Biomedical literature contains critical health information but remains largely inaccessible to non-expert readers due to its technical vocabulary and complex discourse structure. Although plain language summaries (PLS) partially address this gap, their manual creation limits scalability. This paper introduced a new large-scale pseudo-parallel corpus for biomedical text simplification constructed from the Cochrane Database of Systematic Reviews by aligning scientific abstracts with their corresponding PLS our data composed by 6270 pairs of abstract-PLS. We describe a fully automated data collection and preprocessing pipeline and evaluate the effectiveness of the resulting corpus by fine-tuning two transformer-based sequence-to-sequence models, BARTbase and T5-base, using Low-Rank Adaptation (LoRA). Experimental results show that both models successfully reduce textual complexity while preserving semantic content, with T5-base achieving slightly better readability and semantic fidelity. This work contributes a reusable dataset and empirical evidence supporting transformer-based approaches for scalable biomedical text simplification.

Ensuring the quality of system summaries through automatic evaluation is crucial. However, current evaluation metrics for the Arabic language primarily focus on content assessment, ensuring that the summary includes important information, while often neglecting the equally important aspect of linguistic quality, which affects the readability of the summary. To address this issue, we propose a refined method that comprehensively evaluates both the content and linguistic quality of Arabic text summaries. This approach incorporates deep linguistic features alongside semantic features, rather than relying solely on lexical features. Our method aims to predict the “Overall Responsiveness” score by leveraging a combination of content and linguistic quality features. These features are predominantly based on the representation of words, phrases, and sentences as embedding vectors, utilizing either the doc2vec neural language model or multiple large language models inspired by BERT. To evaluate linguistic quality, we incorporate features related to grammaticality, non-redundancy, coherence, and structure. In our experimental setup, we meticulously selected the most relevant features for predicting the “Overall Responsiveness” score. Subsequently, we develop models using various machine learning algorithms, selecting the model with the highest correlation with the human score. Ultimately, we compare the performance of our selected model against multiple existing metrics. The results of our comparison demonstrate that our proposed metric surpasses existing ones, signifying its effectiveness in evaluating Arabic text summaries.

Lexical simplification (LS) is the process of replacing complex words in a sentence with simpler alternatives to improve readability and understandability. However, LS is a challenging task because the target population may have varying literacy levels, resulting in different needs. This study aims to enhance the SimpLex architecture, a LS framework proposed in the literature. Several techniques have been explored to achieve this goal. First, we added new features and trained two additional machine learning models: LightGBM and XGBoost. Additionally, multiple transformer models were examined, including DistilBERT, RoBERTa Large, Facebook BART, and Microsoft DeBERTa. The results demonstrated the effectiveness of LightGBM and XGBoost in predicting complex words. Furthermore, DistilBERT and RoBERTa Large achieved higher SARI scores and lower perplexity compared to the other pre-trained models, except BERT, which had a slightly lower perplexity than the aforementioned models.

Evaluating the quality and effectiveness of text summarization techniques is essential to ensure accurate and informative summaries. In this paper we present a novel method for evaluating Arabic text summarization based on measuring the similarity between embedding keyphrases. We utilize a pretrained BERT model to represent each keyphrase as a dense vector. Then, we quantify the similarity between keyphrases in the system and the human summary. We tested several lengths of keyphrases. After, we built a predictive model by combining different keyphrases based features with diverse linguistic quality features. To evaluate the effectiveness of our method, we compare our embedding keyphrase similarity method to traditional metrics such as ROUGE. Our experiments demonstrate that embedding-based evaluation provides a more nuanced assessment of the summarization quality, considering the semantic similarities between keyphrases. With the incorporation of semantic similarities, our approach captures the contextual relevance of keyphrases and enhances the evaluation process.

The evaluation of text summaries remains a challenging task despite the large number of studies in this field for more than two decades. This paper describes an automatic method for assessing Arabic text summaries. In fact, the proposed method will predict the "Overall Responsiveness" manual score, which is a combination of the content and the linguistic quality of a summary. To predict this manual score, we aggregate, with a regression function, three types of features: lexical similarity features, semantic similarity features and linguistic features. Semantic features include multiple semantic similarity scores based on Bert model. While linguistic features are based on the calculation of entropy scores. To calculate the similarity between a candidate summary and a reference summary, we begin by doing an exact match between n-grams. For the unmatched n-grams, we present them as Bert vectors, and then we compute the similarity between Bert vectors. The proposed method yielded competitive results compared to metrics based on lexical similarity such as ROUGE.

Text summary evaluation represents an important step after building any summarization system. Despite the important number of metrics that have been developed, there are a few metrics that evaluate Arabic text summary. In this paper, we present a new automatic metric for Arabic text summary evaluation. This metric combines ROUGE scores with documents embedding-based scores to build a regression model that predicts the manual score of an Arabic summary. First, we have constructed document embedding models with different vector sizes then, we have used these models to present each candidate and model summary as a document embedding vector. After that, a similarity score between the two document embedding vectors was calculated. Finally, we have combined several similarity scores based on the document embedding representation and the ROUGE scores to predict a manual score. Furthermore, in the combination phase, we have tried multiple regression models to obtain the most optimal predictive model. The obtained result showed that the proposed method outperforms all baseline metrics in the task of text summary evaluation and the task of the summarization system evaluation.

In this paper, we propose to compare four methods for speech understanding of under-resourced languages. The first method is knowledge based method which integrates ontology. The second one is statistical method which integrates CRF discriminative models. The third one is hybrid method which uses CRF models with integration of knowledge base. The fourth one is pattern based method. We have used a spoken Tunisian dialect corpus acquired and annotated to perform experiments. The evaluation is based on semantic representation generated by each method. The obtained results shows that the hybrid method is the best one compared to others, which proves that CRF models with ontology integration is suitable for under-resourced languages.

The present paper introduces a new MultiLing text summary evaluation method. This method relies on machine learning approach which operates by combining multiple features to build models that predict the human score (overall responsiveness) of a new summary. We have tried several single and “ensemble learning” classifiers to build the best model. We have experimented our method in summary level evaluation where we evaluate each text summary separately. The correlation between built models and human score is better than the correlation between baselines and manual score.

In the context of the Personalized Information Retrieval method applied to the Arabic language, this work consists in presenting a personalized ranking method based on a model of supervised learning and its implementation.This method consists of four steps, namely, the user's modeling, the document / query / profile matching, the learning to rank and the result classification.Thus, we proposed a hybrid approach of the user's modeling that relies on both multidimensional and conceptual representations by exploiting Arabic semantic resources.Therefore, to determine the similarity between the document and the profile, we used a learning model that exploits the users' explicit pertinence judgments.In this context, we have proposed learning semantic features related to the user's profile (represented by hierarchies of concepts).The predicted model will then be used in the ordering phase to classify other documents that result from a new query submitted by the user.In this context, we have proposed a novel multi-objective function to order the documents (based on the classic Retrieval Status Value function and the predictive personalized Retrieval Status Value function).Finally, we have explained the evaluation results of the predictive model and the ranking method.These evaluations, which were made based on a training corpus and a test corpus, led to some interesting results.Indeed, the proposed semantic learning criteria connected to the user profile have a significant impact on the performance of our personalized document ranking system.

In this paper, we propose a method that evaluates the content of a text summary using a machine learning approach. This method operates by combining multiple features to build models that predict the PYRAMID scores for new summaries. We have tested several single and "Ensemble Learning" classifiers to build the best model. The evaluation of summarization system is made using the average of the scores of summaries that are built from each system. The results show that our method has achieved good performance in predicting the content score for a summary as well as for a summarization system.

This paper presents an automatic method tier assessing Arabic text summaries. This method is based on a machine learning approach which operates by building a model that combines multiple features to predict the human score "Overall Responsiveness" of an unseen summary. We have tested multiple single and ensemble learning classifiers to build the best predictive model. We experimented our method in summary level evaluation where we evaluate the quality of each text summary separately and in a system level evaluation where the average quality of the text summary system was calculated. We report in this paper the results of our experiments for both evaluation levels; our results prove that the proposed method outperforms the baselines on both tasks.

In this article, we propose a method of text summary's content and linguistic quality evaluation that is based on a machine learning approach. This method operates by combining multiple features to build predictive models that evaluate the content and the linguistic quality of new summaries (unseen) constructed from the same source documents as the summaries used in the training and the validation of models. To obtain the best model, many single and ensemble learning classifiers are tested. Using the constructed models, we have achieved a good performance in predicting the content and the linguistic quality scores. In order to evaluate the summarization systems, we calculated the system score as the average of the score of summaries that are built from the same system. Then, we evaluated the correlation of the system score with the manual system score. The obtained correlation indicates that the system score outperforms the baseline scores.

The Evaluation of a summary's linguistic quality is a difficult task because several linguistic aspects (e.g. grammaticality, coherence, etc.) must be verified to ensure the well formedness of a text's summary. In this paper, we report the result of combining "Adapted ROUGE" scores and linguistic quality features to assess linguistic quality. We build and evaluate models for predicting the manual linguistic quality score using linear regression. We construct models for evaluating the quality of each text summary (summary level evaluation) and of each summarizing system (system level evaluation). We assess the performance of a summarizing system using the quality of a set of summaries generated by the system. All models are evaluated using the Pearson correlation and the Root mean squared error.

Given a set of candidate authors for whom some texts of undisputed authorship exist, attribute texts of unknown authorship to one of the candidates is called Author verification. This problem acquired great attention due to its new applications in forensic analysis, e-commerce and plagiarism detection. The author verification task is of great help in the plagiarism detection process. Indeed, the probability of plagiarism increases where two parts of a document are not assigned to the same author. This paper introduces an analysis framework for hybrid authorship verification. In fact, the proposed method takes advantage of a large set of linguistic features to fully address the identification of the document’s author. These features are explored to build a machine-learning process. We obtained promising results by relying on PAN@CLEF 2014 English literature corpus.

. In this study, we present a new method for profiling the author of an anonymous English text. The aim of author profiling is to determine demographic (age, gender, region, education level) and psychological (personality, mental health) properties of the authors of a text, especially authors of user generated content in social media. To obtain the best classification, authors resort to machine learning methods. Focusing on the works which use the Bayesian networks, all those methods rather apply the Bayesian naïve classifiers which do not yield the best results. Therfore we propose a method based on advanced Bayesian networks for age prediction to over come the mentioned detail problem. We obtained promising results by relying on an English PAN@CLEF 2013 corpus. The obtained results are comparable to the ones obtained by the best state of the art methods. The software and data can be publicly downloaded from www.cicling.org/2016/ data/248/CICLING_248.zip.

The work presented in this paper aims at developing a Personalized Information Retrieval system in Arabic Texts (“PIRAT”) based on the user’s preferences/interests. For this reason, we proposed a user’s modeling and a personalized matching method document-query. The proposed user’s modeling is based on a hybrid representation of the user profile. In this approach, we introduce an algorithm which automatically builds a hierarchical user profile that represents his implicit personal interests and domain. It is to represent the interests and the domain with a conceptual network of nodes linked together through relationships respecting the linking topology defined in the domain of hierarchies and ontologies (hyperonymy, hyponymy, and synonymy). Then, we address the problem of unavailable language resources by building (i) a large Arabic text corpus entitled “WCAT” and (ii) Building our own Arabic queries corpus entitled “AQC2” in order to evaluate the suggested PIRAT system and AXON system. The results of this evaluation are promising.

In this paper, we deal with the author identification issues of the document whose origin is unknown. To overcome these problems, we propose a new hybrid approach combining the statistical and stylistic analysis. Our introduced method is based on determining the lexical and syntactic features of the written text in order to identify the author of the document. These features are explored to build a machine learning process. We obtained promising results by relying on PAN@CLEF2014 English literature corpus. The experimental results are comparable to those obtained by the best state of the art methods.

This paper tackles the author identification problems of documents that have unknown origins. To cope with the author identification challenges, we introduce a hybrid method that amalgamates stylistic analysis with statistical analysis. In fact, the proposed method takes advantage of a large set of stylistic and statistical features to fully address the identification of the document's author. These features are explored to build a machine learning process. We obtained promising results by relying on an English literature corpus.

In this paper we focus on the identification of the author of a written text. We present a new hybrid method that combines a set of stylistic and statistical features in a machine learning process. We tested the effectiveness of the linguistic and statistical features combined with the inter-textual distance "Delta" on the PAN’@CLEF’2015 English corpus and we obtained 0.59 as c@1 precision.

In this paper, we propose a hybrid method for the spoken Tunisian dialect understanding within a limited task. This method couples a discriminative statistical method with a domain ontology. The statistical method is based on conditional random field (CRF) models learned from a little size corpus to perform conceptual labeling task. These models are able to detect the semantic dependency between words. However, the domain ontology is used to add prior knowledge about the task. Our experiments are based on a real spoken Tunisian dialect corpus. The obtained results show that the proposed method is able to improve the performance of CRF models for speech understanding by the integration of the domain ontology. Our method can be exploited for under-resourced languages and Arabic dialects to overcome the lack of linguistic resources .