An Implementation of Symbolic Aboutness Theory

Today information can be globally shared via the Internet and can be accessible from anywhere in the world. The increasing complexity and size of the WWW urges the need of more effective mode for information processing techniques such as information retrieval and filtering, information summarization, topic segmentation, data mining and information discovery, etc. All of them can be fundamentally considered as informational inference processes. For example, the task of information retrieval is to determine whether a document is about (i.e., relevant to) a user’s information request as completely and as precisely as possible. Therefore, aboutness plays a crucial role in the modern information processing procedures. Most current investigations in the information processing areas, however, are heuristically driven and lack of more formal considerations based on a better understanding of aboutness. We argue that an investigation and formal modelling of aboutness informational inference would be greatly helpful to further improve the performance of information process systems by better making use of the deep semantics involved in these processes. In our previous study [Bruza et al 2000], we proposed a commonsense aboutness theory and a set of properties (in form of symbolic inference rules) of aboutness acceptable from a human reasoning perspective. In this report, we give a brief description of the symbolic aboutness theory, and introduce an implementation of the proposed symbolic aboutness theory – “Penguin” Aboutness Inference System. The experimental results show that the proposed aboutness rules are reasonable, and at the same time, provide us a deeper understanding of the advantages and disadvantages of symbolic approach. Moving down to a vector based conceptual level, which is underneath the symbolic level, would provide a solution to the major disadvantages of symbolic approachescomputational complexity and the frame problem.