Social data mining strategies for user modelling with personalisation purposes
- Autores: Sandra Servia Rodríguez
- Directores de la Tesis: Rebeca Díaz Redondo (codir. tes.), Ana Fernández Vilas (codir. tes.)
- Lectura: En la Universidade de Vigo ( España ) en 2015
- Idioma: inglés
- Tribunal Calificador de la Tesis: Carlos García Rubio (presid.), Francisco Javier González Castaño (secret.), Lidia Jackowska Strumillo (voc.)
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
The abundance of information in the online world results in a growing demand for relevant content, making any service in this medium a perfect environment within which personalisation could blossom. The availability of information about users' interests, opinions and so on in online social sites facilitates the effective modelling of users and services, enabling avoidance of the well-known issues in personalisation that come up when new users or items are added to the system -the "cold-start problem"-. In an attempt to externalise the provision of personalisation to online services, thereby allowing them to be exclusively focused on their tasks, the thesis of this dissertation is that an intermediary model of the user constructed by properly mining user generated content in social media can be exploited to create or improve technological social applications. The model we propose is based on representing users' online life by means of what we call "social spheres" and considering only users' data available from public APIs (private messages, retweets, etc.) with users' permission. Two key contributions of this model are (i) a methodology to extract the thematic fields users talk about with their social media contacts and (ii) a measure of the strength of the tie between two individuals from their interaction data available in social media sites. For the former, we use several data mining techniques to represent users' interests or "social contexts" by means of tags of representative words and validate this proposal by using Twitter data. We also show how these social contexts could be used to improve an important marketing application, namely that of advertising recommendation. For the latter, and contrary to previous approaches, we take into account different interaction types and contexts, the time in which interactions occur, the people involved in them and the frequency of interactions with the rest of the user's contacts, finding that our measure assesses with high accuracy users' perceived strength of their social ties. We finally discuss how this model of social spheres may be exploited to improve a wide range of technological applications, from recommender systems to e-mail readers, and describe two of them in detail: an application that helps users gain attention in social media and other designed to find trustworthy users in these media. We also present a prototype of an intermediary service that obtains these social spheres and makes them available to other services.
