The invention relates to an LSTM-based financial news tendency analysis method. The method comprises the steps of performing company name identification based on Baidu encyclopedia query and company name and company code mapping; C; comparing the similarity between sentences and titles by using a doc2vec model, and extracting key sentence groups by comprehensively considering sentence positions, domain verbs and company name information; A; and using word2vec and TFIDF to represent sentences, and using an LSTM model to classify the key sentence group. According to the invention, Baidu encyclopedia query is added as a factor of identification in the company name identification method; t; the effect is better; t; the expansibility is better; t; the problem that the names of non-companies aremisjudged due to too many products is solved; A; according to the method, t, the key sentence group is extracted and introduced into the doc2vec model, t, the similarity calculation accuracy is improved, when sentences are represented, t, the Word2vec is used for training the text, meanwhile, t, the TFIDF method is introduced, t, text context information and the importance degree of words in thetext are fully utilized, and a very good effect can be achieved.