Programming language processing is a field of increasing interest, as more and more models become available, either to address specific tasks or to acquire general knowledge which can then be fine-tuned on downstream tasks. All these models are based on architectures that come from the field of natural language processing, most of them being built on the distributional hypothesis from linguistics. Although this transition from one field to another appears to have occurred naturally, it is not so obvious to claim that this hypothesis will be appropriate for extracting semantics from programs. In this paper, we investigate to which extent a distributional hypothesis can be applied to code embedding. To this end, we first formulate various hypotheses adapted to the specific information contained in programming languages. We then provide a framework to evaluate the effectiveness of these hypotheses through the quality of the resulting embedding spaces. This framework is based on the doc2vec model as a generic language model, as its implementation of the original distributional hypothesis is easy to understand and to adapt to any new ones. Among other tools, we propose a new evaluation method based on program analogies, which measures how well the models capture the underlying structure and meaning of the code. We apply the proposed framework to a set of (distributional) hypotheses and show that we can rule out certain hypotheses in favor of others. Specifically, our study indicates that instruction-based hypotheses capture less semantic information than token-based ones. Furthermore, we observe that distributional hypotheses on tokens are effective in both source code, execution traces, and abstract syntax trees. Additionally, we find that the semantics captured on programs by these three hypotheses are of comparable levels and natures.

Different important studies in Web search results clustering have recently shown increasing performances motivated by the use of external resources. Following this trend, we present a new algorithm called Dual C-Means, which provides a theoretical background for clustering in different representation spaces. Its originality relies on the fact that external resources can drive the clustering process as well as the labeling task in a single step. To validate our hypotheses, a series of experiments are conducted over different standard datasets and in particular over a new dataset built from the TREC Web Track 2012 to take into account query logs information. The comprehensive empirical evaluation of the proposed approach demonstrates its significant advantages over traditional clustering and labeling techniques.

We present in this paper a new approach for the automatic generation of lexical structures from texts. This tedious task is based on the strong hypothesis that simple statistical observations on textual usages can provide pieces of semantics about the lexicon. Using such "naive" observations only, we propose a (pre)-topological framework to formalize and combine various hypothesis on textual data usages and then to derive a structure similar to usual lexical knowledge basis such as WordNet. In addition we also consider the evaluation problem for obtained lexical structures ; a multi-level evaluation strategy is proposed that measures the fitting between a given reference structure and automatically generated structures on different point of views : intrinsic/structural and application-based points of view. The evaluation strategy is then used to quantify the contribution of the new structuring approach with respect to the corresponding solution proposed by (Sanderson et al. 2000) on two case studies that differs on the domain and the size of the lexicon.

Ephemeral clustering has been studied for more than a decade, although with low user acceptance. According to us, this situation is mainly due to (1) an excessive number of generated clusters, which makes browsing difficult and (2) low quality labeling, which introduces imprecision within the search process. In this paper, our motivation is twofold. First, we propose to reduce the number of clusters of Web page results, but keeping all different query meanings. For that purpose, we propose a new polythetic methodology based on an informative similarity measure, the InfoSimba, and a new hierarchical clustering algorithm, the HISGK-means. Second, a theoretical background is proposed to define meaningful cluster labels embedded in the definition of the HISGK-means algorithm, which may elect as best label, words outside the given cluster. To confirm our intuitions, we propose a new evaluation framework, which shows that we are able to extract most of the important query meanings but generating much less clusters than state-of-the-art systems.

This paper deals with clustering for multi-view data, i.e. objects described by several sets of variables or proximity matrices. Many important domains or applications such as information retrieval, biology, chemistry and marketing are concerned by this problematic. The aim of this data mining research field is to search for clustering patterns that perform a consensus between the patterns from different views. This requires to merge information from each view by performing a fusion process that identifies the agreement between the views and solves the conflicts. Various fusion strategies can be applied, occurring either before, after or during the clustering process. We draw our inspiration from the existing algorithms based on a centralized strategy. We propose a fuzzy clustering approach that generalizes the three fusion strategies and outperforms the main existing multi-view clustering algorithm both on synthetic and real datasets.

This paper deals with overlapping clustering, a trade off between crisp and fuzzy clustering. It has been motivated by recent applications in various domains such as information retrieval or biology. We show that the problem of finding a suitable coverage of data by overlapping clusters is not a trivial task. We propose a new objective criterion and the associated algorithm OKM that generalizes the k-means algorithm. Experiments show that overlapping clustering is a good alternative and indicate that OKM outperforms other existing methods.