230 research outputs found
Classifying and mapping e-Tourism data sets
Full text linkHeutzutage kann die Online-Recherche vor der Buchung eines Urlaubs als übliche Ge- wohnheit der Kunden angesehen werden. In diesem Zusammenhang zielen Recommender Systeme darauf ab, die Kunden bei ihrer Suche nach den richtigen Produkten zu unterstüt- zen. Jedoch stehen solche Systeme domänenspezifischen Herausforderungen gegenüber, da Tourismusprodukte typischerweise sehr komplex und mit Emotionen verbunden sind. Um diesen Herausforderungen entgegen zu treten, wurden umfassende Benutzermodelle entwickelt, welche die Präferenzen, die Anforderungen und die Persönlichkeit von Kunden berücksichtigen. Eines dieser Modelle ist das sogenannte Sieben-Faktoren-Modell. In dieser Arbeit werden verschiedene Methoden zur automatisierten Bestimmung der Sieben- Faktoren von Tourismusdestinationen und Hotels untersucht, um Recommender Systeme zu ermöglichen die passendsten Produkte vorzuschlagen. Insbesondere werden explorative Datenanalysen, Clusteranalysen und Regressionsanalysen durchgeführt, um nicht nur die Sieben-Faktoren von Tourismusdestinationen und Hotels zu bestimmen, sondern auch ausschlaggebende Attribute von Tourismusdestinationen und Hotels zu identifizieren. Die Resultate der Clusteranalysen zeigen, dass ähnliche Tourismusdestinationen und auch ähnliche Hotels gruppiert werden können. Die identifizierten Gruppen können mit den Sieben-Faktoren assoziiert werden. Die Ergebnisse der Clusteranalysen ermöglichen es nicht einzelne Faktoren des Sieben-Faktoren-Modells zu bestimmen, aber können für eine direkte Zuordnung verwendet werden. Im Gegensatz zu den Clusteranalysen liefern die Regressionsanalysen einen klaren Beweis dafür, dass die Sieben-Faktoren von Tourismusdestinationen und Hotels unter Berücksichtigung der jeweiligen Attribute bestimmt werden können. Grundsätzlich variiert die Qualität der entwickelten Modelle für verschiedene Faktoren des Sieben-Faktoren-Modells und auch für verschiedene Touris- musprodukte (Destination und Hotels). Der in dieser Arbeit vorgestellte Ansatz kann für neue Datenquellen und auch Produkttypen leicht nachvollzogen werden.Nowadays, researching online before booking a vacation can be seen as a common habit of customers. In this context, Recommender Systems (RSs) are aiming to support the customers to find the right products, but they face domain specific challenges since tourism products are typically very complex and related to emotional experiences. To counteract these challenges, comprehensive user models for capturing the preferences and personality of travelers have been introduced. One of these models is the so-called Seven-Factor Model. This work introduces an automated way for determining the Seven- Factor representation of tourism destinations and hotels to enable a matchmaking for RSs. In particular, exploratory data analyses, cluster analyses, and regression analyses are conducted not only to find a mapping of tourism destinations and hotels onto the Seven- Factors, but also to foster a better understanding of the relationship between destination attributes and the Seven-Factors, and between hotel attributes and the Seven-Factors. The main results show that conceptually meaningful groups of destinations and hotels as well can be identified and associated with the Seven-Factors, but they can only be used for direct allocations rather than for determining each factor of the Seven-Factor Model. Furthermore, the regression analyses provide clear evidence that a tourism destination’s Seven-Factor representation and a hotel’s Seven-Factor representation can be determined by taking the respective attributes into account. In general, the quality of the developed models varies for different factors of the Seven-Factor Model and also for different tourism products (i.e., destination and hotels). Finally, the introduced approach can easily be followed for new data sources and product types
Leveraging the Subtle: Hidden Factors in Recommender Systems
Full text linkRecommender systems are pivotal in various domains, aiding users in their decision-making. However, current systems often overlook subtle factors that significantly impact user preferences and choices. This work aims to bridge this gap by exploring the conceptof implicit item characteristics -- latent features that influence user decision-making in addition to explicit content. The investigation is divided into three key research areas. Firstly, we explore how to systematically identify and expose implicit item characteristics to enhance recommender systems in two key domains: tourism and news. Using advanced analytics such as cluster analysis and multiple linear regression, we map tourist destinations to the established Seven-Factor Model in tourism. In the news, we employ natural language processing techniques to reveal hidden features essential for tailoring recommendations. Secondly, we introduce a novel system called PicTouRe to elicit tourists' implicit preferences through pictures. Leveraging convolutional neural networks, we translate visual preferences into a Seven-Factor profile for each user, simplifying decision-making and capturing both immediate touristic desires and enduring personality traits. Lastly, we enhance news recommender systems by leveraging sentiment and emotions of news articles. Two models, RobustSentiRec and EmoRec, were developed to capture these implicit characteristics, aligning recommendations more closely with user preferences but also raising ethical concerns around potential sentiment and emotional echo chambers. Our findings offer a robust framework for more nuanced, user-sensitive recommendations, opening new avenues for future research and applications in recommender systems
Crystal Structures of Cobalamin-Independent Methionine Synthase (MetE) from Streptococcus mutans: A Dynamic Zinc-Inversion Model
No full textCobalamin-independent methionine synthase (MetE) catalyzes the direct transfer of a methyl group from methyltetrahydrofolate to L-homocysteine to form methionine. Previous studies have shown that the MetE active site coordinates a zinc atom, which is thought to act as a Lewis acid and plays a role in the activation of thiol. Extended X-ray absorption fine structure studies and mutagenesis experiments identified the zinc-binding site in MetE from Escherichia coli. Further structural investigations of MetE from Thermotoga maritima lead to the proposition of two models: "induced fit" and "dynamic equilibrium", to account for the catalytic mechanisms of MetE. Here, we present crystal structures of oxidized and zinc-replete MetE from Streptococcus mutans at the physiological pH. The structures reveal that zinc is mobile in the active center and has the possibility to invert even in the absence of homocysteine. These structures provide evidence for the dynamic equilibrium model. (C) 2011 Elsevier Ltd. All rightshttp://gateway.webofknowledge.com/gateway/Gateway.cgi?GWVersion=2&SrcApp=PARTNER_APP&SrcAuth=LinksAMR&KeyUT=WOS:000295496500012&DestLinkType=FullRecord&DestApp=ALL_WOS&UsrCustomerID=8e1609b174ce4e31116a60747a720701Biochemistry & Molecular BiologySCI(E)PubMed7ARTICLE4688-69741
Exploring Expressed Emotions for Neural News Recommendation
No full textDue to domain-specific challenges such as short item lifetimes and continuous cold-start issues, news recommender systems rely more on content-based methods to deduce reliable user models and make personalized recommendations. Research has shown that alongside the content of an item, the way it is presented to the users also plays a critical role. In this work, we focus on the effect of incorporating expressed emotions within news articles on recommendation performance. We propose a neural news recommendation model that disentangles semantic and emotional modeling of news articles and users. While we exploit the textual content for the semantic representation, we extract and combine emotions of different information levels for the emotional representation. Offline experiments on a real-world dataset show that our approach outperforms non-emotion-aware solutions significantly. Finally, we provide a future outline, where we plan to investigate a) the online performance and b) the explainability/explorability of our approach
An IBC and certificate based hybrid approach to WIMAX security
No full textWiMAX is a promising technology that provides high data throughput with low delays for various user types and modes of operation. These advantages make WiMAX applicable both for infrastructure purposes and end-client usage. Since WiMAX is presented as a network framework and a last-mile technology, it is believed to be capable of handling a wide range of usage scenarios. For example, while the end users have an opportunity to use WiMAX as the primary connection medium for acquiring services such as on-demand video streaming, VoIP connections and mobile bank transactions, the service providers may use it for data relaying purposes among access points. To meet the technical requirements of these various scenarios, majority of the WiMAX research has been conducted on physical and MAC layers; however little has been invested in a comprehensive and efficient security solution, which has resulted in a wide range of security weaknesses and reactive solutions. Many security problems remain to be addressed in different modes and for different user types even in the final security standard of WiMAX, PKMv2. In this thesis, we present a hybrid security solution combining Identity-Based Cryptography (IBC) and certificate based approaches to overcome the existing security problems of WiMAX without degrading service quality. IBC has potential benefits that can provide enhancements to the overall security and efficiency of the security standard. One such enhancement is combining user identity with the public key and therefore eliminating the public key distribution load from the network. However, IBC has a few caveats, such as the necessity of a secure medium to distribute private keys. To compensate for these disadvantages, in this study, IBC is combined with certificate-based security. As a result, the benefits of IBC are maintained while the disadvantages are eliminated. Using the hybrid approach, this study also aims to clarify the key revocation procedures and key lifetimes of WiMAX. To achieve this goal, key renewal intervals are examined and corresponding lifetimes are assigned to the credentials missing in both PKMv2 and PKMv1. Additionally, the key distribution procedures are investigated and a pattern is provided with the message exchange details. To be able to correctly assess the efficiency of this approach, a new mobility model is defined in the evaluation chapter of this thesis. Based on this model, the analysis has shown that our hybrid solution that combines IBC and the certified based security scheme results in a significant bandwidth improvement over the standard’s approach, PKMv2. This work is the first study that unites the advantages of both IBC and the certified-based security scheme for improved security while maintaining low overhead for WiMAX.M.S.Includes bibliographical referencesIncludes abstrac
A recommender system for the matchmaking of event participants
Full text linkViele Teile unseres Lebens werden in letzter Zeit digitalisiert. Insbesondere Veranstaltungen, die traditionell persönlich abgehalten wurden, wechseln nun zu digitalen und hybriden Formaten. Solche Formate erschweren es, die richtigen Personen zu treffen, insbesondere bei großen Veranstaltungen. Ziel dieser Arbeit ist es, ein Empfehlungssystem ("Recommender System") für Business-to-Business (B2B)-Veranstaltungen zu entwerfen, zu implementieren und zu evaluieren, das jedem Teilnehmer eine personalisierte Liste von Teilnehmern zur Verfügung stellt, die für ihn oder sie von Interesse sein könnten. Dabei wird aus vergangenen Interaktionen gelernt und ausschließlich auf implizites Feedback wie Besuche, Lesezeichen, Nachrichten und gebuchte Meetings zurückgegriffen. Wir versuchen auch Empfehlungen für Teilnehmer zu machen, die keine vergangenen Interaktionen haben, indem wir ähnliche Benutzer in den Empfehlungsprozess einbeziehen. Die Daten werden von der Firma b2match bereitgestellt, die eine Online-Plattform zur Verwaltung von Veranstaltungen anbietet. Das Empfehlungssystem wird iterativ mit dem CRISP-DM- Prozess entwickelt, als eigenständiger Service implementiert und in die b2match-Plattform integriert. Um zu überprüfen, ob das Empfehlungssystem gut funktioniert, führen wir eine Offline-Evaluierung gegen Baselines (18 Veranstaltungen für die Entwicklung, 6 Veranstaltungen für die Evaluierung) und eine Online-Evaluierung auf 27 Veranstaltungen durch, die das Empfehlungssystem in der Produktion verwenden. Für die Bewertung mitteln wir die nDCG@10-Ranking-Metrik auf den personalisierten Listen von Teilnehmern aus, wobei jeder Teilnehmer eine Liste erhält. Die Ergebnisse sind vielversprechend. Das entwickelte Empfehlungssystem schneidet in einer Offline-Evaluierung signifikant besser ab als alle Baselines mit einem nDCG@10-Score von 0,1967. Die Ergebnisse der Baselines waren wie folgt: 0,0361 für eine Liste, die aus der Normalverteilung gezogen wurde (p = 0,0044), 0,0716 für ein Popularitätsranking (p = 0,0073), 0,0277 für zufällige Listen (p = 0,0045), 0,0452 für ein Ähnlichkeitsranking (p = 0,0051). Durch Hinzufügen eines hybriden Empfehlungssystems zur Lösung des Cold-Start-Problems konnten wir eine Verbesserung in Bezug auf nDCG@10 von 0,1967 auf 0,2227 (p = 0,0051) erreichen. Das entwickelte Empfehlungssystem erhöht auch die relative Anzahl erfolgreicher Meetings in einer Online-Evaluierung von 0,18% auf 0,31% (p = 0,0005). Unsere Studie kommt zu dem Schluss, dass Matrixfaktorisierungsalgorithmen auf unserem B2B-Event-Datensatz am besten abschneiden. Wenn ein Cold-Start-Szenario auftritt und keine Vorhersage für einen Teilnehmer getroffen werden kann, liefern personalisierte Empfehlungen auf der Grundlage von Interaktionsdaten ähnlicher Teilnehmer vielversprechende Ergebnisse.Recently, many parts of our lives are becoming digitized. Especially events, which were traditionally held in-person, are now migrating to digital and hybrid formats. Such formats make it harder to find the right people to meet, particularly at large events. The goal of this thesis is to design, implement and evaluate a recommender system for business-to-business (B2B) events that provides to each participant a personalized list of participants that might be of interest to them, by learning from past interactions, relying exclusively on implicit feedback, such as visits, bookmarks, messages and meetings booked. We also attempt to make recommendations for participants that have no past interactions, by incorporating similar users into the recommendation process. Data is provided by the company b2match, which provides an online platform for managing events. The recommender system is developed iteratively using the CRISP-DM process, implemented as a standalone service, and integrated into the b2match platform. To verify that the recommender system works well, we do an offline evaluation against baselines (18 events for development, 6 events for evaluation) and an online evaluation on 27 events that use the recommender system in production. For the evaluation we average the nDCG@10 ranking metric on the personalized lists of participants, one list provided for each participant. The results are promising. The developed recommender system performs significantly better than all baselines in an offline evaluation, with an nDCG@10 score of 0.1967. The results of the baselines were as follows: 0.0361 for a list sampled from the normal distribution (p = 0.0044), 0.0716 for a popularity ranking (p = 0.0073), 0.0277 for random lists (p = 0.0045), 0.0452 for a similarity ranking (p = 0.0051). Adding a hybrid recommender to solve the cold start problem, we were able to achieve an improvement in terms of nDCG@10 from 0.1967 to 0.2227 (p = 0.0051). The developed recommender also increases the relative number of successful meetings in an online evaluation from 0.18% to 0.31% (p = 0.0005). Our study concludes that matrix factorization-based algorithms perform best on our B2B event data set. When a cold-start scenario arises and a prediction cannot be made for a participant, providing personalized recommendations based on interaction data from similar participants yields promising results
- …
