From social data to personalized recommendationsa semantic approach

  1. Mohamed, Menna Allah Maged Moustafa Kamel
Dirixida por:
  1. Alberto Gil Solla Director
  2. Manuel Ramos Cabrer Director

Universidade de defensa: Universidade de Vigo

Fecha de defensa: 29 de xuño de 2022

  1. Yolanda Blanco Fernández Presidenta
  2. Vladimir Robles Bykbaev Secretario/a
  3. Abdullah Radi Ewies Daif Vogal
  1. Enxeñaría telemática

Tipo: Tese


The focus of this thesis has been the investigation of crowdsourcing as a concept and the various aspects associated with it. Relevant crowdsourcing systems do not adequately capture the need to improve the worker-to-task alignment in specific crowdsourcing contexts. Most of the time, the corresponding design features are restricted to standard task sorting, search options, and filter mechanisms relying on general-purpose factors like payment or date. These criteria, however, are incapable of addressing the implicit characteristics of specific tasks, nor of considering the workers’ interests and capabilities. A personalized task recommendation approach in crowdsourcing is proposed that seeks to facilitate the matching of worker interests and preferences with the appropriate tasks, potentially benefiting both workers and requesters. To model an effective crowdsourcing system, the best practices for various recommendation systems that share characteristics with crowdsourcing systems were investigated. A crowdsourcing recommendation system based on a push methodology is proposed and evaluated, with the goal of helping workers instantly find the best matching tasks and assisting requesters in quickly identifying the best workers for their tasks. Part of this work also concentrates on a particular area of Digital Humanities which is Cultural Heritage (CH). Cultural Heritage is one of the most diverse domains that has experienced significant digital transformation. The digitization of holdings is vital for different heritage firms to become an important part of the Internet. Galleries, Libraries, Archives, and Museums (GLAMs), and other organizations have begun to transition to digital transformation, but few of them have the resources to describe or archive digital collection items properly. In GLAMs, there is often a need for annotating and enriching different artworks and cultural items. However, the increased size of collections and items, in addition to the poorly described metadata, and the inefficient annotation process, make it hard to search or retrieve the required elements. It also limits the utilization of value-added services and applications that make use of the cultural materials in effective ways. As a result, crowdsourcing is proposed as a solution as it became one of the most famous approaches adopted for the image annotation process. Crowdsourcing plays a significant role when it comes to digital transformation or manipulating large amounts of data. It has recently emerged as a powerful alternative that allows people (usually referred to as users) to perform tasks either voluntarily or for a financial reward on online platforms issued by an organization or other people (usually referred to as requesters). Some of these tasks do not require that users must be experienced in them and other tasks tackle the cognition and abilities of users to achieve a well-defined result, but the main problem with most crowdsourcing platforms is that offering of tasks to workers are executed randomly allowing users to be assigned to tasks that are unrelated to their field or not interested in. The thesis objectives are to first contribute to the application of recommendation systems in the crourcing domain specifically. In a way that eases on system requesters to find best users to perform their tasks, as well as helps the user to pick the preferred tasks for them too that matches their interests and preferences. Second, to utilize the social media to gather necetask data for users and identify relevant information in it, which in turn helps build better users’ profiles. Moreover, to explore semantic similarities using word embeddings techniques to identify relevant connections between users’ profiles and task data to enhance the recommendation process. Furthermore, utilize the Semantic Web technologies and Linked Data ontology to enhance user-profiles and extend the extracted information. Additionally, design and implement a crowdsourcing recommendation system that minimizes the task search time and helps users find their preferred tasks according to their extracted features. Finally, look into another application where crowdsourcing methods can make a significant contribution, which is the Cultural Heritage domain. As a result, another important goal is to assist Cultural Heritage institutions in improving and expanding their digital cultural archives through the use of the crowd in crowdsourcing systems. To the best of our knowledge, the work we have done has revealed the importance of the crowdsourcing domain and how recommendation systems can support and complement working in it. In addition, the vast information available on social networks can be utilized for building profiles to aid in the recommendation process. As well, the use of ontology and semantic technologies can enrich the data and help structure them, and discover interesting associations and connections between them. These connections can be used to enhance the recommendation methods and even discover semantic connections between users’ profiles and the data on the crowdsourcing platforms. Furthermore, it was demonstrated that there are numerous contexts in which synthetic data can complement and enrich a lack of appropriate datasets. The experience gained through the different experiments carried on showed how various classification techniques can aid and enhance the recommendation methods.