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.

Self-supervised information retrieval: a novel approach based on Deep Metric Learning and Neural Language Models

Most of the existing open-source search engines, utilize keyword or tf-idf based techniques to find relevant documents and web pages relative to an input query. Although these methods, with the help of a page rank or knowledge graphs, proved to be effective in some cases, they often fail to retrieve relevant instances for more complicated queries that would require a semantic understanding to be exploited. In this Thesis, a self-supervised information retrieval system based on transformers is employed to build a semantic search engine over the library of Gruppo Maggioli company. Semantic search or search with meaning can refer to an understanding of the query, instead of simply finding words matches and, in general, it represents knowledge in a way suitable for retrieval. We chose to investigate a new self-supervised strategy to handle the training of unlabeled data based on the creation of pairs of ’artificial’ queries and the respective positive passages. We claim that by removing the reliance on labeled data, we may use the large volume of unlabeled material on the web without being limited to languages or domains where labeled data is abundant.

Self-supervised Information Retrieval basato su deep metric learning

La tesi ha lo scopo di ricercare, esaminare ed implementare un sistema di Machine Learning, un Recommendation Systems per precisione, che permetta la racommandazione di documenti di natura giuridica, i quali sono già stati analizzati e categorizzati appropriatamente, in maniera ottimale, il cui scopo sarebbe quello di accompagnare un sistema già implementato di Information Retrieval, istanziato sopra una web application, che permette di ricercare i documenti giuridici appena menzionati.

Graph-based Vision-Language Model for Multi-Modal Retrieval of Fashion Products

Artificial Intelligence is reshaping the field of fashion industry in different ways. E-commerce retailers exploit their data through AI to enhance their search engines, make outfit suggestions and forecast the success of a specific fashion product. However, it is a challenging endeavour as the data they possess is huge, complex and multi-modal. The most common way to search for fashion products online is by matching keywords with phrases in the product's description which are often cluttered, inadequate and differ across collections and sellers. A customer may also browse an online store's taxonomy, although this is time-consuming and doesn't guarantee relevant items. With the advent of Deep Learning architectures, particularly Vision-Language models, ad-hoc solutions have been proposed to model both the product image and description to solve this problems. However, the suggested solutions do not exploit effectively the semantic or syntactic information of these modalities, and the unique qualities and relations of clothing items. In this work of thesis, a novel approach is proposed to address this issues, which aims to model and process images and text descriptions as graphs in order to exploit the relations inside and between each modality and employs specific techniques to extract syntactic and semantic information. The results obtained show promising performances on different tasks when compared to the present state-of-the-art deep learning architectures.

Frame-driven Extraction of Linked Data and Ontologies from Text

Ontology design and population -core aspects of semantic technologies- re- cently have become fields of great interest due to the increasing need of domain-specific knowledge bases that can boost the use of Semantic Web. For building such knowledge resources, the state of the art tools for ontology design require a lot of human work. Producing meaningful schemas and populating them with domain-specific data is in fact a very difficult and time-consuming task. Even more if the task consists in modelling knowledge at a web scale. The primary aim of this work is to investigate a novel and flexible method- ology for automatically learning ontology from textual data, lightening the human workload required for conceptualizing domain-specific knowledge and populating an extracted schema with real data, speeding up the whole ontology production process. Here computational linguistics plays a fundamental role, from automati- cally identifying facts from natural language and extracting frame of relations among recognized entities, to producing linked data with which extending existing knowledge bases or creating new ones. In the state of the art, automatic ontology learning systems are mainly based on plain-pipelined linguistics classifiers performing tasks such as Named Entity recognition, Entity resolution, Taxonomy and Relation extraction [11]. These approaches present some weaknesses, specially in capturing struc- tures through which the meaning of complex concepts is expressed [24]. Humans, in fact, tend to organize knowledge in well-defined patterns, which include participant entities and meaningful relations linking entities with each other. In literature, these structures have been called Semantic Frames by Fill- 6 Introduction more [20], or more recently as Knowledge Patterns [23]. Some NLP studies has recently shown the possibility of performing more accurate deep parsing with the ability of logically understanding the structure of discourse [7]. In this work, some of these technologies have been investigated and em- ployed to produce accurate ontology schemas. The long-term goal is to collect large amounts of semantically structured information from the web of crowds, through an automated process, in order to identify and investigate the cognitive patterns used by human to organize their knowledge.

Query rewriting in information retrieval: automatic context extraction from local user documents to improve query results

The central objective of research in Information Retrieval (IR) is to discover new techniques to retrieve relevant information in order to satisfy an Information Need. The Information Need is satisfied when relevant information can be provided to the user. In IR, relevance is a fundamental concept which has changed over time, from popular to personal, i.e., what was considered relevant before was information for the whole population, but what is considered relevant now is specific information for each user. Hence, there is a need to connect the behavior of the system to the condition of a particular person and his social context; thereby an interdisciplinary sector called Human-Centered Computing was born. For the modern search engine, the information extracted for the individual user is crucial. According to the Personalized Search (PS), two different techniques are necessary to personalize a search: contextualization (interconnected conditions that occur in an activity), and individualization (characteristics that distinguish an individual). This movement of focus to the individual's need undermines the rigid linearity of the classical model overtaken the ``berry picking'' model which explains that the terms change thanks to the informational feedback received from the search activity introducing the concept of evolution of search terms. The development of Information Foraging theory, which observed the correlations between animal foraging and human information foraging, also contributed to this transformation through attempts to optimize the cost-benefit ratio. This thesis arose from the need to satisfy human individuality when searching for information, and it develops a synergistic collaboration between the frontiers of technological innovation and the recent advances in IR. The search method developed exploits what is relevant for the user by changing radically the way in which an Information Need is expressed, because now it is expressed through the generation of the query and its own context. As a matter of fact the method was born under the pretense to improve the quality of search by rewriting the query based on the contexts automatically generated from a local knowledge base. Furthermore, the idea of optimizing each IR system has led to develop it as a middleware of interaction between the user and the IR system. Thereby the system has just two possible actions: rewriting the query, and reordering the result. Equivalent actions to the approach was described from the PS that generally exploits information derived from analysis of user behavior, while the proposed approach exploits knowledge provided by the user. The thesis went further to generate a novel method for an assessment procedure, according to the "Cranfield paradigm", in order to evaluate this type of IR systems. The results achieved are interesting considering both the effectiveness achieved and the innovative approach undertaken together with the several applications inspired using a local knowledge base.

Tecniche di text mining per l'autoorganizzazione della conoscenza

L'elaborato ha come scopo l'analisi delle tecniche di Text Mining e la loro applicazione all'interno di processi per l'auto-organizzazione della conoscenza. La prima parte della tesi si concentra sul concetto del Text Mining. Viene fornita la sua definizione, i possibili campi di utilizzo, il processo di sviluppo che lo riguarda e vengono esposte le diverse tecniche di Text Mining. Si analizzano poi alcuni tools per il Text Mining e infine vengono presentati alcuni esempi pratici di utilizzo. Il macro-argomento che viene esposto successivamente riguarda TuCSoN, una infrastruttura per la coordinazione di processi: autonomi, distribuiti e intelligenti, come ad esempio gli agenti. Si descrivono innanzi tutto le entità sulle quali il modello si basa, vengono introdotte le metodologie di interazione fra di essi e successivamente, gli strumenti di programmazione che l'infrastruttura mette a disposizione. La tesi, in un secondo momento, presenta MoK, un modello di coordinazione basato sulla biochimica studiato per l'auto-organizzazione della conoscenza. Anche per MoK, come per TuCSoN, vengono introdotte le entità alla base del modello. Avvalendosi MoK dell'infrastruttura TuCSoN, viene mostrato come le entità del primo vengano mappate su quelle del secondo. A conclusione dell'argomento viene mostrata un'applicazione per l'auto-organizzazione di news che si avvale del modello. Il capitolo successivo si occupa di analizzare i possibili utilizzi delle tecniche di Text Mining all'interno di infrastrutture per l'auto-organizzazione, come MoK. Nell'elaborato vengono poi presentati gli esperimenti effettuati sfruttando tecniche di Text Mining. Tutti gli esperimenti svolti hanno come scopo la clusterizzazione di articoli scientifici in base al loro contenuto, vengono quindi analizzati i risultati ottenuti. L'elaborato di tesi si conclude mettendo in evidenza alcune considerazioni finali su quanto svolto.

Dynamic networks, text analysis and Gephi: the art math

In numerosi campi scientici l'analisi di network complessi ha portato molte recenti scoperte: in questa tesi abbiamo sperimentato questo approccio sul linguaggio umano, in particolare quello scritto, dove le parole non interagiscono in modo casuale. Abbiamo quindi inizialmente presentato misure capaci di estrapolare importanti strutture topologiche dai newtork linguistici(Degree, Strength, Entropia, . . .) ed esaminato il software usato per rappresentare e visualizzare i grafi (Gephi). In seguito abbiamo analizzato le differenti proprietà statistiche di uno stesso testo in varie sue forme (shuffolato, senza stopwords e senza parole con bassa frequenza): il nostro database contiene cinque libri di cinque autori vissuti nel XIX secolo. Abbiamo infine mostrato come certe misure siano importanti per distinguere un testo reale dalle sue versioni modificate e perché la distribuzione del Degree di un testo normale e di uno shuffolato abbiano lo stesso andamento. Questi risultati potranno essere utili nella sempre più attiva analisi di fenomeni linguistici come l'autorship attribution e il riconoscimento di testi shuffolati.