A Non-Training Approach to Generating High Semantic Titles for Chinese Documents

Due to the abundance of available data, manual title generation may become unfeasible. Traditionally, information retrieval has been applied to perform automatic title generation by exploring and searching for keywords from a document. However, titles generated through such a direct combination approach may not satisfy Chinese grammatical rules, and also may not express the semantic meaning explicitly. Thus, we propose a non-learning approach based on conceptual schema to generating titles automatically through sentence modification and recombination. Experimental results prove that our model can satisfy the automatic title generation requirements.