A Text Similarity Calculation Method with Feature Weighted Fusion

Text similarity calculation is an important component of many text mining techniques. Most of the existing studies only consider model fusion of textual features or semantic features, which are not suitable for long text processing scenarios due to the time consumption of topic extraction for words of the whole document. This article introduces related work of text similarity calculation, including TF-IDF algorithm, lexical features and LDA(Latent Dirichlet Allocation) model. Then a text similarity calculation method that combines text word features and text implicit topic features is proposed, making full use of the surface and hidden information of text to calculate text similarity. The proposed method first improves traditional word mover's distance (WMD) algorithm by introducing lexical features of text, and then calculates text similarity by linear weighting of the improved WMD algorithm and LDA model algorithm. Experimental results indicate that compared with traditional WMD algorithm and LDA model algorithm, the proposed method obtains better text similarity calculation results.