Abstractive Long Document Summarization: Studio e Sperimentazione di Modelli Generativi Retrieval-Augmented

In questa tesi si trattano lo studio e la sperimentazione di un modello generativo retrieval-augmented, basato su Transformers, per il task di Abstractive Summarization su lunghe sentenze legali. La sintesi automatica del testo (Automatic Text Summarization) è diventata un task di Natural Language Processing (NLP) molto importante oggigiorno, visto il grandissimo numero di dati provenienti dal web e banche dati. Inoltre, essa permette di automatizzare un processo molto oneroso per gli esperti, specialmente nel settore legale, in cui i documenti sono lunghi e complicati, per cui difficili e dispendiosi da riassumere. I modelli allo stato dell’arte dell’Automatic Text Summarization sono basati su soluzioni di Deep Learning, in particolare sui Transformers, che rappresentano l’architettura più consolidata per task di NLP. Il modello proposto in questa tesi rappresenta una soluzione per la Long Document Summarization, ossia per generare riassunti di lunghe sequenze testuali. In particolare, l’architettura si basa sul modello RAG (Retrieval-Augmented Generation), recentemente introdotto dal team di ricerca Facebook AI per il task di Question Answering. L’obiettivo consiste nel modificare l’architettura RAG al fine di renderla adatta al task di Abstractive Long Document Summarization. In dettaglio, si vuole sfruttare e testare la memoria non parametrica del modello, con lo scopo di arricchire la rappresentazione del testo di input da riassumere. A tal fine, sono state sperimentate diverse configurazioni del modello su diverse tipologie di esperimenti e sono stati valutati i riassunti generati con diverse metriche automatiche.

COMPENDIUM: A text summarization system for generating abstracts of research papers

This article analyzes the appropriateness of a text summarization system, COMPENDIUM, for generating abstracts of biomedical papers. Two approaches are suggested: an extractive (COMPENDIUM E), which only selects and extracts the most relevant sentences of the documents, and an abstractive-oriented one (COMPENDIUM E–A), thus facing also the challenge of abstractive summarization. This novel strategy combines extractive information, with some pieces of information of the article that have been previously compressed or fused. Specifically, in this article, we want to study: i) whether COMPENDIUM produces good summaries in the biomedical domain; ii) which summarization approach is more suitable; and iii) the opinion of real users towards automatic summaries. Therefore, two types of evaluation were performed: quantitative and qualitative, for evaluating both the information contained in the summaries, as well as the user satisfaction. Results show that extractive and abstractive-oriented summaries perform similarly as far as the information they contain, so both approaches are able to keep the relevant information of the source documents, but the latter is more appropriate from a human perspective, when a user satisfaction assessment is carried out. This also confirms the suitability of our suggested approach for generating summaries following an abstractive-oriented paradigm.

A novel concept-level approach for ultra-concise opinion summarization

The Web 2.0 has resulted in a shift as to how users consume and interact with the information, and has introduced a wide range of new textual genres, such as reviews or microblogs, through which users communicate, exchange, and share opinions. The exploitation of all this user-generated content is of great value both for users and companies, in order to assist them in their decision-making processes. Given this context, the analysis and development of automatic methods that can help manage online information in a quicker manner are needed. Therefore, this article proposes and evaluates a novel concept-level approach for ultra-concise opinion abstractive summarization. Our approach is characterized by the integration of syntactic sentence simplification, sentence regeneration and internal concept representation into the summarization process, thus being able to generate abstractive summaries, which is one the most challenging issues for this task. In order to be able to analyze different settings for our approach, the use of the sentence regeneration module was made optional, leading to two different versions of the system (one with sentence regeneration and one without). For testing them, a corpus of 400 English texts, gathered from reviews and tweets belonging to two different domains, was used. Although both versions were shown to be reliable methods for generating this type of summaries, the results obtained indicate that the version without sentence regeneration yielded to better results, improving the results of a number of state-of-the-art systems by 9%, whereas the version with sentence regeneration proved to be more robust to noisy data.

Génération de résumés par abstraction

Cette thèse présente le résultat de plusieurs années de recherche dans le domaine de la génération automatique de résumés. Trois contributions majeures, présentées sous la forme d'articles publiés ou soumis pour publication, en forment le coeur. Elles retracent un cheminement qui part des méthodes par extraction en résumé jusqu'aux méthodes par abstraction. L'expérience HexTac, sujet du premier article, a d'abord été menée pour évaluer le niveau de performance des êtres humains dans la rédaction de résumés par extraction de phrases. Les résultats montrent un écart important entre la performance humaine sous la contrainte d'extraire des phrases du texte source par rapport à la rédaction de résumés sans contrainte. Cette limite à la rédaction de résumés par extraction de phrases, observée empiriquement, démontre l'intérêt de développer d'autres approches automatiques pour le résumé. Nous avons ensuite développé un premier système selon l'approche Fully Abstractive Summarization, qui se situe dans la catégorie des approches semi-extractives, comme la compression de phrases et la fusion de phrases. Le développement et l'évaluation du système, décrits dans le second article, ont permis de constater le grand défi de générer un résumé facile à lire sans faire de l'extraction de phrases. Dans cette approche, le niveau de compréhension du contenu du texte source demeure insuffisant pour guider le processus de sélection du contenu pour le résumé, comme dans les approches par extraction de phrases. Enfin, l'approche par abstraction basée sur des connaissances nommée K-BABS est proposée dans un troisième article. Un repérage des éléments d'information pertinents est effectué, menant directement à la génération de phrases pour le résumé. Cette approche a été implémentée dans le système ABSUM, qui produit des résumés très courts mais riches en contenu. Ils ont été évalués selon les standards d'aujourd'hui et cette évaluation montre que des résumés hybrides formés à la fois de la sortie d'ABSUM et de phrases extraites ont un contenu informatif significativement plus élevé qu'un système provenant de l'état de l'art en extraction de phrases.

prism: rinfrangere distribuzioni di probabilità in riassunti astrattivi mediante strategie di decoding

Negli ultimi quattro anni la summarization astrattiva è stata protagonista di una evoluzione senza precedenti dettata da nuovi language model neurali, architetture transformer-based, elevati spazi dimensionali, ampi dataset e innovativi task di pre-training. In questo contesto, le strategie di decoding convertono le distribuzioni di probabilità predette da un modello in un testo artificiale, il quale viene composto in modo auto regressivo. Nonostante il loro cruciale impatto sulla qualità dei riassunti inferiti, il ruolo delle strategie di decoding è frequentemente trascurato e sottovalutato. Di fronte all'elevato numero di tecniche e iperparametri, i ricercatori necessitano di operare scelte consapevoli per ottenere risultati più affini agli obiettivi di generazione. Questa tesi propone il primo studio altamente comprensivo sull'efficacia ed efficienza delle strategie di decoding in task di short, long e multi-document abstractive summarization. Diversamente dalle pubblicazioni disponibili in letteratura, la valutazione quantitativa comprende 5 metriche automatiche, analisi temporali e carbon footprint. I risultati ottenuti dimostrano come non vi sia una strategia di decoding dominante, ma come ciascuna possieda delle caratteristiche adatte a task e dataset specifici. I contributi proposti hanno l'obiettivo di neutralizzare il gap di conoscenza attuale e stimolare lo sviluppo di nuove tecniche di decoding.

SAB(IO): A BIOLOGICALLY PLAUSIBLE CONNECTIONIST APPROACH TO AUTOMATIC TEXT SUMMARIZATION

An implementation of a computational tool to generate new summaries from new source texts is presented, by means of the connectionist approach (artificial neural networks). Among other contributions that this work intends to bring to natural language processing research, the use of a more biologically plausible connectionist architecture and training for automatic summarization is emphasized. The choice relies on the expectation that it may bring an increase in computational efficiency when compared to the sa-called biologically implausible algorithms.

Summarization of specialized discourse: the case of medical articles in Spanish

In this article, we present the current state of our work on a linguistically-motivated model for automatic summarization of medical articles in Spanish. The model takes into account the results of an empirical study which reveals that, on the one hand, domain-specific summarization criteria can often be derived from the summaries of domain specialists, and, on the other hand, adequate summarization strategies must be multidimensional, i.e., cover various types of linguistic clues. We take into account the textual, lexical, discursive, syntactic and communicative dimensions. This is novel in the field of summarization. The experiments carried out so far indicate that our model is suitable to provide high quality summarizations.

Exploring the use of prediction and summarization to increase students' listening and reading comprehension /

This study examined the effect of expHcitly instructing students to use a repertoire of reading comprehension strategies. Specifically, this study examined whether providing students with a "predictive story-frame" which combined the use of prediction and summarization strategies improved their reading comprehension relative to providing students with generic instruction on prediction and summarization. Results were examined in terms of instructional condition and reading ability. Students from 2 grade 4 classes participated in this study. The reading component of the Canadian Achievement Tests, Second Edition (CAT/2) was used to identify students as either "average or above average" or "below average" readers. Students received either strategic predication and summarization instruction (story-frame) or generic prediction and summarization instruction (notepad). Students were provided with new but comparable stories for each session. For both groups, the researcher modelled the strategic tools and provided guided practice, independent practice, and independent reading sessions. Comprehension was measured with an immediate and 1-week delayed comprehension test for each of the 4 stories, hi addition, students participated in a 1- week delayed interview, where they were asked to retell the story and to answer questions about the central elements (character, setting, problem, solution, beginning, middle, and ending events) of each story. There were significant differences, with medium to large effect sizes, in comprehension and recall scores as a fimction of both instructional condition and reading ability. Students in the story-frame condition outperformed students in the notepad condition, and average to above average readers performed better than below average readers. Students in the story-frame condition outperformed students in the notepad condition on the comprehension tests and on the oral retellings when teacher modelling and guidance were present. In the cued recall sessions, students in the story-frame instructional condition recalled more correct information and generated fewer errors than students in the notepad condition. Average to above average readers performed better than below average readers across comprehension and retelling measures. The majority of students in both instructional conditions reported that they would use their strategic tool again.

Text Summarization by Sentence Extraction and Syntactic Pruning

Nous présentons une méthode hybride pour le résumé de texte, en combinant l'extraction de phrases et l'élagage syntaxique des phrases extraites. L'élagage syntaxique est effectué sur la base d’une analyse complète des phrases selon un parseur de dépendances, analyse réalisée par la grammaire développée au sein d'un logiciel commercial de correction grammaticale, le Correcteur 101. Des sous-arbres de l'analyse syntaxique sont supprimés quand ils sont identifiés par les relations ciblées. L'analyse est réalisée sur un corpus de divers textes. Le taux de réduction des phrases extraites est d’en moyenne environ 74%, tout en conservant la grammaticalité ou la lisibilité dans une proportion de plus de 64%. Étant donné ces premiers résultats sur un ensemble limité de relations syntaxiques, cela laisse entrevoir des possibilités pour une application de résumé automatique de texte.

A complex network approach to text summarization

Automatic summarization of texts is now crucial for several information retrieval tasks owing to the huge amount of information available in digital media, which has increased the demand for simple, language-independent extractive summarization strategies. In this paper, we employ concepts and metrics of complex networks to select sentences for an extractive summary. The graph or network representing one piece of text consists of nodes corresponding to sentences, while edges connect sentences that share common meaningful nouns. Because various metrics could be used, we developed a set of 14 summarizers, generically referred to as CN-Summ, employing network concepts such as node degree, length of shortest paths, d-rings and k-cores. An additional summarizer was created which selects the highest ranked sentences in the 14 systems, as in a voting system. When applied to a corpus of Brazilian Portuguese texts, some CN-Summ versions performed better than summarizers that do not employ deep linguistic knowledge, with results comparable to state-of-the-art summarizers based on expensive linguistic resources. The use of complex networks to represent texts appears therefore as suitable for automatic summarization, consistent with the belief that the metrics of such networks may capture important text features. (c) 2008 Elsevier Inc. All rights reserved.

Static Video Summarization through Optimum-Path Forest Clustering

This paper introduces the Optimum-Path Forest (OPF) classifier for static video summarization, being its results comparable to the ones obtained by some state-of-the-art video summarization techniques. The experimental section has been conducted using several image descriptors in two public datasets, followed by an analysis of OPF robustness regarding one ad-hoc parameter. Future works are guided to improve OPF effectiveness on each distinct video category.

Summarization in Math Class

This action research study of twenty students in my sixth grade mathematics classroom examines the implementation of summarization strategies. Students were taught how to summarize concepts and how to explain their thinking in different ways to the teacher and their peers. Through analysis of students’ summaries of concepts from lessons that I taught, tests scores, and student journals and interviews, I discovered that summarizing mathematical concepts offers students an engaging opportunity to better understand those concepts and render that understanding more visible to the teacher. This analysis suggests that non-traditional summarization, such as verbal and written strategies, and strategies involving movement and discussions, can be useful in mathematics classrooms to improve student understanding, engagement in learning tasks, and as a form of formative assessment.

Extractive summarization using complex networks and syntactic dependency

The realization that statistical physics methods can be applied to analyze written texts represented as complex networks has led to several developments in natural language processing, including automatic summarization and evaluation of machine translation. Most importantly, so far only a few metrics of complex networks have been used and therefore there is ample opportunity to enhance the statistics-based methods as new measures of network topology and dynamics are created. In this paper, we employ for the first time the metrics betweenness, vulnerability and diversity to analyze written texts in Brazilian Portuguese. Using strategies based on diversity metrics, a better performance in automatic summarization is achieved in comparison to previous work employing complex networks. With an optimized method the Rouge score (an automatic evaluation method used in summarization) was 0.5089, which is the best value ever achieved for an extractive summarizer with statistical methods based on complex networks for Brazilian Portuguese. Furthermore, the diversity metric can detect keywords with high precision, which is why we believe it is suitable to produce good summaries. It is also shown that incorporating linguistic knowledge through a syntactic parser does enhance the performance of the automatic summarizers, as expected, but the increase in the Rouge score is only minor. These results reinforce the suitability of complex network methods for improving automatic summarizers in particular, and treating text in general. (C) 2011 Elsevier B.V. All rights reserved.

Towards a unified framework for opinion retrieval, mining and summarization

The exponential increase of subjective, user-generated content since the birth of the Social Web, has led to the necessity of developing automatic text processing systems able to extract, process and present relevant knowledge. In this paper, we tackle the Opinion Retrieval, Mining and Summarization task, by proposing a unified framework, composed of three crucial components (information retrieval, opinion mining and text summarization) that allow the retrieval, classification and summarization of subjective information. An extensive analysis is conducted, where different configurations of the framework are suggested and analyzed, in order to determine which is the best one, and under which conditions. The evaluation carried out and the results obtained show the appropriateness of the individual components, as well as the framework as a whole. By achieving an improvement over 10% compared to the state-of-the-art approaches in the context of blogs, we can conclude that subjective text can be efficiently dealt with by means of our proposed framework.