关键词: 文本相似度, Doc2vec, 双向长短期记忆网络

Abstract: In view of the inefficiency of manual approval texts in current government systems, this paper introduces text similarity into e-government. In the current network model based on text similarity, there is a huge matrix of generated word vectors, which requires a lot of time to train, and only uses the context of the context to generate word vectors, ignoring the relationship between the word order and semantics of the document. In order to improve efficiency and reduce training cost, this paper proposes a Do-Bi-LSTM text similarity calculation method, which first converts the text in the training data set into a vector through the Doc2vec language model. This method adds a text vector on the basis of the word vector, so can capture the interrelationship between sentences and between paragraphs. Then the obtained vector is trained as the input of the Bi-LSTM network model. Finally, compared with the LSTM network model and the traditional deep network model, the experiment shows that the accuracy of the method is greatly improved and feasible.

Key words: text similarity, Doc2vec, bi-directional long short-term memory