Polysemy is a very common phenomenon in modern languages. Most of previous work treat the multiple meanings of a word equally. However, under many circumstances, there exists a primal meaning for the word. Many of the new appearing meanings of the words are either originated from a primal meaning, or are merely literal references to the original word, e.g., apple (fruit), Apple (Inc), and Apple (movie). When constructing knowledge base from on-line encyclopedia data, it would be more efficient to be aware of the information about the importance of the senses. In this paper, we would like to explore a way to automatically recommend the primal meaning of a word based on the textual descriptions of the multiple meanings of a word from online encyclopedia websites. We propose a hybrid model that captures both the pattern of the description and the relationship between different descriptions with both distantly supervised and unsupervised models. The experiment results show that our method yields very good result with a p@1 score of 80.6%, and a MAP of 89.0%, surpassing the Word-Sum baseline by a big margin (P@1 is 62.7% and MAP is 76.8%).