Chinese long text similarity calculation
What is it about?
The neural network model has achieved good results in the similarity calculation task of sentences or short texts. However, the effect of existing similarity algorithms on long texts is not ideal, and they cannot truly extract richer semantic information hidden in the structure of long text documents. This article aims to build a learning model that can more accurately express the semantics of long texts and solve the bottleneck of calculating long text similarity.
Why is it important?
This article ingeniously integrates the characteristics of Chinese long texts in terms of grammatical structure into the Bert model, proposing a semantic progressive fusion model from word → sentence → text, which maximizes the preservation of the true semantics of long texts and improves the accuracy of long text similarity calculation.