Problem generation method and device

Abstract

The invention discloses a problem generation method and device, and the method comprises the steps: taking an original problem sample and a target problem sample as training samples, and carrying outthe inverse reinforcement learning training of a discriminator composed of a machine reading model and a scoring function, wherein the target problem sample is a semantic similarity problem generatedby inputting the original problem sample into a problem generation model; and taking the overlap ratio score output by the scoring function as a return; taking the original problem sample as a training sample, and carrying out reinforcement learning training on a generator formed by the problem generation model, so as to further form adversarial training between the generator and the discriminator and taking the target problem sample as an adversarial sample corresponding to the original problem sample, wherein the problem generation model after adversarial training can be used for generatingtarget problems with similar semantics.

Description

technical field [0001] This description relates to the field of computer applications, in particular to a method and device for generating questions. Background technique [0002] In the field of machine question answering, question-answer pairs can be used as training samples to train a question answering system that can perform machine question answering; in order to improve the robustness of the question answering system, people usually need to pre-test the question answering system to be used before the question answering system is trained. Training samples are used for data enhancement; specifically, other questions that have similar semantics to the original question but have different expressions can be generated by certain means, and then new question-answer pairs can be generated; for example, the original "earth to the moon The distance is -about 380,000 kilometers"-the answer is right, then a new question similar to this question "how far is the earth to the moon"...

AI Technical Summary

Problems solved by technology

[0005] However, under the traditional implementation method, the training process of the above-mentioned problem generation model often cannot take into account the effect and cost; Directly mark the questions generated in the training phase and perform supervised learning training, which needs to consume more human resources

[0006] Moreover, the optimization and calling scenarios of the above-mentioned question generation models are often different from the actual application scenarios of the question answering system, which may lead to problems that the generated questions are not suitable for the final application scenarios

Images