Unstructured P2P social search query routing algorithms for agentified social networks
- Autores: Albert Trias Mansilla
- Directores de la Tesis: Josep Lluis de la Rosa Esteva (dir. tes.)
- Lectura: En la Universitat de Girona ( España ) en 2013
- Idioma: español
- Tribunal Calificador de la Tesis: Boleslaw Karl Szymanski (presid.), Jordi Sabater Mir (secret.), Borisw Galitsky (voc.)
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
In 2010 Horowitz and Kamvar explained that there are mainly two paradigms for finding relevant information. The village paradigm is to ask someone who might satisfy an information need, while the library paradigm is based on accessing a document that might accomplish that goal.
Social search has become more important with the advent of online social networks. Social search is based on the capacity to use information that is provided by people to satisfy the information needs of other people instead of using data selected by authorities and indexed in a catalogue. Social search includes the village paradigm, in which people satisfy the information needs instead of the indexed documents of the library paradigm. Out of these two paradigms, only the library paradigm has been automated online; with social search, the first steps to automate the village paradigm have been made.
In fact, the information search process has been automated from the library paradigm perspective only, while the older paradigm, the village, has been almost not automated at all.
Nevertheless, the village paradigm presents some benefits compared with the library paradigm; people can adapt the answer content as a function of who is requesting the information, or furthermore, people can perform clarifications of the answer content; at the same time, the explanations in documents remain static and are the same for all of the readers, and no clarifications can be performed.
The proliferation of online social networks along with the advances in artificial intelligence allow us to consider village paradigm automation. Agent technology is the design metaphor that I have chosen to automate the village paradigm.
In this thesis, I analyse the village paradigm, seeking the aspects that might be automated; among these aspects, I worked out the selection of candidates in unstructured peer-to-peer (P2P) social networks because the social networks are inherently P2P. We consider unstructured architectures because social networks are bottom-up, while structured architectures usually use methods to organise nodes and methods, such as distributed hash tables (DHT), to search content. Moreover, in the literature, there are approximations of unstructured P2P social networks that are applied in web searches, as is the case for 6Search.
Although P2P architectures are considered to be more scalable, in the case of unstructured P2P architectures, the query-routing methods based on Breadth First Search (BFS), such as the case of 6Search, generate large quantities of messages that threaten their scalability in the domain of social webs.
In this thesis, I propose a social search protocol that uses stop messages to improve scalability; to be effective, stop messages require the addition of delays in the propagation of the question messages. Next, I study the effect of social network topologies on the proposed protocol; with this aim, we used 21 social network topologies. We used Movielens data in our simulations. Then, I propose a search algorithm that reduces the number of acquaintances (or contacts) that the question message is sent to. Deciding how many acquaintances to whom the question will be sent is not straightforward; selecting only a few has the risk of the question being dropped, ignored or simply forgotten, and no answer might be received, while selecting too many threatens the scalability. Question Waves (QW), the social search algorithm proposed, contributes to relaxing this decision. The idea of QW comprises delaying the question messages that are addressed to less trusted acquaintances compared to those who are more trusted; in this way, acquaintances with lower trust will be asked later, and more trusted acquaintances will be asked without a delay or with a small delay. With this algorithm, the expectation of obtaining a relevant answer from the requested candidates decays with time, so the algorithm proceeds to send another question message to additional acquaintances after a prudent time, which is indicated by the delay explained before. As a result, not only is scalability improved, but the answers also come sorted by the probability of being relevant; in this way, the algorithm contributes to obtaining more relevant answers in the first answers that are received. The answers are automatically sorted (ranked), and as a consequence, the QW algorithm obtains higher precision in a predefined-size window of answers.
I simulated the QW algorithm over two datasets. The first dataset corresponds to knowledge exchanges (KE) that we generated randomly, and the second dataset corresponds to opinion exchanges that were built with data from Movielens, a dataset for recommender systems.
Although the work that we conducted is promising, yet not far from applicability, it is far from a complete solution of village paradigm automation. We have opened a new line of research within our research group, which is that we will continue to study, with more emphasis on the applicability, improving the agents¿ behaviour and the QW algorithm and studying the environment properties.
