent roles correspond well with the positions assigned to users using traditional graph-based measures.

Résumé: In this paper, we address the issue of privacy preserving data-mining. Specifically, we consider a scenario where each member j of T parties has its own private database. The party j builds a private classifier hj for predicting a binary class variable y. The aim of this paper consists of aggregating these classifiers hj in order to improve individual predictions. More precisely, the parties wish to compute an efficient linear combination over their classifier in a secure manner.

Commentaires: special issue of selected papers from the 1st ECML/PKDD Workshop on Privacy and Security issues in Data Mining and Machine Learning

Résumé: Analyzing the social roles inside on-line communities became a big challenge nowadays. The on-line communities formed around exchange platforms (e.g., forums) create an increasing source of data for analyzing user’s behavior. This paper proposes an exploratory analysis of communities in news website based on its sub-communities. Actually, we assume that people who participate in forum debate in news websites focus their participation in one or a very few topics (also called context) i.e., they formed the sub-communities. These sub-communities, will help us to find the contextual celebrity: the pertinent users in the sub-communities. We based our analysis on a dataset composed by 11,143 users writing more than 35,000 posts on 57 different forums grouped in 3 topics, and on social networks enriched with relations extracted from the content of the users’ posts.

Résumé: The expansion of web user roles is, nowadays, a fact due to the ability of users to interact, discuss, exchange ideas and opinions, form social networks with each other through the web. The interaction level among users leads to the appearance of several social roles which can be characterized as positions, behaviors, or virtual identities. These roles may be developed in social networks, and they keep changing and evolving over time. In this article, a survey of the state-of-the-art approaches is presented regarding the identification of roles within the context of a social network. It is shown that social roles exist as a function of each other; they appear and evolve through user interaction. Different approaches are analyzed and additional characteristics that should be taken into account during the roles analysis are discussed.

Résumé: Web forums are a huge data source. They allow people to interact with unknown individuals. Studying forums shows that the interaction is not obvious only through the structure but also through the content of the post. Taking into account this observation, we extract a social network with different kinds of relationships i.e. the structural relation, the name and the text quotations relation. We present here the promising results we obtain, and the difficulties we face while extracting the quotations in this kind of textual content. These results are obtained from real data (from two information websites) which make the validation difficult. So, we create a validation protocol composed of two steps and based on human raters. Finally, we will see the objective of this work which is understanding interactions in order to extract the social roles of individuals.

---------

Ref HAL: hal-00616247_v2
Résumé: We propose a system which employs conceptual knowledge to improve topic models by removing unrelated words from the simplified topic description. We use WordNet to detect which topical words are not conceptually similar to the others and then test our assumptions against human judgment. Results obtained on two different corpora in different test conditions show that the words detected as unrelated had a much greater probability than the others to be chosen by human evaluators as not being part of the topic at all. We prove that there is a strong correlation between the said probability and an automatically calculated topical fitness and we discuss the variation of the correlation depending on the method and data used.

---------

Ref HAL: hal-00616245_v1
Résumé: The growing number of statistical topic models led to the need to better evaluate their output. Traditional evaluation means estimate the model's fitness to unseen data. It has recently been proven than the output of human judgment can greatly differ from these measures. Thus the need for methods that better emulate human judgment is crucial. In this paper we present a system that computes the utility of individual topics from a given model on the basis of information drawn from a given ontology, in this case WordNet. The notion of utility is regarded as the ability to attribute concepts to each topic and separate words related to the topic from the unrelated ones based on that concept. In multiple experiments we prove the correlation between the automatic evaluation method and the answers received from human evaluators, for two different corpora and difficulty levels. By changing the evaluation focus from a statistical one to a conceptual one we were able to detect which topics are conceptually meaningful and rank them accordingly.

---------