Research On Keyword Extraction Method Based On Graph Structure

As one of the most important research branches in the field of natural language processing,keyword extraction technology has been applied in various fields and it is the core technology in text classification and content retrieval.With a large amount of text knowledge,keyword extraction technology can classify documents more precisely according to the different focus directions of extracting keywords,or use the extracted keywords as search tags of documents,which can greatly improve the efficiency of knowledge retrieval.Keyword extraction technology has also been used in various aspects such as search engine,intelligent question and answer,text generation and sentiment computing.At present,keyword extraction algorithms are mainly used to extract keywords from small-scale and short text corpus and are more difficult to perform keyword discovery on large-scale data.With the excellent performance of deep learning in various fields,it has become possible to use deep learning methods to achieve keyword extraction.However,most of the existing deep learning-based keyword extraction algorithms are to obtain the semantic feature representation of words through the deep learning model first,and then use traditional methods to process the features to extract keywords,but there are few methods to directly use deep learning models to obtain text keywords.To address the above problems,this thesis conducts research on keyword extraction on large-scale data and how to use deep learning methods to achieve keyword extraction,and the main research results are as follows.1.A keyword extraction algorithm based on hybrid feature network is proposed.The entropy weight method is used to fuse the weighted structural network and the weighted semantic network of the text to construct the lexical influence network of the text.At the same time,the index of link influence of nodes is defined to judge the importance of word nodes in the lexical influence network.The semantic tree of the keyword node is constructed and analyzed in detail,and the algorithm effect is judged by the semantic tree.2.A keyword extraction algorithm based on graph neural networks is proposed.An algorithm for automatic score generation of candidate words is proposed to transform the keyword extraction problem into a linear prediction problem.By fusing the attention mechanism with the graph convolutional layer,the lexical characteristics and text network features are effectively combined.Batch processing of the data through the subnet of the network is proposed,which greatly reduces the time originally spent due to large-scale matrix operations.Finally,by comparing with existing baseline methods,the proposed method can match or surpass the existing baseline methods,and the new proposed network structure has better keyword extraction effect than the traditional structure,which truly realizes the deep learningbased keyword extraction method.