Logic Based Models for Information Retrieval in Social Media Contexts
The massive volume of information available in social media nowadays has created an urgent need for information retrieval models to assist users in various retrieval tasks in this domain. One of the most prominent fields where users gain important information from social media is the process of evaluating products or services based on customers’ reviews. However, the usage of such data in information retrieval tasks raises many major issues. Firstly, social contributions contain information from a wide variety of users, therefore the credibility of the information becomes questionable. Secondly, the absence of information in social contributions does not necessarily mean "nothing", as in many cases, missing information would be implicitly meaningful. Finally, social contributions may contain contradictions that can confuse the users and limit the usefulness of the information. Approaches for modelling information retrieval based on social media contributions have been widely discussed, including probabilistic multi-valued logic-based models. The main strength of these models is their ability to address states like “unknown” and “inconsistent” for information matching besides the states of the traditional binary logic (true and false). Therefore, probabilistic multi-valued logic is well suited to model information retrieval in social contributions-based contexts. However, so far, they have not been utilised to model information retrieval in such environments. In this thesis, we investigate the utilisation of two types of multi-valued probabilistic logics for information retrieval tasks in a social contributions based environment. In the first part of this thesis, we investigated the utilisation of four-valued and subjective logics in the domain of hotel reviews in a system-oriented study. In the second part, we have conducted user studies, to test the effectiveness of a logical model as an algorithm to rank items in a laptop store. Our results have shown powerful abilities of the multi-valued logical models in ranking tasks.