The authors overview an efficient and simple representation model of sentence semantics in the context of a paraphrase identification problem. A dependency tree is chosen as the main structure to represent connections between words in a sentence. To represent word semantics, pre-trained word representation models are used. Based on these two key components, several features helping to precisely identify paraphrases are designed. The conducted experiments proved this model to be efficient. The model application results are quite close to those of state-of-the-art models.
展开▼