Method and apparatus for determining semantic matching degree

Abstract

The invention provides a method and an apparatus for determining a semantic matching degree. The method comprises the steps of obtaining a first statement and a second statement; dividing the first statement and the second statement into x and y statement segments respectively; performing convolution operation on a word vector in each statement segment of the first statement and a word vector in each statement segment of the second statement to obtain a three-dimensional tensor; combining and / or screening adjacent vectors in one-dimensional vectors in x rows and y columns until the three-dimensional tensors are combined into a one-dimensional target vector; and according to the target vector, determining the semantic matching degree of the first statement and the second statement. According to embodiments of the invention, the to-be-matched statements are divided into the statement segments taking the word vectors as units according to a word order; the statement segments in the two statements are subjected to the convolution operation in pairs to obtain the three-dimensional tensors, wherein each three-dimensional tensor comprises local matching information of the statement segments of each statement; and a semantic matching result obtained based on the three-dimensional tensors is more accurate.

Description

technical field [0001] The present invention relates to the field of semantic matching, and more specifically, to a method and device for determining the degree of semantic matching. Background technique [0002] Semantic matching technology has become one of the core technologies of natural language processing, and has played an important role in many commercial systems, for example, voice assistants (such as Siri, Google Now), machine translation (Machine Translation), and chat robots (such as Microsoft Xiaobing) and so on. [0003] In the prior art, each sentence to be matched is usually decomposed into multiple word vectors, and each word vector has a fixed dimension. On this basis, the sentence to be matched is represented by the linear superposition of the word vectors it contains. The degree of semantic matching of two sentences is described as the inner product between the vectors representing the two sentences. [0004] Although the above semantic matching method...

AI Technical Summary

Problems solved by technology

This method of overall statement matching expresses all the information of the statement through a vector, often ignoring the impact of the matching degree of local segments between sentences on the final semantic matching result, resulting in inaccurate matching results

Images