Question and answer matching method, system and storage medium

A question answer matching system for providing a matching answer to a user query based on a convolutional neural network, the system comprising one or more processors and one or more non-transitory computer-readable storage media storing instructions executable by the one or more processors to cause the system to perform operations comprising:

• obtaining, for a historical query, a positive sample comprising a best answer corresponding to the historical query, a negative sample comprising a randomly sampled answer not corresponding to the historical query, and another negative sample comprising an answer with literal similarities to the historical query, to construct a pre-constructed dataset corresponding to the historical query;
• training the convolutional neural network using the pre-constructed dataset as training data, wherein the convolutional neural network comprises one or more word embedding layers sharing a pre-trained word vector and one or more convolutional layers;
• transforming the user query and one of one or more suggested answers corresponding to the user query by using the pre-trained word vector to obtain vector representations of the user query and the one of one or more suggested answers corresponding to the user query;
• performing a convolutional operation through the one or more convolutional layers on the vector representations of the user query and the one of one or more suggested answers, respectively, to extract features;
• and
• mapping convolution results of the vector representations of the user query and the one of one or more suggested answers into a sample annotating space, to obtain a matching result of the user query
• wherein the convolutional operation on the vector representation of the user query does not share a configuration parameter with the convolutional operation on the vector representation of the one or more suggested answers.

A non-transitory computer-readable storage medium for providing a matching answer to a user query based on a convolutional neural network, configured with instructions executable by one or more processors to cause the one or more processors to perform operations comprising:

• obtaining, for a historical query, a positive sample comprising a best answer corresponding to the historical query, a negative sample comprising a randomly sampled answer not corresponding to the historical query, and another negative sample comprising an answer with literal similarities to the historical query, to construct a pre-constructed dataset corresponding to the historical query;
• training the convolutional neural network using the pre-constructed dataset as training data, wherein the convolutional neural network comprises one or more word embedding layers sharing a pre-trained word vector and one or more convolutional layers;
• transforming the user query and one of one or more suggested answers corresponding to the user query by using the pre-trained word vector to obtain vector representations of the user query and the one of one or more suggested answers corresponding to the user query;
• performing a convolutional operation through the one or more convolutional layers on the vector representations of the user query and the one of one or more suggested answers, respectively, to extract features;
• and
• mapping convolution results of the vector representations of the user query and the one of one or more suggested answers into a sample annotating space, to obtain a matching result of the user query;
• wherein the convolutional neural network comprises a first word embedding layer corresponding to the user query, and a second word embedding layer corresponding to the one of one or more suggested answers corresponding to the user query.