Scholar Profiling Mining Algorithm Based On Generative Pre Training Model

Scholar profiling construction has long been considered a very important and challenging task in the field of information mining and social analysis.In the face of the data characteristics of the Internet era,such as large amounts of data,high data noise,and large amounts of redundant data,traditional information mining methods based on knowledge engineering,conditional random fields,and machine learning have been unable to meet user needs.In response to these situations,this article is based on a generative pre training model and conducts deep mining of scholar experience information and basic information to construct a high-precision scholar profiling.Firstly,in order to solve the problems caused by the complex semantic relationships and entity alignment in existing research,this paper proposes a scholar experience mining algorithm that combines generative multi-turn question answering and contrastive learning.This algorithm is based on the GLM pre training model,changing the text mining task into generative multi-turn question and answer task,and capturing the characteristics of scholar experience information through the Masked Multi-Head Attention mechanism.It uses positive and negative sample similarity distances to calculate contrastive learning losses,enhancing the feature learning ability of the model,and more accurately and efficiently mining scholar experience text information.Secondly,in order to address the impact of ignoring the addition of prompt tags and neighborhood features to pre training models in existing research on scholar basic information mining tasks,this paper proposes a scholar information mining algorithm combining prompt features and neighbor features.This algorithm adds identification prompt to scholars’ basic information,and considers the web address and text content features of its left and right neighborhood nodes in the coding stage,integrating target node features and neighborhood node features,further improving the model’s feature learning ability.Finally,this article conducts experiments on scholar experience mining and scholar basic information mining in the public dataset AMiner-data,and conducts a large number of comparative experiments with other basic methods.It conducts in-depth analysis of key modules and important parameters,which proves the practicality and effectiveness of this method for mining high-precision scholar portrait tasks in various aspects.