Research On QA Answer Selection Algorithm Based On Deep Learning

Answer selection is an important part of intelligent question answering.Given a problem and the corresponding pool of candidate answers,answer selection task aims to find the correct answers from the pool of candidate answers.The core content is how to construct the matching relationship between the question and the candidate answer,so as to sort the candidate answer according to the matching relation,and the answer selection task can also be regarded as a kind of sentence matching task.This paper mainly studies the application of deep learning in answer selection task,and proves the effectiveness of models through experiments.The main research contents of this paper include the following aspects:1.We study the vector representation based answer selection model,and compare the effectiveness of CNN and biLSTM on learning sentence meaning representations.We verify the effectiveness of the combination of the two models.The sentence vector representation learned from the combination of the two models can contain two kinds of semantic information learned from neural network.2.We study the matching feature learning based answer selection model,and focus on two aspects:alignment feature learning methods and aggregation methods.Then build the best matching feature learning model and compare it with the existing work.3.We combine the deep model with the shallow features,and do experiments on the Chinese answer selection data set.It is proved that the shallow features can play a complementary role to the deep model.4.We apply the answer selection model to short text matching task,thus prove the versatility of the answer selection model on other sentence matching tasks.