Pertinence d'un document | EPFL Graph Search

Une recherche scientifique ne peut avoir cette qualité de « scientifique » qu’après avoir répondu à un ensemble de critères. Parmi ces critères, la pertinence occupe une place primordiale, elle constitue la moelle épinière de toute recherche académique. En effet, c’est le degré de la pertinence qui donne beaucoup de crédibilité à un tel document. Par elle ce dernier aura beaucoup de chance d’être publiée et d’être tenue comme référence scientifique de recherche. 

La pertinence peut être définie de plusieurs façons selon le champ scientifique dont on parle. Dans le domaine de la recherche documentaire,  (Sylvie, 2002) estime qu’il s’agit du « degré de corrélation entre une question et la réponse apportée ». Autrement dit, il s’agit du point de rencontre entre deux notions transversales de recherche de l’information à savoir : « le bruit documentaire » et « le silence documentaire ». On parle du silence documentaire lorsqu’une requête de recherche dans une base de données ou dans un

Integration of scientific and social networks

Ehsaneddin Asgari

In this paper, we address the problem of scientific-social network integration to find a matching relationship between members of these networks (i.e. The DBLP publication network and the Twitter social network). This task is a crucial step toward building a multi environment expert finding system that has recently attracted much attention in Information Retrieval community. In this paper, the problem of social and scientific network integration is divided into two sub problems. The first problem concerns finding those profiles in one network, which presumably have a corresponding profile in the other network and the second problem concerns the name disambiguation to find true matching profiles among some candidate profiles for matching. Utilizing several name similarity patterns and contextual properties of these networks, we design a focused crawler to find high probable matching pairs, then the problem of name disambiguation is reduced to predict the label of each candidate pair as either true or false matching. Because the labels of these candidate pairs are not independent, state-of-the-art classification methods such as logistic regression and decision tree, which classify each instance separately, are unsuitable for this task. By defining matching dependency graph, we propose a joint label prediction model to determine the label of all candidate pairs simultaneously. Two main types of dependencies among candidate pairs are considered for designing the joint label prediction model which are quite intuitive and general. Using the discriminative approaches, we utilize various feature sets to train our proposed classifiers. An extensive set of experiments have been conducted on six test collection collected from the DBLP and the Twitter networks to show the effectiveness of the proposed joint label prediction model.

Springer2014

Simple Unsupervised Keyphrase Extraction using Sentence Embeddings

Kamil Bennani-Smires, Martin Jaggi, Claudiu-Cristian Musat

Keyphrase extraction is the task of automatically selecting a small set of phrases that best describe a given free text document. Keyphrases can be used for indexing, searching, aggregating and summarizing text documents, serving many automatic as well as human-facing use cases. Existing supervised systems for keyphrase extraction require large amounts of labeled training data and generalize very poorly outside the domain of the training data. At the same time, unsupervised systems found in the literature have poor accuracy, and often do not generalize well, as they require the input document to belong to a larger corpus also given as input. Furthermore, both supervised and unsupervised methods are often too slow for real-time scenarios and suffer from over-generation. Addressing these drawbacks, in this paper, we introduce an unsupervised method for keyphrase extraction from single documents that leverages sentence embeddings. By selecting phrases whose semantic embeddings are close to the embeddings of the whole document, we are able to separate the best candidate phrases from the rest. We show that our embedding-based method is not only simpler, but also more effective than graph-based state of the art systems, achieving higher F-scores on standard datasets. Simplicity is a significant advantage, especially when processing large amounts of documents from the Web, resulting in considerable speed gains. Moreover, we describe how to increase coverage and diversity among the selected keyphrases by introducing an embedding-based maximal marginal relevance (MMR) for new phrases. A user study including over 200 votes showed that, although reducing the phrase semantic overlap leads to no gains in terms of F-score, our diversity enriched selection is preferred by humans.

2018

Intégration de connaissances syntaxiques et sémantiques dans les représentations vectorielles de textes

Romaric Besançon

The notion of similarity between texts is fundamental for many applications of Natural Language Processing. For example, this notion is particularly useful for the applications designed for the management of information in large textual databases, such as Information Retrieval or Automatic Text Structuring. Information Retrieval is the search of the most relevant documents according to an information need expressed by a query, and can be implemented by the search of the documents most similar to the query. Automatic Text Structuring is often viewed as the clustering of documents according to their similarity measures. The similarity between documents relies on their representation. The most used textual representation is the Vector Space model, in which each document is represented by a vector, and the similarity between documents is then computed by a distance measure in this space, for instance, the cosine of the vectors representing the documents. We first present several vector space models used for the computation of similarities between documents. Then, we focus on the problem of the integration of additional knowledge in the vector space representation, and the impact of this integration on the results obtained for several tasks. We fist consider the integration of co-occurrences in the representation model, and we focus on the DSIR model (Distributional Semantics based In-formation Retrieval). We show that this model has a probabilistic theoretical basis. We then consider the use of syntactic information to compute the co-occurrence frequencies. We also consider the integration of knowledge about compounds in the representation, taking into account morpho-syntactic and semantic variants of the considered compounds. We finally address the issue of word sense disambiguation, using synonymy relations to derive a vector space representation for which each dimension is associated to a meaning and not to a term. For all these methods, we propose several evaluations : we first consider a validation for the notion of similarity derived from a vector space representation, in a multi-lingual framework : the idea is to verify that the similarity between two documents in one language is close to the similarity between their translations in another language. We also propose an evaluation of the different models considered in a standard Information Retrieval evaluation framework. We finally consider the evaluation of the models on a Word Sense Disambiguation task.

EPFL2002

Intégration de connaissances syntaxiques et sémantiques dans les représentations vectorielles de textes

Romaric Besançon

EPFL2002