Software companies spend a lot of resources on debugging, a process where previous solutions can help in solving current problems. The bug tickets, containing this information, are often time-consuming to read. To minimize the time spent on debugging and to make sure that the knowledge from prior solutions is kept in the company, an evaluation was made to see if summaries could make this process more efficient. Abstractive and extractive summarization models were tested for this task and fine-tuning of the bert-extractive-summarizer was performed. The model-generated summaries were compared in terms of perceived quality, speed, similarity to each other, and summarization length. The average description summary contained part of the description needed and the found solution was either well documented or did not answer the problem at all. The fine-tuned extractive model and the abstractive model BART provided good conditions for generating summaries containing all the information needed. / Vid mjukvaruutveckling går mycket resurser åt till felsökning, en process där tidigare lösningar kan hjälpa till att lösa aktuella problem. Det är ofta tidskrävande att läsa felrapporterna som innehåller denna information. För att minimera tiden som läggs på felsökning och säkerställa att kunskap från tidigare lösningar bevaras inom företaget, utvärderades om sammanfattningar skulle kunna effektivisera detta. Abstrakta och extraherande sammanfattningsmodeller testades för uppgiften och en finjustering av bert-extractive- summarizer gjordes. De genererade sammanfattningarna jämfördes i avseende på upplevd kvalitet, genereringshastighet, likhet mellan varandra och sammanfattningslängd. Den genomsnittliga sammanfattningen innehöll delar av den viktigaste informationen och den föreslagna lösningen var antingen väldokumenterad eller besvarade inte problembeskrivningen alls. Den finjusterade BERT och den abstrakta modellen BART visade goda förutsättningar för att generera sammanfattningar innehållande all den viktigaste informationen.

No description available.

No description available.

Background Healthcare professionals spend large amounts of time on documentation tasks in contemporary healthcare. One such documentation task is the discharge summary which summarizes a care episode. However, research shows that many discharge summaries written today are of lacking quality. One method which has the po- tential to alleviate the situation is natural language processing, specifically text summarization, as it could automatically summarize patient notes into a discharge summary. Aim This thesis aims to provide initial knowledge on the topic of summarization of Swedish clinical text into discharge summaries. Furthermore, this thesis aims to provide knowledge specifically on performing summarization using the Stockholm EPR Gastro ICD-10 Pseudo Corpus II dataset, consisting of Swedish electronic health record data. Method Using the design science framework, an artefact was produced in the form of a model, based on a pre-trained Swedish BART model, which can summarize patient notes into a discharge summary. This model was developed using the Hugging Face library and evaluated both via ROUGE scores as well as via a manual evaluation performed by a now retired healthcare professional. Results The discharge summaries produced from a test set by the artefact model achieved ROUGE-1/2/L/S scores of 0.280/0.057/0.122/0.068. The manual evaluation im- plies that the artefact is prone to fail to accurately include clinically important information, that the artefact produces text with low readability, and that the artefact is very prone to produce severe hallucinations. Conclusion The artefact’s performance is worse than the results of previous studies on the topic of summarization of patient notes into discharge summaries, in terms of ROUGE scores. The manual evaluation of the artefact performance suggests sev- eral shortcomings in its capabilities to accurately summarize a care episode. Since this was the first major work conducted on the topic of text summarization using the Stockholm EPR Gastro ICD-10 Pseudo Corpus II dataset, there are many possible directions for future works.

Text plays a pivotal role in our daily lives, encompassing various forms such as social media posts, news articles, books, reports, and more. Consequently, Natural Language Processing (NLP) has garnered widespread attention. This technology empowers us to undertake tasks like text classification, entity recognition, and even crafting responses within a dialogue context. However, despite the expansive utility of NLP, it frequently necessitates a critical decision: whether to place trust in a model's predictions. To illustrate, consider a state-of-the-art (SOTA) model entrusted with diagnosing a disease or assessing the veracity of a rumor. An incorrect prediction in such scenarios can have dire consequences, impacting individuals' health or tarnishing their reputation. Consequently, it becomes imperative to establish a reliable method for evaluating the reliability of an NLP model's predictions, which is our focus-uncertainty estimation on NLP. Though many works have researched uncertainty estimation or NLP, the combination of these two domains is rare. This is because most NLP research emphasizes model prediction performance but tends to overlook the reliability of NLP model predictions. Additionally, current uncertainty estimation models may not be suitable for NLP due to the unique characteristics of NLP tasks, such as the need for more fine-grained information in named entity recognition. Therefore, this dissertation proposes novel uncertainty estimation methods for different NLP tasks by considering the NLP task's distinct characteristics. The NLP tasks are categorized into natural language understanding (NLU) and natural language generation (NLG, such as text summarization). Among the NLU tasks, the understanding could be on two views, global-view (e.g. text classification at document level) and local-view (e.g. natural language inference at sentence level and named entity recognition at token level). As a result, we research uncertainty estimation on three tasks: text classification, named entity recognition, and text summarization. Besides, because few-shot text classification has captured much attention recently, we also research the uncertainty estimation on few-shot text classification. For the first topic, uncertainty estimation on text classification, few uncertainty models focus on improving the performance of text classification where human resources are involved. In response to this gap, our research focuses on enhancing the accuracy of uncertainty scores by bolstering the confidence associated with winning scores. we introduce MSD, a novel model comprising three distinct components: 'mix-up,' 'self-ensembling,' and 'distinctiveness score.' The primary objective of MSD is to refine the accuracy of uncertainty scores by mitigating the issue of overconfidence in winning scores while simultaneously considering various categories of uncertainty. seamlessly integrate with different Deep Neural Networks. Extensive experiments with ablation settings are conducted on four real-world datasets, resulting in consistently competitive improvements. Our second topic focuses on uncertainty estimation on few-shot text classification (UEFTC), which has few or even only one available support sample for each class. UEFTC represents an underexplored research domain where, due to limited data samples, a UEFTC model predicts an uncertainty score to assess the likelihood of classification errors. However, traditional uncertainty estimation models in text classification are ill-suited for UEFTC since they demand extensive training data, while UEFTC operates in a few-shot scenario, typically providing just a few support samples, or even just one, per class. To tackle this challenge, we introduce Contrastive Learning from Uncertainty Relations (CLUR) as a solution tailored for UEFTC. CLUR exhibits the unique capability to be effectively trained with only one support sample per class, aided by pseudo uncertainty scores. A distinguishing feature of CLUR is its autonomous learning of these pseudo uncertainty scores, in contrast to previous approaches that relied on manual specification. Our investigation of CLUR encompasses four model structures, allowing us to evaluate the performance of three commonly employed contrastive learning components in the context of UEFTC. Our findings highlight the effectiveness of two of these components. Our third topic focuses on uncertainty estimation on sequential labeling. Sequential labeling involves the task of assigning labels to individual tokens in a sequence, exemplified by Named Entity Recognition (NER). Despite significant advancements in enhancing NER performance in prior research, the realm of uncertainty estimation for NER (UE-NER) remains relatively uncharted but is of paramount importance. This topic focuses on UE-NER, seeking to gauge uncertainty scores for NER predictions. Previous models for uncertainty estimation often overlook two distinctive attributes of NER: the interrelation among entities (where the learning of one entity's embedding depends on others) and the challenges posed by incorrect span predictions in entity extraction. To address these issues, we introduce the Sequential Labeling Posterior Network (SLPN), designed to estimate uncertainty scores for the extracted entities while considering uncertainty propagation from other tokens. Additionally, we have devised an evaluation methodology tailored to the specific nuances of wrong-span cases. Our fourth topic focuses on an overlooked question that persists regarding the evaluation reliability of uncertainty estimation in text summarization (UE-TS). Text summarization, a key task in natural language generation (NLG), holds significant importance, particularly in domains where inaccuracies can have serious consequences, such as healthcare. UE-TS has garnered attention due to the potential risks associated with erroneous summaries. However, the reliability of evaluating UE-TS methods raises concerns, stemming from the interdependence between uncertainty model metrics and the wide array of NLG metrics. To address these concerns, we introduce a comprehensive UE-TS benchmark incorporating twenty-six NLG metrics across four dimensions. This benchmark evaluates the uncertainty estimation capabilities of two large language models and one pre-trained language model across two datasets. Additionally, it assesses the effectiveness of fourteen common uncertainty estimation methods. Our study underscores the necessity of utilizing diverse, uncorrelated NLG metrics and uncertainty estimation techniques for a robust evaluation of UE-TS methods. / Doctor of Philosophy / Text is integral to our daily activities, appearing in various forms such as social media posts, news articles, books, and reports. We rely on text for communication, information dissemination, and decision-making. Given its ubiquity, the ability to process and understand text through Natural Language Processing (NLP) has become increasingly important. NLP technology enables us to perform tasks like text classification, which involves categorizing text into predefined labels, and named entity recognition (NER), which identifies specific entities such as names, dates, and locations within text. Additionally, NLP facilitates generating coherent and contextually appropriate responses in conversational agents, enhancing human-computer interaction. However, the reliability of NLP models is crucial, especially in sensitive applications like medical diagnoses, where errors can have severe consequences. This dissertation focuses on uncertainty estimation in NLP, a less explored but essential area. Uncertainty estimation helps evaluate the confidence of NLP model predictions. We propose new methods tailored to various NLP tasks, acknowledging their unique needs. NLP tasks are divided into natural language understanding (NLU) and natural language generation (NLG). Within NLU, we look at tasks from two perspectives: a global view (e.g., document-level text classification) and a local view (e.g., sentence-level inference and token-level entity recognition). Our research spans text classification, named entity recognition (NER), and text summarization, with a special focus on few-shot text classification due to its recent prominence. For text classification, we introduce the MSD model, which includes three components to enhance uncertainty score accuracy and address overconfidence issues. This model integrates seamlessly with different neural networks and shows consistent improvements in experiments. For few-shot text classification, we develop Contrastive Learning from Uncertainty Relations (CLUR), designed to work effectively with minimal support samples per class. CLUR autonomously learns pseudo uncertainty scores, demonstrating effectiveness with various contrastive learning components. In NER, we address the unique challenges of entity interrelation and span prediction errors. We propose the Sequential Labeling Posterior Network (SLPN) to estimate uncertainty scores while considering uncertainty propagation from other tokens. For text summarization, we create a benchmark with tens of metrics to evaluate uncertainty estimation methods across two datasets. This benchmark helps assess the reliability of these methods, highlighting the need for diverse, uncorrelated metrics. Overall, our work advances the understanding and implementation of uncertainty estimation in NLP, providing more reliable and accurate predictions across different tasks.

隨著網路世界的興起，在面臨選擇難題時，民眾不僅會接收口耳相傳的資訊，也會以關鍵字上網搜尋目標資訊，但是在海量資料的浪潮中，如何快速的整合資料是一大挑戰。電影影評文章摘要可以幫助民眾進電影院前了解電影的資訊，透過這樣的方式確認電影是自身有興趣的電影。 本研究以電影：復仇者聯盟2影評66篇4616句、蝙蝠俠對超人：正義曙光60篇9345句、動物方城市60篇5545句、星際效應50篇4616句、高年級實習生62篇5622句為資料來源，以分群概念結合摘句之方法生成影評摘要。其中，利用K-Means演算法將五部電影的多篇影評特徵詞、句子進行分群後，使用TFIDF評比各分群語句的重要性來選取高權重語句，再以WWA方法挑選分群中不同面向的語句，最後以相似度計算最佳範本與各分群內容的相似度來決定每一群聚的排序順序，產生一篇具有相似內容段落和段落順序的影評多篇摘要。 研究結果顯示，原本五部電影影評對最佳範本之相似度為15.87%，經由本研究方法產生之摘要對最佳範本單篇摘要之相似度為21.19%。另外，因為影評中各分群的順序是比對最佳範本相似度而產生的排序，整篇摘要會具有與最佳範本相似段落排序的摘要內容，其中內容包含了電影影評中廣泛提到的相似內容，不同的相似段落讓文章摘要的呈現更具廣泛性。藉由此摘要方法，可以幫助民眾藉由自動化彙整、萃取的摘要快速了解相關電影資訊內容和協助決策。 / Abstract As Facing the Big Data issue, there are too many information on the website for reader to understand. How to perform and summarize essential information quickly is a challenge. People who want to go to a movie will also face this situation. Before choosing movies, they will search relative information of the movies. However, there are many film reviews all over the websites. Automatic text summarization can efficiently extract important information for readers, and conclude concepts of reviews on the websites. Through this method, readers can easily comprehend the best idea of all the reviews and save their time. The research presents a multi-concept and extractive film review summary for readers. It generates film review summary from the most popular blog platform, PIXNET, with extract-based method and clustering concept. The method using K-Means algorism let the film review summary focus on specific film to cluster the sentences by features, and having statistical sense and WWA method to measure the weight of sentences in order to choose the representative sentences. On the last step, it will compare to templates to decide the sequence of classified sentences and summary all represent sentences from each cluster. The research provides a multi-concept and extractive film review summary for people. From the result, there are five movies, which are used summary method increase the average similarity to 21.19% that comparing between the film reviews summary and templates summary. It shows that the automatic film reviews summarization can extract the important sentences from the reviews. Also, with comparing template method to order the cluster, it can sequentially list the cluster of the sentences to generate a movie review, which saves readers’ time and easily comprehend.

This work focuses on the summarization task for a set of articles on the same topic. It discusses several possible ways of summarizations and ways to assess their final quality. The implementation of the described algorithms and their application to selected texts constitutes a part of this work. The input texts come from several Czech news servers and they are represented as deep syntactic trees (the so called tectogrammatical layer).

Le contenu généré dans les médias sociaux comme Twitter permet aux utilisateurs d'avoir un aperçu rétrospectif d'évènement et de suivre les nouveaux développements dès qu'ils se produisent. Cependant, bien que Twitter soit une source d'information importante, il est caractérisé par le volume et la vélocité des informations publiées qui rendent difficile le suivi de l'évolution des évènements. Pour permettre de mieux tirer profit de ce nouveau vecteur d'information, deux tâches complémentaires de recherche d'information dans les médias sociaux ont été introduites : la génération de résumé rétrospectif qui vise à sélectionner les tweets pertinents et non redondant récapitulant "ce qui s'est passé" et l'envoi des notifications prospectives dès qu'une nouvelle information pertinente est détectée. Notre travail s'inscrit dans ce cadre. L'objectif de cette thèse est de faciliter le suivi d'événement, en fournissant des outils de génération de synthèse adaptés à ce vecteur d'information. Les défis majeurs sous-jacents à notre problématique découlent d'une part du volume, de la vélocité et de la variété des contenus publiés et, d'autre part, de la qualité des tweets qui peut varier d'une manière considérable. La tâche principale dans la notification prospective est l'identification en temps réel des tweets pertinents et non redondants. Le système peut choisir de retourner les nouveaux tweets dès leurs détections où bien de différer leur envoi afin de s'assurer de leur qualité. Dans ce contexte, nos contributions se situent à ces différents niveaux : Premièrement, nous introduisons Word Similarity Extended Boolean Model (WSEBM), un modèle d'estimation de la pertinence qui exploite la similarité entre les termes basée sur le word embedding et qui n'utilise pas les statistiques de flux. L'intuition sous- jacente à notre proposition est que la mesure de similarité à base de word embedding est capable de considérer des mots différents ayant la même sémantique ce qui permet de compenser le non-appariement des termes lors du calcul de la pertinence. Deuxièmement, l'estimation de nouveauté d'un tweet entrant est basée sur la comparaison de ses termes avec les termes des tweets déjà envoyés au lieu d'utiliser la comparaison tweet à tweet. Cette méthode offre un meilleur passage à l'échelle et permet de réduire le temps d'exécution. Troisièmement, pour contourner le problème du seuillage de pertinence, nous utilisons un classificateur binaire qui prédit la pertinence. L'approche proposée est basée sur l'apprentissage supervisé adaptatif dans laquelle les signes sociaux sont combinés avec les autres facteurs de pertinence dépendants de la requête. De plus, le retour des jugements de pertinence est exploité pour re-entrainer le modèle de classification. Enfin, nous montrons que l'approche proposée, qui envoie les notifications en temps réel, permet d'obtenir des performances prometteuses en termes de qualité (pertinence et nouveauté) avec une faible latence alors que les approches de l'état de l'art tendent à favoriser la qualité au détriment de la latence. Cette thèse explore également une nouvelle approche de génération du résumé rétrospectif qui suit un paradigme différent de la majorité des méthodes de l'état de l'art. Nous proposons de modéliser le processus de génération de synthèse sous forme d'un problème d'optimisation linéaire qui prend en compte la diversité temporelle des tweets. Les tweets sont filtrés et regroupés d'une manière incrémentale en deux partitions basées respectivement sur la similarité du contenu et le temps de publication. Nous formulons la génération du résumé comme étant un problème linéaire entier dans lequel les variables inconnues sont binaires, la fonction objective est à maximiser et les contraintes assurent qu'au maximum un tweet par cluster est sélectionné dans la limite de la longueur du résumé fixée préalablement. / User-generated content on social media, such as Twitter, provides in many cases, the latest news before traditional media, which allows having a retrospective summary of events and being updated in a timely fashion whenever a new development occurs. However, social media, while being a valuable source of information, can be also overwhelming given the volume and the velocity of published information. To shield users from being overwhelmed by irrelevant and redundant posts, retrospective summarization and prospective notification (real-time summarization) were introduced as two complementary tasks of information seeking on document streams. The former aims to select a list of relevant and non-redundant tweets that capture "what happened". In the latter, systems monitor the live posts stream and push relevant and novel notifications as soon as possible. Our work falls within these frameworks and focuses on developing a tweet summarization approaches for the two aforementioned scenarios. It aims at providing summaries that capture the key aspects of the event of interest to help users to efficiently acquire information and follow the development of long ongoing events from social media. Nevertheless, tweet summarization task faces many challenges that stem from, on one hand, the high volume, the velocity and the variety of the published information and, on the other hand, the quality of tweets, which can vary significantly. In the prospective notification, the core task is the relevancy and the novelty detection in real-time. For timeliness, a system may choose to push new updates in real-time or may choose to trade timeliness for higher notification quality. Our contributions address these levels: First, we introduce Word Similarity Extended Boolean Model (WSEBM), a relevance model that does not rely on stream statistics and takes advantage of word embedding model. We used word similarity instead of the traditional weighting techniques. By doing this, we overcome the shortness and word mismatch issues in tweets. The intuition behind our proposition is that context-aware similarity measure in word2vec is able to consider different words with the same semantic meaning and hence allows offsetting the word mismatch issue when calculating the similarity between a tweet and a topic. Second, we propose to compute the novelty score of the incoming tweet regarding all words of tweets already pushed to the user instead of using the pairwise comparison. The proposed novelty detection method scales better and reduces the execution time, which fits real-time tweet filtering. Third, we propose an adaptive Learning to Filter approach that leverages social signals as well as query-dependent features. To overcome the issue of relevance threshold setting, we use a binary classifier that predicts the relevance of the incoming tweet. In addition, we show the gain that can be achieved by taking advantage of ongoing relevance feedback. Finally, we adopt a real-time push strategy and we show that the proposed approach achieves a promising performance in terms of quality (relevance and novelty) with low cost of latency whereas the state-of-the-art approaches tend to trade latency for higher quality. This thesis also explores a novel approach to generate a retrospective summary that follows a different paradigm than the majority of state-of-the-art methods. We consider the summary generation as an optimization problem that takes into account the topical and the temporal diversity. Tweets are filtered and are incrementally clustered in two cluster types, namely topical clusters based on content similarity and temporal clusters that depends on publication time. Summary generation is formulated as integer linear problem in which unknowns variables are binaries, the objective function is to be maximized and constraints ensure that at most one post per cluster is selected with respect to the defined summary length limit.

Due to the increasing amount and complexity of the data stored in the enterprises\' databases, the task of knowledge discovery is nowadays vital to support strategic decisions. However, the mining techniques used in the process usually have high computational costs that come from the need to explore several alternative solutions, in different combinations, to obtain the desired knowledge. The most common mining tasks include data classification, labeling and clustering, outlier detection and missing data prediction. Traditionally, the data are represented by numerical or categorical attributes in a table that describes one element in each tuple. Although the same tasks applied to traditional data are also necessary for more complex data, such as images, graphs, audio and long texts, the complexity and the computational costs associated to handling large amounts of these complex data increase considerably, making most of the existing techniques impractical. Therefore, especial data mining techniques for this kind of data need to be developed. This Ph.D. work focuses on the development of new data mining techniques for large sets of complex data, especially for the task of clustering, tightly associated to other data mining tasks that are performed together. Specifically, this Doctoral dissertation presents three novel, fast and scalable data mining algorithms well-suited to analyze large sets of complex data: the method Halite for correlation clustering; the method BoW for clustering Terabyte-scale datasets; and the method QMAS for labeling and summarization. Our algorithms were evaluated on real, very large datasets with up to billions of complex elements, and they always presented highly accurate results, being at least one order of magnitude faster than the fastest related works in almost all cases. The real data used come from the following applications: automatic breast cancer diagnosis, satellite imagery analysis, and graph mining on a large web graph crawled by Yahoo! and also on the graph with all users and their connections from the Twitter social network. Such results indicate that our algorithms allow the development of real time applications that, potentially, could not be developed without this Ph.D. work, like a software to aid on the fly the diagnosis process in a worldwide Healthcare Information System, or a system to look for deforestation within the Amazon Rainforest in real time / O crescimento em quantidade e complexidade dos dados armazenados nas organizações torna a extração de conhecimento utilizando técnicas de mineração uma tarefa ao mesmo tempo fundamental para aproveitar bem esses dados na tomada de decisões estratégicas e de alto custo computacional. O custo vem da necessidade de se explorar uma grande quantidade de casos de estudo, em diferentes combinações, para se obter o conhecimento desejado. Tradicionalmente, os dados a explorar são representados como atributos numéricos ou categóricos em uma tabela, que descreve em cada tupla um caso de teste do conjunto sob análise. Embora as mesmas tarefas desenvolvidas para dados tradicionais sejam também necessárias para dados mais complexos, como imagens, grafos, áudio e textos longos, a complexidade das análises e o custo computacional envolvidos aumentam significativamente, inviabilizando a maioria das técnicas de análise atuais quando aplicadas a grandes quantidades desses dados complexos. Assim, técnicas de mineração especiais devem ser desenvolvidas. Este Trabalho de Doutorado visa a criação de novas técnicas de mineração para grandes bases de dados complexos. Especificamente, foram desenvolvidas duas novas técnicas de agrupamento e uma nova técnica de rotulação e sumarização que são rápidas, escaláveis e bem adequadas à análise de grandes bases de dados complexos. As técnicas propostas foram avaliadas para a análise de bases de dados reais, em escala de Terabytes de dados, contendo até bilhões de objetos complexos, e elas sempre apresentaram resultados de alta qualidade, sendo em quase todos os casos pelo menos uma ordem de magnitude mais rápidas do que os trabalhos relacionados mais eficientes. Os dados reais utilizados vêm das seguintes aplicações: diagnóstico automático de câncer de mama, análise de imagens de satélites, e mineração de grafos aplicada a um grande grafo da web coletado pelo Yahoo! e também a um grafo com todos os usuários da rede social Twitter e suas conexões. Tais resultados indicam que nossos algoritmos permitem a criação de aplicações em tempo real que, potencialmente, não poderiam ser desenvolvidas sem a existência deste Trabalho de Doutorado, como por exemplo, um sistema em escala global para o auxílio ao diagnóstico médico em tempo real, ou um sistema para a busca por áreas de desmatamento na Floresta Amazônica em tempo real

A sumarização multidocumento consiste na tarefa de produzir automaticamente um único sumário a partir de um conjunto de textos derivados de um mesmo assunto. É imprescindível que seja feito o tratamento de fenômenos que ocorrem neste cenário, tais como: (i) a redundância, a complementaridade e a contradição de informações; (ii) a uniformização de estilos de escrita; (iii) tratamento de expressões referenciais; (iv) a manutenção de focos e perspectivas diferentes nos textos; (v) e a ordenação temporal das informações no sumário. O tratamento de tais fenômenos contribui significativamente para que seja produzido ao final um sumário informativo e coerente, características difíceis de serem garantidas ainda que por um humano. Um tipo particular de coerência estudado nesta tese é a coerência local, a qual é definida por meio de relações entre enunciados (unidades menores) em uma sequência de sentenças, de modo a garantir que os relacionamentos contribuirão para a construção do sentido do texto em sua totalidade. Partindo do pressuposto de que o uso de conhecimento discursivo pode melhorar a avaliação da coerência local, o presente trabalho propõe-se a investigar o uso de relações discursivas para elaborar modelos de coerência local, os quais são capazes de distinguir automaticamente sumários coerentes dos incoerentes. Além disso, um estudo sobre os erros que afetam a Qualidade Linguística dos sumários foi realizado com o propósito de verificar quais são os erros que afetam a coerência local dos sumários, se os modelos de coerência podem identificar tais erros e se há alguma relação entre os modelos de coerência e a informatividade dos sumários. Para a realização desta pesquisa foi necessário fazer o uso das informações semântico-discursivas dos modelos CST (Cross-document Structure Theory) e RST (Rhetorical Structure Theory) anotadas no córpus, de ferramentas automáticas, como o parser Palavras e de algoritmos que extraíram informações do córpus. Os resultados mostraram que o uso de informações semântico-discursivas foi bem sucedido na distinção dos sumários coerentes dos incoerentes e que os modelos de coerência implementados nesta tese podem ser usados na identificação de erros da qualidade linguística que afetam a coerência local. / Multi-document summarization is the task of automatically producing a single summary from a collection of texts derived from the same subject. It is essential to treat many phenomena, such as: (i) redundancy, complementarity and contradiction of information; (ii) writing styles standardization; (iii) treatment of referential expressions; (iv) text focus and different perspectives; (v) and temporal ordering of information in the summary. The treatment of these phenomena contributes to the informativeness and coherence of the final summary. A particular type of coherence studied in this thesis is the local coherence, which is defined by the relationship between statements (smallest units) in a sequence of sentences. The local coherence contributes to the construction of textual meaning in its totality. Assuming that the use of discursive knowledge can improve the evaluation of the local coherence, this thesis proposes to investigate the use of discursive relations to develop local coherence models, which are able to automatically distinguish coherent summaries from incoherent ones. In addition, a study on the errors that affect the Linguistic Quality of the summaries was conducted in order to verify what are the errors that affect the local coherence of summaries, as well as if the coherence models can identify such errors, and whether there is any relationship between coherence models and informativenessof summaries. For thisresearch, it wasnecessary theuseof semantic-discursive information of CST models (Cross-document Structure Theory) and RST (Rhetorical Structure Theory) annoted in the corpora, automatic tools, parser as Palavras, and algorithms that extract information from the corpus. The results showed that the use of semantic-discursive information was successful on the distinction between coherent and incoherent summaries, and that the information about coherence can be used in error detection of linguistic quality that affect the local coherence.