The ability to understand natural language text is far from being emulated in machines. One of the main hurdles to overcome is that computers lack both the common and common-sense knowledge humans normally acquire during the formative years of their lives. In order to really understand natural language, a machine should be able to grasp such kind of knowledge, rather than merely relying on the valence of keywords and word co-occurrence frequencies. In this work, the largest existing taxonomy of common knowledge is blended with a natural-language-based semantic network of common-sense knowledge, and multi-dimensional scaling is applied on the resulting knowledge base for open-domain opinion mining and sentiment analysis.