SemRec : social semantic recommendations

Abstract

With the vast amount of data being consumed in social networks, ranking and recommending items to the user is becoming more and more of a challenge. This Final Year Project focuses on merging information from different social networks to build a unified user model, semantically analyse this information and in turn offer intelligent ranking according to the best interests of the user. First we consider each heterogeneous social network stream separately. We gather and enhance all the possible users data (like the user posts and interactions). To enrich the users data we use Alchemy API, one of the world's leading Natural Language Processing (NLP) tools. This tool is able to extract the topic and concepts (apart from other information) that are relevant to a corpus of text. We then build the user model by storing all the gathered information in RDF format by re-using existing ontologies where possible and extending with new properties where needed. The different feeds are then merged from the social networks into one and the ranking algorithm is applied based on the data mined in the previous stage. As for recommendations we focus on the re-ordering of the merged news feed(s) of the user. Finally the test and evaluate our work in the following ways: (i) we set up a small social network scenario to see how the algorithm performs on a controlled data set. (ii) We also test our application by giving it to real social network users and ask for their feedback via a survey, which contains both qualitative and quantitative questions. (iii) The users were also given the option to rank each post's ranking given by our algorithm, allowing us to collect quantitative feedback regards the individual post's sorting. The overall feedback obtained was positive, with the majority of the people stating that the intelligent ranks given were interesting and that they think topics should in fact be considered when dealing with similar algorithms.