Query expansion aims to address the term mismatch problem by adding relevant terms to the original query. Traditional approaches, particularly those based on relevance feedback, have shown significant improvements in information retrieval, but rely mainly on lexical information for selecting expansion terms. In contrast, contextual models such as BERT have proven effective in capturing semantic relationships between queries and documents, leading to improved retrieval performance. In this work, we propose an extension of the classical relevance model that incorporates BERT-based contextual information into query model estimation. Specifically, our approach leverages passage-level signals to better estimate the importance of expansion terms. The resulting model is combined with the original relevance model to form a hybrid query model for document retrieval. Experiments on four TREC collections demonstrate that our approach outperforms the baseline relevance model and traditional query expansion methods. It also surpasses two BERT-based query expansion models, CEQE and SRoc, across most evaluation metrics, while achieving competitive performance compared to the dense retrieval model ColBERT. These results highlight the effectiveness of the proposed contextualized hybrid expansion approach.

Term position feature is widely and successfully used in IR and Web search engines, to enhance the retrieval effectiveness. This feature is essentially used for two purposes: to capture query terms proximity or to boost the weight of terms appearing in some parts of a document. In this paper, we are interested in this second category. We propose two novel query-independent techniques based on absolute term positions in a document, whose goal is to boost the weight of terms appearing in the beginning of a document. The first one considers only the earliest occurrence of a term in a document. The second one takes into account all term positions in a document. We formalize each of these two techniques as a document model based on term position, and then we incorporate it into a basic language model (LM). Two smoothing techniques, Dirichlet and Jelinek-Mercer, are considered in the basic LM. Experiments conducted on three TREC test collections show that our model, especially the version based on all term positions, achieves significant improvements over the baseline LMs, and it also often performs better than two state-of-the-art baseline models, the chronological term rank model and the Markov random field model.

Term position feature is widely and successfully used in IR and Web search engines, to enhance the retrieval effectiveness. This feature is essentially used for two purposes: to capture query terms proximity or to boost the weight of terms appearing in some parts of a document. In this paper, we are interested in this second category. We propose two novel query‐independent techniques based on absolute term positions in a document, whose goal is to boost the weight of terms appearing in the beginning of a document. The first one considers only the earliest occurrence of a term in a document. The second one takes into account all term positions in a document. We formalize each of these two techniques as a document model based on term position, and then we incorporate it into a basic language model (LM). Two smoothing techniques, Dirichlet and Jelinek‐Mercer, are considered in the basic LM. Experiments conducted on three TREC test collections show that our model, especially the version based on all term positions, achieves significant improvements over the baseline LMs, and it also often performs better than two state‐of‐the‐art baseline models, the chronological term rank model and the Markov random field model.

RESUME. La plupart des modèles de RI se basent généralement sur la combinaison de trois facteurs dans leur fonction de pondération, qui sont : la fréquence du terme dans le document (TF), la fréquence du terme dans la collection (ou l'IDF) et la taille du document. Quelques approches ont proposé d'intégrer la position du terme dans le document dans l'objectif de surpondérer les termes qui apparaissent au début du document. Dans cet article, nous nous situons dans cette perspective. Précisément, nous proposons deux nouvelles techniques d'estimation du poids d'un terme en se basant sur ses positions dans le document. La première technique considère uniquement la positon de la première apparition du terme dans le document; la seconde technique prend en compte toutes les positions du terme dans le document. Nous avons ensuite intégré les facteurs obtenus dans un modèle de langue pour la RI. Deux techniques de lissage sont considérées dans ce modèle de langue: Dirichlet et Jelinek-Mercer. Les résultats expérimentaux obtenus sur deux collections de test TREC, montrent que notre modèle améliore significativement les deux modèles de langue de base: Dirichlet et Jelinek-Mercer. Notre modèle surpasse aussi un modèle de l'état de l'art, qui est le modèle CTR, basé sur la position du terme dans le document.

Most existing Information Retrieval model including probabilistic and vector space models are based on the term independence hypothesis. To go beyond this assumption and thereby capture the semantics of document and query more accurately, several works have incorporated phrases or other syntactic information in IR, such attempts have shown slight benefit, at best. Particularly in language modeling approaches this extension is achieved through the use of the bigram or n-gram models. However, in these models all bigrams/n-grams are considered and weighted uniformly. In this paper we introduce a new approach to select and weight relevant n-grams associated with a document. Experimental results on three TREC test collections showed an improvement over three strongest state-of-the-art model baselines, which are the original unigram language model, the Markov Random Field model, and the positional language model.

RESUME. Dans cet article nous presentons une nouvelle technique de reformulation de requete. Cette technique considere la requete comme un ensemble de termes composes et un ensemble de termes simples. Pour determiner les termes d'expansion on additionne les poids des relations d'un terme candidat avec chacun des termes de la requete (simple, compose). Un terme candidat est choisi s’il est fortement en relation avec la plupart des termes de la requete. Cette technique est modelisee dans le cadre de modele de langue. Les tests effectues sur deux collections TREC ont montre des ameliorations par rapport a deux modeles: le modele uni-gramme et le modele d’expansion de requetes base sur la mesure KullbackLeibler Divergence (KLD). ABSTRACT. In this paper we present a new technique for query expansion. This technique considers the query as a set of compound terms and a set of simple terms. To determine the expansion terms we add the weight of a term relationships with each of the candidate query terms (simple, compound). A candidate term is selected if it is strongly related with most query terms. This technique is modeled in the context of the language model. Tests on two TREC collections showed improvements compared to the uni-gram model and KLD expansion model. MOTS-CLES : Expansion de requete, Termes composes, Modele de langue, cooccurrence.

RESUME. Plusieurs caracteristiques ont ete utilisees pour estimer la probabilite a priori d’un document comme : la longueur du document, la structure des liens, le facteur temps. Cependant, ces caracteristiques dependent seulement du document lui meme. Or, dans le contexte du web une page web fait partie en general d’un site web. L’idee que nous explorons dans cette article est l’utilisation des caracteristiques du site contenant la page concernee pour conditionner la probabilite a priori de pertinence de la page. Une fois cette probabilite est calculee nous la combinons avec le score obtenu par le contenu de la page web. Cette combinaison des deux evidences est realisee sous le cadre de modele de langue. Afin de valider notre idee, nous avons effectue des tests sur la collection TREC « .GOV » ; ou nous avons compare les differentes versions de notre modele avec deux modeles : le modele uni gramme qui ne considere que le contenu de la page, et le modele combinant le contenu d’une page web et la probabilite a priori de la page obtenu en utilisant seulement une caracteristique sur la page (nombre de liens entrants). Les resultats obtenus montrent que notre modele est tres prometteur.

Most traditional information retrieval systems are based on single terms indexing. However, it is admitted that semantic content of a document (or a query) cannot be accurately captured by a simple set of independent keywords. Although, several works have incorporated phrases or other syntactic information in IR, such attempts have shown slight benefit, at best. Particularly in language modeling approaches this is achieved through the use of the big ram or n-gram models. However, in these models all big rams/n-grams are considered and weighted uniformly. In this paper we introduce a new approach to weight and consider only certain types of N-grams "compound terms". Experimental results on three test collections showed an improvement.

Nowadays, resources on the web increase considerably. In this immense data warehouse, current information retrieval systems do not allow users to obtain results to their requests that meet exactly their needs. Mainly this is due to indexing techniques used (key words, thesaurus). In order to improve the relevance of information retrieval, an ontology-based approach called OBIREX is proposed in this paper. OBIREX is based on the use of ontology of the domain for indexing a collection of documents and the use of semantic links between documents to allow the inference of all relevant documents. This approach is tested on the domain of e-learning of computer science in the context of the semantic web. Some results obtained are also presented.

Usually, user accesses to information on the web using search engines. However, a significant factor, which is the quality of the restored documents, is underestimated by most of these engines. This is due to not taking into account all dimensions of a web document during the indexing and retrieval process. In recent years, language models have been proposed on Information Retrieval and increased in popularity, due to their simplicity, clear probabilistic meaning. In addition these models offer possibility of integrating a priori document information. Several features have been used to estimate the prior probability of a document such as: document length. However these features depend only on the document. A web page is part of website. The idea that we explore in this article is to use website features which contains the concerned page to estimate the page prior.

RESUME. La plupart des systemes de recherche d’information classiques se basent sur une indexation par termes simples. Cependant, ces derniers delivrent beaucoup de resultats en reponse aux requetes des utilisateurs. Ceci est du en partie au fait que le contenu semantique d'un document (ou d'une requete) ne peut pas etre capture precisement par un simple ensemble de mots cles independants. Deux directions sont explorees pour incorporer la semantique dans les modeles de langage. La premiere se base sur l’exploitation des liens entre termes tout en utilisant une meme unite d’indexation. La seconde se base sur l’utilisation d’unites d’indexation plus complexes en plus de l’utilisation de termes simples. Dans ce papier est detaillee l’approche que nous proposons pour incorporer la dimension semantique de document, et qui rentre dans le cadre de la seconde direction.

RESUME. La plupart des systemes de recherche d’information classiques se basent sur une indexation par termes simples. Cependant, ces derniers delivrent beaucoup de resultats en reponse aux requetes des utilisateurs. Ceci est du en partie au fait que le contenu semantique d'un document (ou d'une requete) ne peut pas etre capture precisement par un simple ensemble de mots cles independants. Deux directions sont explorees pour incorporer la semantique dans les modeles de langage. La premiere se base sur l’exploitation des liens entre termes tout en utilisant une meme unite d’indexation. La seconde se base sur l’utilisation d’unites d’indexation plus complexes en plus de l’utilisation de termes simples. Dans ce papier est detaillee l’approche que nous proposons pour incorporer la dimension semantique de document, et qui rentre dans le cadre de la seconde direction.

De nos jours, les ressources disponibles sur le web augmentent considérablement.Dans cet immense entrepôt de données, les systèmes de recherche d'information actuels ne permettent pas de retourner aux utilisateurs les documents répondant exactement à leurs besoins exprimés par une requête sur une collection de documents.Cela est dû, en grande partie, aux techniques d'indexation utilisées (mots-clés, thésaurus, etc.).Afin d'améliorer la pertinence de la recherche d'information, nous proposons dans ce papier une approche qui se base sur l'utilisation d'une ontologie de domaine pour l'indexation d'une base de documents et l'utilisation des liens sémantiques entre documents ou fragments de documents de la collection, pour permettre l'inférence de tous les documents pertinents.Cette approche est testée sur le domaine de l'e-learning de l'informatique dans le contexte du web sémantique.Quelques résultats obtenus sont également présentés.ABSTRACT.Nowadays, resources on the web increase considerably.In this immense data warehouse, current information retrieval systems do not allow users to obtain results to their requests that meet exactly their needs.Mainly this is due to indexing techniques used (key words, thesaurus).In order to improve the relevance of information retrieval, we propose in this paper an approach based on the use of an ontology of the domain for indexing a collection of documents and the use of semantic links between documents or fragments of documents of the collection to allow the inference of all relevant documents.This approach is tested on the domain of e-learning of computer science in the context of the semantic web.Some results obtained are also presented.

Nowadays, resources on the web increase considerably. In this immense data warehouse, current information retrieval systems do not allow users to obtain results to their requests that meet exactly their needs. Mainly this is due to indexing techniques used (key words, thesaurus). In order to improve the relevance of information retrieval, we propose in this paper an approach based on the use of an ontology of the domain for indexing a collection of documents and the use of semantic links between documents or fragments of documents of the collection to allow the inference of all relevant documents. This approach is tested on the domain of e-learning of computer science in the context of the semantic web. Some results obtained are also presented. MOTS-CLÉS : Recherche d'information, ontologie, logique, e-learning, web sémantique.