Research And Implementation Of Semantic Filtering Technology Of Public Opinion Text Based On Knowledge Graph

Online public opinion is published,reposted and disseminated in the form of text in cyberspace,which leading to frequent and frequent online public opinion problems,posing a challenge to online content governance.In practical public opinion governance work,it is necessary to filter out relevant information of specific topic content from large-scale data,while the current content filtering technology mainly adopts algorithm based on keyword matching,which lacks semantic understanding and has a gap with the natural demand expression of public opinion content governors,resulting in low accuracy,high noise and information redundancy of filtering results,requiring a lot of time and manpower to eliminate meaningless information,leading to low efficiency of public opinion content governance.Based on knowledge graph,public opinion text semantic filtering technology can perform deep semantic understanding and analysis of text content,effectively improve the accuracy of filtering results,and has important significance for building a clear cyberspace.Based on the problem that the mainstream public opinion text filtering technology lacks semantic understanding,resulting in low filtering accuracy,a characteristic classification public opinion knowledge graph based on causal relationship in the education field is designed and implemented.Based on the analysis of the characteristics of online public opinion text in the education field,such as multiple sources,diverse structures and complex semantics,a classification system that conforms to the actual situation of public opinion work is designed,and the construction is completed by a entity recognition and relation extraction technology,supplemented by a small amount of manual annotation.The graph contains noun entity knowledge graph and event behavior causal graph,which provides domain semantic association data support for semantic filtering.Aiming at the problem of logical semantic association loss and high vector dimension when knowledge graph is vectorize,a low dimensional and high fidelity hyperbolic contrastive embedding knowledge graph representation model is designed and implemented.The model uses hyperbolic embedding representation method to learn specific relation curvature to preserve the hierarchy of knowledge graph;relies on parameterized distance preservation operation to capture logical relation;introduces contrastive learning to drive model training.Experiments show that the model can use 32-dimensional vectors to represent the domain knowledge graph,and achieve the best performance on MRR,H@10 and H@3 indicators.Aiming at the problem of insufficient semantic depth and inconsistent semantic feature matching degree between long and short texts in public opinion text filtering,a knowledge graph semantic matching filtering model based on entity linking is designed and implemented.The model uses entity linking technology to represent the text to be filtered as an entity list and extracts the semantic association subnetwork of the query statement,and performs semantic matching calculation and sorting by using the late interaction paradigm ranking model,and sets a threshold that conforms to the domain characteristics to complete semantic filtering.Experiments show that our modal is significantly improved compared with the baseline model.To verify the effect and feasibility of the model designed in this paper,a semantic filtering system for public opinion text based on knowledge graph is designed and implemented.The system can complete the semantic filtering task from end to end,and can meet the work needs of public opinion governance personnel after functional testing.