A fuzzy model to enhance user profiles in microblogging sites using deep relations

Francisco, Manuel; Castro, Juan L.

VL / 401 - BP / 133 - EP / 149
Social Networking Sites (SNS) have entailed a revolution for society. They have given a say to everyone regardless of their status and this has been translated into loads of data. The task of profiling users constitutes a way to learn from this data in order to show users only the content that is relevant to them. Several recommendation system techniques have been used to address this problem, being mainly based on what the user explicitly says about themselves, on what the user publishes in the SNS and on the similarities between users. However, in social media context, it is also possible to use relations between users. Considering basic relations like follower or followee to extract information from them may result in noise, since they do not imply that users share interest or even ideas. In this work, we present a fuzzy framework to enrich user profiles with complex properties in order to have an even better representation of them. We use basic relations defined by SNSs to complete the information available in user profiles with topics of interest and ideas towards them and to define deep relations that will enable new ways of analysis. We use these deep relations to create clusters of similar users that, ultimately, will allow the expansion of properties from known users to the rest of the cluster. We tested our proposal with a dataset of Tweets in Spanish related to a political event. Our experiments prove the potential that this approach has for a lot of applications in microblogging context. (C) 2020 Elsevier B.V. All rights reserved.

Access level