Research On Question Answering System For Freshmen Register In Colleges And Universities Oriented To Knowledge Graph And Semantic Recognition

The content of university orientation work comes from different departments,and the information channels are also diverse.For example,there are not only enrollment process,student loan policy,academic education,life guide and other documents,but also some specific life guide and school norms.They can also come from campus network,Internet,e-mail,wechat,instant messaging,forums and other crowdfunding Q&A information Information sources are diverse,fragmentary and discrete,and some of them still have some problems,such as poor professionalism,low accuracy,scattered data,large redundancy and noise.Therefore,traditional question answering systems often use keyword oriented model or simple depth model,which despises domain knowledge and semantic recognition technology,resulting in poor experience of question answering system.Therefore,this paper introduces the knowledge graph technology of colleges and universities to enhance the reliability in the process of Q&A.at the same time,compared with the open domain Q&A system,the sample data of the university orientation Q&A system is less and the labeling cost is high.In the case of small sample data,the depth model is faced with the problems of difficult training and poor effect.This paper introduces the deep learning method,The language model is pre trained,and the training speed is accelerated and the effect is good.To sum up,based on the knowledge graph of colleges and universities,combined with deep learning methods and natural language processing means,the research content of the paper is completed from the following aspects:A named entity recognition model based on dual flow self attention multi direction graph is constructed.Aiming at the problem that traditional named entity recognition methods can not take into account both semantic disambiguation and effective extraction of text sequence features in Chinese environment,a named entity recognition model based on dual flow self attention multi direction graph is proposed.Firstly,the model is pre trained on the general corpus data set with rich samples,and then refined on the freshmen's Q&A data set with few samples through model migration and parameter sharing.In the text embedding stage,the corpus data is formed into graph form to further enrich and optimize the related entity dictionary.Then,the improved graph neural network is used to embed the corpus data combined with the designed entity dictionary,so as to effectively reduce the error of Chinese word segmentation.In the feature extraction stage,a dual flow self attention mechanism based on dynamic weighted fusion is proposed,which uses the feature contribution value to enhance the semantic recognition ability of the model.In the model output stage,the extracted features are input into the standard classifier to predict the global optimal results.Through the above work,we can effectively and accurately extract the entities with strong relevance to the university orientation business in the user questions,and pave the way for the next work.A Chinese short text classification model is proposed,which combines the multi head self attention mechanism.In view of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the problem of the Chinese short text knowledge utilization in the question answering process and the semantic ambiguity and sparse characteristics of the short text itself,this paper constructs a Chinese short text classification model which combines the multi head self attention mechanism: firstly,by using the pre training language model,the dynamic text representation method is used to improve the semantic recognition ability of the language model;then,through the Finally,to improve the ability of the model to extract the local features of the text of Q&A sentences,a Chinese text short text classification model can be obtained by increasing convolutional neural network,which can extract features at different granularity levels,so as to obtain higher-order representation with different semantics,thus enhancing the question and answer The ability to classify Chinese short text text question sentences in application scenarios.Develop the new question and answer system in Colleges and universities.Firstly,the system structure is provided by using software engineering technology;secondly,the knowledge graph of the new business in Colleges and universities is constructed;then,the above research results are selected with the mature technologies and products such as Yanshen and other mature technologies and products through the in-depth learning framework with the real robot which can conduct human-computer dialogue.The system integration method is used to integrate the software to develop the online Q&A system to complete the paper;finally,the Department The system is testing,and the visualization technology is applied,and the test display system has achieved a good application effect.