Research And Implementation Of Network Representation Learning Algorithm Based On Time Sequence Information

The network structure can help understand the connections between different entities in real life.Exploring the network structure can provide in-depth understanding of network information.For example,link prediction,community discovery,node classification,etc.,all benefit from the development of network analysis technology.However,because the network structure is irregular and contains a large number of nodes and connections,it will be very complicated to process nodes directly on the network.And subsequent machine learning tasks cannot directly use the network structure data as input,so how to express the network structure is a problem worth studying.Network representation learning is to study the conversion of network structure data into low-dimensional real-valued vectors on the basis of preserving the original network structure information as much as possible,and the obtained vectors can be directly input into subsequent machine learning tasks.Most existing network representation learning methods usually focus on static network structure,in which nodes and edges are fixed and will not change over time,and do not consider the dynamic changes of the network.But most of the networks in the real world are dynamic networks,that is,the networks will change over time.This information that changes according to time is an important part of the network and reflects the evolution of the network over time.Aiming at the above problems,this paper aims at the time information of establishing connections between nodes in dynamic networks,learns the evolution process of the network,so as to learn the lowdimensional vector representation of nodes,which is used to improve the prediction accuracy of network evolution.This article first introduces the current research status of network representation learning and analyzes the problems of existing methods,then elaborates the dynamic network representation learning method based on time series information proposed in this paper.At the same time,the design and implementation of the method in this paper are given,and the experimental results prove the accuracy and rationality of the dynamic network representation learning algorithm proposed in this paper.Finally,the conclusion and future work of this article are given.The main work of this paper is as follows:(1)This article first proposes a random walk process based on time sequence information,that is,adding time information to the random walk process to respect the dynamic evolution of the network over time,that is,the process of connecting edges must conform to the current time sequence.At the same time,considering that historical neighbors will affect the selection of current neighbors,that is,neighbors that have established connections in the early stage will also establish connections with a certain probability in the next moment,and the impact will gradually decrease with the increase of time,that is,neighbors with frequent contacts in the next There will be a high probability of establishing a connection at any time,so the Hawks model is introduced to simulate its influence changes.(2)In targeted optimization,this paper proposes two node optimization strategies based on timing information to dynamically adjust the number and length of node generation sequences.By controlling the stopping probability of each random walk,the purpose of dynamically adjusting the length of the node sequence is achieved.At the same time,the number of generated node sequences can be dynamically adjusted by distinguishing early edges and recent edges.(3)During the experiment,this paper uses three real data sets: Academic data set,College Msg data set and friends data set to verify the rationality of the proposed method.It can be seen from the experimental results that the dynamic network representation learning method based on time series information proposed in this paper has achieved better results in tasks such as link prediction and node classification.At the same time,the parameter sensitivity problem was analyzed,and the reasons for the different experimental results caused by different values of the parameters in the experimental process were discussed,and its rationality was verified.