Citation-context Based Academic Literature Summarization Method

With the continuous increase of scientific research output,it is difficult for academic researchers to retrieve and access to the explosive growth of academic literature,the workload of scientific research is also increasing.That academic literature automatic summarization method obtains information efficiently and accurately from the vast literature by means of modern computer technology has become the research hotspot.The dissertation takes citation context as the research object,aiming at the shortcomings of the current citation-based summarization of academic literature,and we propose the summarization method of academic literature based on citation context,which improves the quality of summarization of academic literature.The main research contents and achievements of the dissertation are as follows:(1)The classification algorithm of citation context based on convolution neural network is designed.Firstly,the structural characteristics of the academic literature are analyzed,and the classification model based on the academic literature structure is proposed to solve the problem of incomplete coverage of the summarization information.In order to locate the discouse facet of citation context in the cited article,simulating the application of convolutional neural network in the image field,designing a simple neural network model and the word vector based on deep learning is used as the input of the sentence to realize the citation context classification based on convolution neural network.The CNN-static and CNN-non-static modes are compared with the traditional SVM-based classification algorithm,the experimental results indicate that the citation context classification based on convolution neural network achieves high accuracy in both modes,which CNN-non-static accuracy is up to 79.91% and the overall average increased by 3.12%,it can solve effectively the discourse fact of citation context of the distribution of the problem.The results of the evaluation also confirm that the method improves the information and readability of the summarization.(2)The algorithm of citation context extraction based on vector space model is constructed.The relationship between citation and citation context is analyzed,the feature selection and weight calculation of citation sentences are carried out,and the vector space model of citation and cited article is constructed,the semantic relationship between citation and citation context is calculated by using cosine distance,realizing that the use of citation to extract the citation context from the cited literature.The results of the extraction show that the consistency of citation and citation context is low,the extraction accuracy is low and the overall average is 17.66%.It also shows that the citation does not accurately reflect the cited article,and the summarization of the academic literature based on citation is inconsistent with the cited article information.(3)The improved algorithm for sentence ranking based on graph is proposed.The traditional graph-based ranking considers only the importance of the sentence,and does not take into account the redundancy between the sentences,resulting in the generation of summarization information redundancy problem.Using the semantic position relation of words between sentences and combining the semantic similarity relation between sentences to evaluate the redundancy of sentences,the importance and redundancy of weighted sentences are sorted by their comprehensive scores,which solves the problem of redundancy of summarization sentence information.The outcome of the summarization evalution shows that the Rouge values of the summarization are improved by this method,the quality of the summarization is also improved,making it closer to the gold summary.