Effective and precise passage retrieval is crucial in contemporary information retrieval systems, particularly in light of the exponential expansion of digital document collections. Conventional approaches relying on lexical matching frequently struggle to capture semantic subtleties, often overlooking passages that are semantically pertinent but lexically divergent from user queries. Recent advancements in transformer-based models, such as BERT and its variants, have substantially enhanced retrieval performance by facilitating a deeper semantic comprehension. Nonetheless, the computational demands of these models present scalability challenges when applied to large datasets. In this work, we propose an innovative passage retrieval framework that seamlessly integrates lexical-semantic relevance with contextual information, markedly boosting retrieval precision and coherence. Our methodology unfolds in three stages: first, a Cross-Encoder BERT model assesses semantic relevance by analyzing fine-grained interactions between queries and passages. Second, a Bi-Encoder architecture generates initial embeddings, which are subsequently refined using a Graph Attention Network (GAT) to incorporate structural and contextual relationships among passages. Third, a dynamic reranking mechanism improves ranking by merging semantic relevance scores with contextual similarity evaluations, ensuring that retrieved passages are both contextually coherent and semantically precise. Evaluations conducted on benchmark datasets against state-of-the-art baselines reveal that our hybrid approach enhances retrieval accuracy and coherence. Furthermore, by integrating graph-based embedding refinements, our model mitigates the scalability constraints typically associated with Cross-Encoder architectures.

In recent years, social networks have become the most exploited sources of information, such as Facebook, Instagram, LinkedIn, and Twitter, have been considered to be the main sources of non-credible information. The presence of false information in these social networks has a very negative impact on the credibility of conversations. In this article, we propose a new approach, called CREDEEP (Conversational credibility based on deep learning). CREDEEP is based on: (i) the combination of post and user features in order to detect credible and not credible conversations; (ii) the integration of multi-dense layers to represent features more deeply and to improve the results. In order to study the performance of our approach, we have used the standard PHEME dataset. We compared our approach with the main approaches we have studied in the literature. The obtained results confirm the performance of our model in terms of precision, recall, and F1-measure.

In the medical field, a doctor must have a comprehensive knowledge by reading and writing narrative documents, and he is responsible for every decision he takes for patients. Unfortunately, it is very tiring to read all necessary information about drugs, diseases and patients due to the large amount of documents that are increasing every day. Consequently, so many medical errors can happen and even kill people. Likewise, there is such an important field that can handle this problem, which is the information extraction. There are several important tasks in this field to extract the important and desired information from unstructured text written in natural language. The main principal tasks are named entity recognition and relation extraction since they can structure the text by extracting the relevant information. However, in order to treat the narrative text we should use natural language processing techniques to extract useful information and features. In our paper, we introduce and discuss the several techniques and solutions used in these tasks. Furthermore, we outline the challenges in information extraction from medical documents. In our knowledge, this is the most comprehensive survey in the literature with an experimental analysis and a suggestion for some uncovered directions.

Since the development of information technologies, there is a huge amount of electronic documents that was written by medical specialists and are rich of useful information needed to make critical decisions in several medical tasks. Thus, a doctor must have a big knowledge and he is responsible for every decision he takes for patients. In fact, the doctor should read, with full concentration, many electronic narrative documents to collect the necessary information. Unfortunately, it's too tiring to read all necessary information about drugs, diseases and patient due to the large amount of documents that are increasing every day. Consequently, so many medical errors can happen and even can cause fatalities. On the other hand, information extraction is such a good field that can handle this problem. One of the most important main task in this field is the Named Entity Recognition (NER) and its role was to identify the medical named entities, such as drug, disease or treatment, from an unstructured text written in natural language. However, in order to treat the narrative text, natural language tasks should be performed before NER. In our paper, we introduce a named entity recognition method, called NESSMa (Named Entity tagging by Surrounding Sequence Matching), and based on sequence tagging, it is able to annotate the words of a sentence using Bidirectional Long Short-Term Memory neural network with Conditional Random Field (BiLSTM-CRF) model. We pass the Bidirectional Encoder Representations from Transformers (BERT) word embedding as feature together with the Part of Speech (PoS) of the word and the cue sequence information. The cue sequence information indicates if a word belongs to a named entity surrounding sequence based on word edit distance. For that, we have automatically constructed a dictionary of named entities’ surrounding sequences for each entity type using a train set. As expected, experiments shows that adding the cue sequence information is able to improve the results according to F1-measure and outperform state-of-the-art methods.

Twitter is a communication medium and a collaboration system that allows broadcasting short messages called tweets. In contrast to traditional blogs, media-sharing and social networks services, microblogs (tweets) are textual messages submitted in real-time to report an idea, an actual interest, or an opinion. The size of these messages may be limited by a maximum number of characters. This constraint, related to the size of message, causes the use of a particular vocabulary. The aim is to exchange a maximum of information in as little characters as possible. In this respect, we will focus on the Tweet Contextualization task. The purpose of this task is to allow the reader a better understanding of the tweet. This paper deals with a new Tweet Contextualization approach using a semantic query expansion. The main idea of our proposed method is to enhance queries (tweets) in order to produce more informative contexts. The effectiveness of our method is proved through an experimental study conducted on the INEX 2014 collection.

The great success of social networks is due to their ability to offer Internet users a space of free expression where they can produce a large amount of information which provides every day with a new challenge for data analysts. The ease of use of social media encourages users to increasingly express their opinions either by using simple words expressing feelings or by using irony and sarcasm. The new challenges are to extract and analyze this mass of information which can then be used in different applications such as sentiment analysis and sarcasm detection. Sarcasm detection is a subarea of sentiment analysis, opinion mining, and emotion mining which are all representing the process of automatic identification of people’s orientation or sentiment toward individuals, products, services, issues, and events. Sarcasm detection, which is the fact of deciding if a text is ironic or not, could be used, for example, to improve the precision of the sentiment analysis. In most of the existing approaches, sarcasm detection is a binary classification; each text is classified as sarcastic or non-sarcastic; however, since tweets are generally written by humans and humans are by default fuzzy in their emotions and expressions, we can’t 100% confirm that a text is sarcastic or not. In addition, tweets are expressed in natural language which is full of ambiguity and non-precision, which motivates us more to adopt fuzzy logic, not just to detect sarcasm but to give it a score. In this manuscript, we propose a fuzzy sarcasm detection approach using social information such as replies, historical tweets and likes, etc multiplying each by a degree of importance. The evaluation shows that the use of fuzzy logic has led us to improve the precision metric of the classification and to improve the accuracy of our approach. Using degrees of importance gave us the best values for recall, precision, and accuracy measures compared to existing approaches.

Categorization of profiles in social networks represents one of the most interesting current challenges. Indeed, nowadays social networks have an inevitable role in the decision-making of Internet users. Detecting categories of profiles, which can play the role of influencers, can make a difference in understanding the structures and the functions of networks. With the exponential growth of data, and since we are talking about machine learning methods, classical techniques have become of poor efficiency. In this article, we introduce Weighted Categories Propagation (WCP), an improvement of existing methods which benefit from the labels propagation to consider the influence of social relationships on profile categorization. The efficiency of WCP is experimentally validated. Experiments carried out on the data set of RepLab 2014 indicate that WCP outperforms similar methods.

Retrieving relevant information from Twitter is always a challenging task given its vocabulary mismatch, sheer volume and noise. Representing the content of text tweets is a critical part of any microblog retrieval model. For this reason, deep neural networks can be used for learning good representations of text data and then conduct to a better matching. In this paper, we are interested in improving both representation and retrieval effectiveness in microblogs. For that, a Hybrid-Deep Neural-Network-based text representation model is proposed to extract effective features’ representations for clustering oriented microblog retrieval. HDNN combines recurrent neural network and feedforward neural network architectures. Specifically, using a bi-directional LSTM, we first generate a deep contextualized word representation which incorporates character n-grams form FasText. However, these contextual embedded existing in a high-dimensional space are not all important. Some of them are redundant, correlated and sometimes noisy making the learning models over-fitting, complex and less interpretable. To deal with these problems, we proposed a Hybrid-Regularized-Autoencoder-based method which combines autoencoder with Elastic Net regularization for an effective unsupervised feature selection and extraction. Our experimental results show that the performance of clustering and especially information retrieval in microblogs depend heavily on features’ representation.

Tweet contextualization (TC) is a new issue that aims to answer questions of the form What is this tweet about? The idea of this task was imagined as an extension of a previous area called multi-document summarization (MDS), which consists in generating a summary from many sources. In both TC and MDS, the summary should ideally contain most relevant information of the topic that is being discussed in the source texts (for MDS) and related to the query (for TC). Furthermore of being informative, a summary should be coherent, i.e. well written to be readable and grammatically compact. Hence, coherence is an essential characteristic in order to produce comprehensible texts. In this paper, we propose a new approach to improve readability and coherence for tweet contextualization based on bipartite graphs. The main idea of our proposed method is to reorder sentences in a given paragraph by combining most expressive words detection and HITS (Hyperlink- Induced Topic Search) algorithm to make up a coherent context.

Today, Twitter has become one of the most popular micro-blogging service with a large amount of information on various topics produced by millions of users every day. When searching for useful information in Twitter, users need to assess high quality content that meets their needs. However, the study of effective information retrieval in such microblog is still a challenge because there is a large difference in the quality level of relevant tweets returned in the search results for a given query. Therefore, looking for an effective microblog retrieval requires distinguishing the high-quality tweet content among thousands of results. Existing works are based on hand-crafted features (e.g., number of re-tweets, number of followers, etc.) using hard-hand-engineering to indicate the quality of tweets. In this paper, we focus on the problem of ranking tweets, and particularly on retrieving high quality content. We propose a ranking approach based on k-means clustering to distinguish high quality from low quality tweets. The clustering algorithm is based on learning features from deep learning autoencoder and hand-crafted features from tweets’ content and authors’ profiles. We used information gain as a feature importance measure to find the optimal set of features having stronger power in clustering the data. By conducting a pilot feature analysis study, we demonstrate the impact of the learned features to identify tweets’ quality in the clustering process. Our experimental results show that the integration of learned features has shown significant improvement in the quality of clustering and especially on the ranking performance compared to the use of hand-crafted features only.

The explosion of social media and microblogging services has gradually increased the microblogging data and particularly tweets data. In microblogging services such as Twitter, the users may become overwhelmed by the rise of data. Although, Twitter allows people to micro-blog about a broad range of topics in real time, it is often hard to understand what these tweets are about. In this work, we study the problem of Tweet Categorization (TC), which aims to automatically classify tweets based on their topic. The accurate TC, however, is a challenging task within the 140-character limit imposed by Twitter. The majority of TC approaches use lexical features such as Bag of Words (BoW) and Bag of Entities (BoE) extracted from a Tweet content. In this paper, we propose a semantic approach of improving the accuracy of TC based on feature expansion from external Knowledge Bases (KBs) and the use of eXtended WordNet Domain as a classifier. In particular, we propose a deep enrichment strategy to extend tweets with additional features by exploiting the concepts present in the semantic graph structures of the KBs. Then, our supervised categorization relies only on the ontological knowledge and classifier training is not required. Empirical results indicate that this enriched representation of text items can substantially improve the TC performance.

The main objective in XML Retrieval is to select the relevant elements of XML document instead of the whole document. Many open issues appear when considering Relevance Feedback (RF) in XML documents. They are mainly related to the form of XML documents, which mix content and structure information and to the new information granularity. In this paper, a new flexible method of relevance feedback in XML retrieval using two sources of evidence is described. We propose to use the context criterion to select terms to extend the initial query and to use generative structures to express structural constraints. Both approaches are applied in different combined forms. Experiments are carried out with the INEX evaluation campaign and results show the effectiveness of our approach.

Relevance Feedback (RF) is a technique allowing to enrich an initial query according to the user feedback. The goal is to express more precisily the user's needs. Some open issues appear when considering semi-structured documents like XML documents. Most of the RF approaches proposed in XML retrieval are simple adaptations of traditional RF to the new granularity of information. They enrich queries by adding terms extracted from relevant elements instead of terms extracted from whole documents. In this paper we show how structural constraints can also be used in RF. We propose a new approach that is able to extend the initial query by adding one or more generative structures. This approach is applied to unstructured queries. Experiments are carried out on INEX collection and results show the interest of our method.