• Digital Humanities and Cultural Heritage
• Embargo close
• Publications close
• UPCommons. Portal del coneixement o... close

Aquesta tesi presenta una estratègia innovadora sobre l'ús de transaccions bancàries per generar representacions d'empreses amb embeddings. El treball investiga dos mètodes diferents, un encoder basat en Xarxes Neuronals Convolucionals (CNNs) per processar sèries temporals i una combinació del model d'Embeddings d'OpenAI i l'arquitectura Transformer per al processament del llenguatge natural contingut en les dades transaccionals. El CNN Encoder utilitza sèries temporals de transaccions agregades setmanalment mentre que el Transformer aprofita la informació textual present en les transaccions bancàries, processant-la com a text. Aquestes metodologies han estat seleccionades per aprendre representacions vectorials de menor dimensionalitat capaces de capturar la informació continguda en les transaccions bancàries. Aquestes representacions vectorials poden ser aplicades a diferents tasques subjacents com l'estimació dels ingressos anuals recurrents (ARR), el creixement o el model de negoci de cada empresa. Aquest treball estableix una base per a l'aplicació de l'aprenentatge automàtic en el procés d'avaluació del risc per a la subscripció de crèdits, apuntant a la necessitat d'un conjunt de dades més extens, millors enriquidors de transaccions i models més sofisticats. Serveix com a punt de partida per a investigacions futures sobre l'ús de transaccions bancàries per comprendre el comportament financer de les empreses. This thesis presents an innovative approach to utilize banking transactions to generate company representations in the form of embeddings. The work investigates two different methodologies, a Convolutional Neural Network (CNN) encoder for time series data and a combination of OpenAI Embeddings model and the Transformer architecture for natural language processing of transaction data. The CNN Encoder makes use of transaction time series data aggregated into weekly windows, while the Transformer-based approach leverages the rich textual information present in transaction data, treating it as natural language. These methodologies are chosen to learn lower-dimensional embeddings capable of capturing the information in banking transactions that can be applied across various downstream tasks, such as estimating the Annual Recurring Revenue (ARR), Growth, and Business Model of a company. This work lays a foundation for the application of machine learning in the underwriting process, pointing to the need for more extensive datasets, better transaction enrichers, and more sophisticated models. It serves as a starting point for further research in leveraging banking transaction data for understanding the financial behavior of companies.

Rentals United és una empresa de gestió de canals, de manera que ofereix solucions de programari per a propietaris i gestors de propietats enfocades al seu lloguer per vacances. Ja que opera en un sector molt volàtil, és una prioritat per a l'empresa mantenir els clients existents i aprofitar les noves oportunitats que sorgeixen. L'objectiu principal d'aquest treball és partir de l'actual sistema de Rentals United per a la retenció de clients i construir un nou conjunt de processos de transformació de les dades. El seu origen radica en els sistemes d'emmagatzematge de dades de l'empresa i, mitjançant mètodes i algoritmes basats en dades, dóna lloc a una taula final de dades on cada client hi és representat. D'una banda, en aquesta taula es poden trobar indicacions per tal de prevenir la pèrdua de clients, per mitjà d'accions construïdes a partir d'arbres de decisió. D'altra banda, hi ha oportunitats de creixement personalitzades i quantificades per tal que els clients puguin ampliar el seu catàleg de productes actual (augmentant així els ingressos de l'empresa). Les mètriques es divideixen en diferents àrees i es calculen cada dia, en qüestió de minuts. Al mateix temps, es reforcen processos paral·lels, destinats a incrementar la fiabilitat global de les dades i a trobar possibles inconsistències. Aquest treball permet l'extracció d'una llista de clients amb un risc de pèrdua estimat elevat. A més, s'obtenen estadístiques individuals i agregades, les quals donen lloc a diferents informes gràfics que els membres de Rentals United ja estan utilitzant a diari per a entendre el comportament dels clients. S'avaluen diferents sistemes de processament del llenguatge per a aprofundir en la integració dels resultats del projecte en l'operació habitual de l'empresa. De fet, es millora un bot de conversa existent per a incorporar les propostes del projecte. A més, es plantegen i defineixen les bases d'un model supervisat que faci servir aquestes dades, pendent de ser implementat en el futur. En resum, el coneixement de l'empresa sobre el comportament dels clients s'amplia i es fa més usable. Rentals United is a channel management company that offers software solutions for property managers and owners to handle their vacation rental properties. It operates in a very volatile sector; thereby it is a priority for the company to maintain existing clients and to leverage new opportunities. The main objective of this thesis is to depart from Rentals United's current approaches to customer retention in order to build a solid new pipeline. It originates in the company's data warehouses and, through data-driven methods and algorithms, gives rise to a final data table where each client is represented. On the one hand, guidance to prevent churns may be found in this table by means of a set of decision tree - based actionable insights. On the other hand, there are customized and quantified growth opportunities for customers to expand their current portfolio (increasing the company's revenue). Metrics are segmented into different areas and re-computed every day, in a matter of minutes. At the same time, parallel data processes are reinforced to boost data reliability and find inconsistences. This work allows the extraction of a shortlist of clients with estimated high churn risk at present. Moreover, user-by-user and aggregated statistics are obtained and feed reports that Rentals United members are already using to understand clients' behaviour. Different approaches to language processing are tested to further integrate the results into the company's operation flow. In fact, an existing chat bot is strengthened to incorporate the insights of the project. Furthermore, the groundworks for a supervised model to exploit the new data are laid and well-defined to be implemented in the future. In summary, the company's knowledge about the customers' operation is widened and made more usable.

Con el ánimo de hacer estructuras cada vez más seguras, los ingenieros buscamos día a día nuevas técnicas y nuevos métodos que permitan minimizar los riesgos y las probabilidades de fallo de las edificaciones y sus elementos ante eventos naturales. Tal es el caso de las acciones generadas por las fuerzas producto de presiones de viento y sismos, las cuales actúan sobre las fachadas de los edificios que en la gran mayoría de los casos están constituidas por muros en obra de fábrica. En ocasiones se generan daños que llevan al colapso de los elementos no estructurales, debido a la baja o nula ductilidad que presenta un muro de fabrica sin ningún tipo de reforzamiento. Es por ello que este trabajo está enfocado en investigar el comportamiento a flexión fuera del plano de muros de fábrica no armados y muros de fábrica armada, para este caso se ha comparado dos diferentes alternativas de refuerzo de junta horizontal, una la convencional mediante cerchas o escalerillas conformadas por barras de acero, y la otra mediante una técnica específica con una malla compuesta de alambres de acero de alta resistencia entretejidos con fibra de vidrio (Murfor compact, de Bekaert), para ello se ha dividido los 15 muros de fabrica en 5 grupos: 4 fábricas tipo URM (muros de fabrica no armadas), 2 fábricas tipo RM-2M con Murfor® RND/Z-4-080 de 2 armaduras de tendel (separación 0.6 m), 2 fábricas tipo RM-2MC con Murfor® Compact I-80 de 2 armaduras de tendel (separación 0.6 m), 2 fábricas tipo RM-3M con Murfor® RND/Z-4-080 de 3 armaduras de tendel (separación 0.4 m) y 4 fábricas tipo RM-3MC con Murfor® Compact I-80 de 3 armaduras de tendel (separación 0.4 m). With the aim of making structures increasingly safer, engineers seek new techniques and methods every day that allow minimizing the risks and probabilities of failure of buildings and their elements in the face of natural events. This is the case with the actions generated by forces resulting from wind pressures and earthquakes, which act on the facades of buildings that in the majority of cases are made up of masonry walls. On occasions, damages are generated that lead to the collapse of non-structural elements, due to the low or zero ductility presented by an un-reinforced masonry wall. That is why this work is focused on investigating the out-of-plane flexural behavior of unreinforced and reinforced masonry walls. For this case, two different alternatives for horizontal joint reinforcement have been compared: the conventional method using steel bars formed as trusses or ladders, and the other using a specific technique with a composite mesh made up of high-strength steel wires woven with fiberglass (Murfor compact, from Bekaert). For this purpose, the 15 masonry walls have been divided into 5 groups: 4 URM-type masonry walls (unreinforced masonry walls), 2 RM-2M-type masonry walls with Murfor® RND/Z-4-080 reinforcement and 2 tendons (0.6 m spacing), 2 RM-2MC-type masonry walls with Murfor® Compact I-80 reinforcement and 2 tendons (0.6 m spacing), 2 RM-3M-type masonry walls with Murfor® RND/Z-4-080 reinforcement and 3 tendons (0.4 m spacing), and 4 RM-3MC-type masonry walls with Murfor® Compact I-80 reinforcement and 3 tendons (0.4 m spacing). Amb l'ànim de fer estructures cada cop més segures, els enginyers busquem dia a dia noves tècniques i nous mètodes que permetin minimitzar els riscos i les probabilitats de fallada de les edificacions i els seus elements davant d'esdeveniments naturals. Aquest és el cas de les accions generades per les forces producte de pressions de vent i sismes, les quals actuen sobre les façanes dels edificis que en la gran majoria dels casos estan constituïdes per murs a obra de fàbrica. De vegades es generen danys que porten al col·lapse dels elements no estructurals, a causa de la baixa o nul·la ductilitat que presenta un mur de fàbrica sense cap mena de reforçament. És per això que aquest treball està enfocat a investigar el comportament a flexió fora del pla de murs de fàbrica no armats i murs de fàbrica armada. escaletes conformades per barres d'acer, i l'altra mitjançant una tècnica específica amb una malla composta de filferros d'acer d'alta resistència entreteixits amb fibra de vidre (Murfor compact, de Bekaert), per això s'han dividit els 15 murs de fàbrica en 5 grups: 4 fàbriques tipus URM (murs de fàbrica no armades), 2 fàbriques tipus RM-2M amb Murfor® RND/Z-4-080 de 2 armadures de tendel (separació 0.6 m), 2 fàbriques tipus RM-2MC amb Murfor® Compact I-80 de 2 armadures de tendell (separació 0.6 m), 2 fàbriques tipus RM-3M amb Murfor® RND/Z-4-080 de 3 armadures de tendell (separació 0.4 m) i 4 fàbriques tipus RM-3MC amb Murfor ® Compact I-80 de 3 armadures de tendell (separació 0.4 m).

Food wastage is a major problem in today’s world and the distribution of food surplus is undoubtedly one of the leading solutions to the problem. Surplus distribution is widely practised across the globe and there are several communities for the same as well. However, the success of these movements are determined by the active participation of volunteers. In addition, volunteer availability is an issue that is dependent on several different factors. An incentive module for the volunteers can normalize the availability of the volunteers. Such incentive modules are prone to identity duplication attacks and the implemented solutions do not really encourage volunteer hierarchical structures. In this paper, an incentive module that resists identity duplication and promotes the hierarchical structure between volunteers is presented. In addition, an improved matching module to match the food donors to the receivers utilizing the available volunteers optimally is also proposed "Fatos Xhafa’s work is supported by Project “Optimizing Carsharing and Ridesharing Mobility in Smart Sustainable Cities” (grant 21S09355-001) of the Ajuntament of Barcelona and the Fundació “la Caixa” under the framework of the Barcelona Science Plan 2020–2023" Peer Reviewed

Integrating multi-sensor systems to sort and monitor complex waste streams is one of the most recent innovations in the recycling industry. The complementary strengths of Laser-Induced Breakdown Spectroscopy (LIBS) and computer vision systems offer a novel multi-sensor solution for the complex task of sorting aluminum (Al) post-consumer scrap into alloy groups. This study presents two novel methods for fusing RGB and Depth images with LIBS using Deep Learning models. The first method is a single-output model that combines LIBS UNET and two DenseNets in a late fusion framework. The second method is a multiple-output model that uses the structure of the single-output model to enhance learning and avoid overfitting. In particular, the network has two outputs that enable the regularization of the individual sensors. A data set of 773 aluminum scrap pieces was created with two sets of ground truth-values, corresponding to the two envisaged sorting tasks, to train and evaluate the developed models. The first sorting task is separating Cast and Wrought (C&W) aluminum. The second is the division of the post-consumer aluminum scrap into three commercially interesting fractions. The single-output model performs best for separating C&W, with a Precision, Recall, and F1-score of 99%. The multiple-output model performs best for classifying the three selected commercial fractions, with a Precision, Recall, and F-score of 86%, 83%, and 84%, respectively. The presented data fusion method for LIBS and computer vision images encompasses the great potential for sorting post-consumer aluminum scrap. By sorting mixed post-consumer aluminum scrap in alloy groups, more wrought-to-wrought recycling can occur, and quality losses can be mitigated during recycling. Peer Reviewed

Recent developments in robotic demanufacturing raise the potential to increase the cost-efficiency of recycling and recovering resources from Waste of Electrical and Electronic Equipment (WEEE). However, the industrial adoption of robotic demanufacturing for mixed WEEE streams requires tailored instructions for every product model. Considering the large variation in product models, it is not expected to be feasible in the coming decade to rely only on computer vision technologies to define the tailored instructions required for robust and time-efficient robotic demanufacturing. Therefore, the presented research developed a generic retrieval system named You Only Demanufacture Once (YODO) based on content-based image retrieval (CBIR) to identify the product model and retrieve product model-specific demanufacturing instructions. The system compares the visual features represented on a color image of the WEEE with a database of known descriptions representing previously imaged WEEE to find a match or to figure out whether the analyzed product model is new to the system. The performance of YODO is evaluated with a case study for laptop model identification, where a large dataset is created including 4089 images of a representative laptop waste stream. The results demonstrate a top-1 retrieval mean average precision (mAP) of 93.75%. After running YODO on 3600 laptops, the system learned 1079 unique product models, and the presented results show an 85% chance that the next laptop presented to the system is already registered in the database, allowing the retrieval of relevant information for robotic demanufacturing. This corroborates that a fast learning rate can be achieved, allowing a YODO system to support the robotic demanufacturing by making prior product-specific learnings available. Award-winning

The resistance of the citrus rootstock Forner-Alcaide 5 to Tylenchulus semipenetrans was assessed in comparison to the susceptible Carrizo citrange (S) in two replant situations. Changes in soil and root population densities were monitored after planting the new trees to detect tree infection and nematode population growth over time. The efficacy of pre-plant control methods such as fumigation with 1,3 Dichloropropene (1,3D) or 2-year fallowing was also assessed. Fumigation and fallowing reduced juveniles + males/250 cm3 soil to undetectable levels for 3 and 2 years, respectively, and postponed root parasitism on Forner-Alcaide 5 for 3 and 2.5 years, respectively. Forner-Alcaide 5 supported lower densities of juveniles + males/250 cm3 soil and females/g root than Carrizo citrange in both replant situations The efficacy of the resistant rootstock was enhanced when combined with 1,3D fumigation or 2-year fallowing because the pre-planting control reduced densities of juveniles + males in soil before establishing the orchard, and the resistant rootstock, the female densities in the roots. Selection for virulence within the field population did not occurr after 9-years of exposure of Forner-Alcaide 5 to the nematode. The durability of the resistance can be enhanced by combining rootstock resistance with pre-planting control methods in order to delay root parasitism, and therefore, diminish the selection pressure exerted on the resistant rootstock. Forner-Alcaide 5 exibited higher scion diameter and yield than Carrizo citrange. info:eu-repo/semantics/acceptedVersion