Font Size: a A A

Research And Implement For Question Answering Based On Deep Learning And Knowledge Graph Embedding

Posted on:2019-05-14Degree:MasterType:Thesis
Country:ChinaCandidate:W M ZhangFull Text:PDF
GTID:2348330545981066Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
With the development of Internet technology,the network information becomes more and more inclusive and huge.Faced with such a huge data,how to accurately find useful information has become an important task for the development of the internet.The emergence of question answering system satisfies the need of users to quickly obtain the target information.The question answering system enable to provide the target answer directly without too many choices.The goal of question answering system is to quickly return simple and accurate answers according to user input problems.Visual question answering system(VQA)is a form of question answering system.It can answer a question raised by users by according to the content of a reference image.At present,the visual question answering system is regarded as a classification task.The commonly method is to extract image feature and question feature according to the provided image and question,and then merge the image feature and question feature to do classify.But the accuracy and model performance need to be improved.In addition,nowadays visual question answering system is hard to handle questions required common sense.To solve these problems,we propose a visual question answering system based on two-step joint attention network and a visual question answering system based on knowledge graph embedding.Visual question answering system based on the two-step joint attention network introduced two-step joint attention model to VQA.Two-step joint attention model visual model enable to focus the image and question from coarse-drained parts to fine-grained parts while ignoring the interference information in the image and question to achieve accuracy improvement by adopting CNN and LSTM respectively.The visual question answering system based on knowledge graph embedding introduced knowledge base into VQA and innovatively proposed an idea that obtain entity representation from image by using CNN.,which achieve effective experimental results for questions that require common sense.We designed and implemented the mobile terminal APP to display proposed visual question answering systems,which allows people to intuitively experience the system performance.
Keywords/Search Tags:visual question answering, attention, knowledge graph embedding, common sense
PDF Full Text Request
Related items