Semantic feature extraction method based on Sensor-BERT and Word2Vec

The invention discloses a semantic feature extraction method based on Sension-BERT and Word2Vec, and belongs to the field of data communication of a data center, log data is cleaned by using a preprocessing method, confidential information in the log data is removed, a plurality of conjunctions in a log are divided into single words by using a word segmentation technology, and the conjunctions are extracted by using the single words. Performing template analysis by using a Drain analyzer to obtain a log template event; the log template event is input into a pre-trained Sension-BERT model and a Word2Vec method, semantic vectors obtained through Sension-BERT contain sentence-level semantic information and position information of words in sentences, detailed word vectors of the words in each sentence can be obtained through Word2Vec, semantic vectors of a second part are obtained through weighted average, and the semantic vectors of the second part are obtained through weighted average; fusing the two parts of s