FIGURE 2.
The architecture for extracting ncRNA structure feature through NLP method with local fusion strategy. As (B) shown, each ncRNA from database is divided into many triple symbols by 3-mers composition, and GloVe is used to generate embedding vector of 43 symbols. Then, as (A) shown, each ncRNA will be split into some consecutive subsequences with no overlap. All the triple symbols embedding vector of each subsequence can be obtained from (B). Finally, the representation of ncRNA can be obtained through calculating the average of all symbol vectors in each subsequence respectively, and concatenating all average vectors.