The Design And Implementation Of Retrieval System Of Similar Software Project Based On Deep Learning

With the rapid development of internet technology,open source has become a development trend of the software industry.Many software developers contribute a lot of software resources and technical experience to the open source community.Although the rich software resources provide convenience for users,due to its large number,rapid update,and large correlation between different resources,coupled with the existing retrieval system,it is impossible to intelligently analyze users' input intentions,resulting in rigid search results.It is still difficult to quickly retrieve similar projects.Based on the deep learning algorithm,this thesis proposes the design and implementation of a retrieval system of similar software project.The system uses SpringMVC+Spring+MyBatis(SSM)framework,uses Word2vec to preprocess data,and neural network models such as long-term and short-term memory network,convolutional neural network and Doc2vec model to deeply analyze the user's requirements,solving the problem that the search results are not smart enough and accurate to help users to find the projects they want more efficiently and accurately.This thesis mainly describes the design and implementation of retrieval system of similar software project based on neural network model.System's function modules are divided into registration-login module,search module,search list display module and project detail module.Firstly,the project background of the system is summarized,and then the research status of information retrieval system and deep learning algorithm are analyzed.Then the principle of the architecture technology SSM framework and deep learning algorithm used by the system is introduced.Then introduce the requirements design,architecture design,module design and database design of the system.Then describes the designs,models,implementation and evaluation of LSTM,CNN and Doc2vec in detail,the design and implementation of each module are as also.Finally,gives the testing results of the system.The core of the three algorithms in this thesis is to transform the vectorized user input into a probability distribution vector based on the project category using the neural network classification model,and cluster based on the classification result.These vectors form the software project type space.It is beneficial to use the distance formula to calculate the vector similarity to get the similarity of the software project.The similar software project retrieval system implemented in this thesis analyzes users'input through deep learning algorithm It enhances the understanding of user's query intention,and also compensates for the lack of user's ability to express demand to some extent.The user can freely choose the retrieval algorithm and give his own satisfaction to the retrieval result,so that an equal and mutual feedback mechanism is established between the system and the user.