Among other criteria, a pattern may be interesting if it is not redundant with other discovered patterns. A general approach to determining redundancy is to consider a probabilistic model for frequencies of patterns, based on those of patterns already mined, and compare observed frequencies to the model. Such probabilistic models include the independence model, partition models or more complex models which are approached via randomization for a lack of an adequate tool in probability theory allowing a direct approach. We define constrained independence, a generalization to the notion of independence. This tool allows us to describe probabilistic models for evaluating redundancy in frequent itemset mining. We provide algorithms, integrated within the mining process, for determining non-redundant itemsets. Through experimentations, we show that the models used reveal high rates of redundancy among frequent itemsets and we extract the most interesting ones.

One of the prevalent learning tasks involving images is content-based image classification. This is a difficult task especially because the low-level features used to digitally describe images usually capture little information about the semantics of the images. In this paper, we tackle this diffic ulty by enriching the semantic content of the image representation by using external knowledge. The underlying hypothesis of our work is that creating a more semantically rich representation for images would yield higher machine learning performances, without the need to modify the learning algorithms themselves. The external semantic information is presented under the form of non-positional image labels, therefore positioning our work in a weakly supervised context. Two approaches are proposed: the first one leverages the labels into the visual vocabulary construction algorithm, the result being dedicated visual vocabularies. The second approach adds a filtering phase as a pre-processing of the vocabulary construction. Known positive and known negative sets are constructed and features that are unlikely to be associated with the objects denoted by the labels are filtered. We apply our proposition to the task of content-based image classification and we show that semantically enriching the image representation yields higher classification performances than the baseline representation.

In this paper, we propose a new time-aware dissimilarity measure that takes into account the temporal dimension. Observations that are close in the description space, but distant in time are considered as dissimilar. We also propose a method to enforce the segmentation contiguity, by introducing, in the objective function, a penalty term inspired from the Normal Distribution Function. We combine the two propositions into a novel time-driven constrained clustering algorithm, called TDCK-Means, which creates a partition of coherent clusters, both in the multidimensional space and in the temporal space. This algorithm uses soft semi-supervised constraints, to encourage adjacent observations belonging to the same entity to be assigned to the same cluster. We apply our algorithm to a Political Studies dataset in order to detect typical evolution phases. We adapt the Shannon entropy in order to measure the entity contiguity, and we show that our proposition consistently improves temporal cohesion of clusters, without any significant loss in the multidimensional variance.

Many studies have shown the limits of the support/confidence framework used in Apriori‐like algorithms to mine association rules. There are a lot of efficient implementations based on the antimonotony property of the support, but candidate set generation (e.g., frequent item set mining) is still costly. In addition, many rules are uninteresting or redundant and one can miss interesting rules like nuggets. We are thus facing a complexity issue and a quality issue. One solution is to not use frequent itemset mining and to focus as soon as possible on interesting rules using additional interestingness measures. We present here a formal framework that allows us to make a link between analytic and algorithmic properties of interestingness measures. We introduce the notion of optimonotony in relation with the optimal rule discovery framework. We then demonstrate a necessary and sufficient condition for the existence of optimonotony. This result can thus be applied to classify the measures. We study the case of 39 classical measures and show that 31 of them are optimonotone. These optimonotone measures can thus be used with an underlying pruning strategy. Empirical evaluations show that the pruning strategy is efficient and leads to the discovery of nuggets using an optimonotone measure and without the support constraint.

Today's world is characterized by the multiplicity of interconnections through many types of links between the people, that is why mining social networks appears to be an important topic. Extracting information from social networks becomes a challenging problem, particularly in the case of the discovery of community structures. Mining bibliographical data can be useful to find communities of researchers. In this paper we propose a formal definition to consider the similarity and dissimilarity between individuals of a social network and how a graph-based clustering method can extract research communities from the DBLP database.

Clustering methods usually require to know the best number of clusters, or another parameter, e.g. a threshold, which is not ever easy to provide. This paper proposes a new graph-based clustering method called GBC which detects automatically the best number of clusters, without requiring any other parameter. In this method based on regions of influence, a graph is constructed and the edges of the graph having the higher values are cut according to a hierarchical divisive procedure. An index is calculated from the size average of the cut edges which self-detects the more appropriate number of clusters. The results of GBC for 3 quality indices (Dunn, Silhouette and Davies-Bouldin) are compared with those of K-Means, Ward's hierarchical clustering method and DBSCAN on 8 benchmarks. The experiments show the good performance of GBC in the case of well separated clusters, even if the data are unbalanced, non-convex or with presence of outliers, whatever the shape of the clusters.

Associative classification presents various methods whose common characteristic is the class prediction from the class association rules (rules whose consequent one is one of the class modalities). According to and, this new approach offers better results than the traditional approaches based on rules such as the decision trees. It also offers a great flexibility with the unstructured data. However, this approach suffers from a huge mass of generated rules which leads to a waste of time and space. In this work, we propose a new associative classification method. This method is based on FCP-Growth-P, an algorithm which generates only class itemsets and integrates for pruning the specialization condition of Li. Thus one saves both execution time and storage space. The phase of classification is based on a reduced base of the most significant rules leading to each class, which ensures the speed of the method. Examples are classified using the results given by the vote of these various rules weighted by its quality measure.

Many studies have shown the limits of support/confidence framework used in Apriori-like algorithms to mine association rules. One solution to cope with this limitation is to get rid of frequent itemset mining and to focus as soon as possible on interesting rules. Many works have focused on the algorithmic properties of the confidence. In particular, the all-confidence which is a transformation of the confidence, has the antimonotone property. In this paper, we generalize the all-confidence by associating to any measure its corresponding all-measure. We present a formal framework which allows us to make the link between analytic and algorithmic properties of the all-measure. We then propose the notion of all-monotony which corresponds to the monotony property of the all-measures. Our results show that although being very interesting, all-monotony is a demanding property.