Calculation of sentence vector similarity based on fasttext model of weighted fusion

Traditional text similarity algorithms suffer from inadequate feature extraction, and most of them do not calculate text similarity from features in both sentence and word dimensions. To solve these problems, a sentence vector similarity calculation method based on fasttext model of weighted fusion is proposed. The method generates word vectors by fasttext model, obtains sentence vectors consisting of word vectors weighted by word frequency and word order, gets sentence-level similarity by calculating cosine similarity, and gains word-level similarity by W-WRD algorithm. The final similarity result is obtained by this two levels of sentence vector similarity. Experimental results show that the model still has high accuracy on small data sets, outperforms the model before fusion, performs well on multiple data sets, and has good transferability and stability.