Chinese Analogy Search Considering Multi Relations

There are some specific relations between the entities of the nature. The same relationship exists between two pairs of entities, simply to say, what A is to B as C is to D. For two unfamiliar entities (A, B), We cannot say precisely the relationship between them. When given an instance C similar to A, we cannot say out the entity D corresponding to B. Latent relation search provides a new way of searching D after given the question (A, B) and (C, ?). This paper proposed a Latent relation search method based on words frequency and weight of words. We explore all the relations between entity A and B by using K-means clustering method and then determine entity D corresponding to C in different relationship types. The proposed method achieves an MRR of 0.773.