Multi-role identification of sentences using Relevance Vector Space

This paper presents a solution to classifying sentences with multi-labels. This problem is an essential part to a semantic search process. Sentences or keywords with correctly automated labelling can enhance the efficiency and performance of the search. The technique introduces a vector space of relevance for keywords and sentences with necessary operations. Concepts and motivation are explained with mathematical models for the two vectors and their operations. Experiments of multi-role identification of sentences are conducted with abstracts of scientific articles. Procedures of executing models in the experiment are explained step by step. In comparison, nine other techniques of multi-label classification are used on the same data set. The evaluation is done on 4-fold cross validation basis. The study reveals a successful result of multi-role identification of sentences with higher accuracy than other nine techniques.