Loic Lecerf

Scalable indexing for layout based document retrieval and ranking

In this paper we propose a schema for querying large documents collections by document layout. We develop a model of layout indexing of a collection adapted for the quick retrieval of top k relevant documents. Fort the sake of scalability, we avoid a direct evaluation of the similarity between a query and each document in the collection; their similarity is instead approximated by the similarity between their projections on the set of representative blocks which are inferred from the collection on the indexed step. The technique also proposes new functions for the relevance ranking and the cluster pruning that ensure a scalable retrieval and ranking.

Scalable Feature Selection for Multi-class Problems

Scalable feature selection algorithms should remove irrelevant and redundant features and scale well on very large datasets. We identify that the currently best state-of-art methods perform well on binary classification tasks but often underperform on multi-class tasks. We suggest that they suffer from the so-called accumulative effect which becomes more visible with the growing number of classes and results in removing relevant and unredundant features. To remedy the problem, we propose two new feature filtering methods which are both scalable and well adapted for the multi-class cases. We report the evaluation results on 17 different datasets which include both binary and multi-class cases.

Semi-supervised visual clustering for spherical coordinates systems

In this paper we propose a method that combines the advanced data analysis of the automatic statistical methods and the flexibility and manual parameter tuning of interactive visual clustering. We present the Semi-Supervised Visual Clustering (SSVC) interface; its main contribution is the learning of the optimal projection distance metric for the star and spherical coordinate visualization systems. Beyond the conventional manual setting, it couples the visual clustering with the automatic setting where the projection distance metric is learned from the available set of user feedbacks in the form of either item similarities or direct item annotations. Moreover, SSVC interface allows for the hybrid setting where some parameters are manually set by the user while the remaining parameters are determined by the optimal distance algorithm.

Document annotation by active learning techniques

We present a system for the semantic annotation of layout-oriented documents, with an integrated learning component. We introduce probabilistic learning methods on tree-like documents and we present different active learning techniques for training document annotation models. We report some preliminary results of deploying such active learning techniques on an important case of document collection annotation.

Scalable Feature Extraction from Noisy Documents

We cope with the metadata recognition in layout-oriented documents. We address the problem as a classification task and propose a method for automatic extraction of relevant features, in presence of content and structural noise, caused by scanning, OCR and segmentation problems. The method is based on the automatic analysis of documents and requires no particular preprocessing. %nor the domain knowledge. The method mines the documents and determines frequent patterns, which are both literal patterns and their generalization. We also propose a sampling technique which processes a sample of documents and uses the {\it Chernoff bounds} to estimate the pattern frequency in the entire dataset.As a number of frequent patterns as feature candidates grows, the method applies a scalable feature selection method to determine the most relevant features to a given classification task. A series of evaluations on two collections show that the method performs comparably to the manual work on rule writing made by domain experts.