Research On Summarization Generation Technology Based On Deep Learning

In this era of information explosion,text summarization provides people with a simple and fast way to read,and it has become one of the research hotspots.The current text summarization methods are based on two deep learning frameworks,namely Seq2Seq and Transformer,which have great improvement over the traditional methods.But there are still two shortcomings:(1)For the Seq2Seq model,the generated summary always loses some key information that can express the main content of the source text.Meanwhile,it tends to generate the same words when decoding,resulting in repetition.(2)For the Transformer model,its encoder is too single,which leads to incomplete semantic encoding.In view of the above problems,the corresponding methods are proposed:In order to solve the first problem,we propose an abstractive summarization model with a feature extractor and a semantic enhancer.First,the key features of source text are analyzed by feature extractor to filter the noise and achieve more accurate encoding.And then in order to further enhance semantic relevance and reduce repetition,we apply a semantic enhancer to indicate semantic relevance within the source text.Experimental results show that the performance of the model is greatly improved,and the generated summaries have high coherence and readability.In order to solve the second problem,we propose an abstractive summarization model of multi-attention fusion.In terms of attention range,the fusion of global attention mechanism and local attention mechanism is carried out to realize the combination of multiple attention focus and enrich the encoding content.On the granularity of attention implementation,the integration of word level attention mechanism and sentence level attention mechanism is carried out to realize the modeling of sentence as a whole and improve the encoding integrity.Experimental results show that the summaries generated by the model are richer and more comprehensive.Finally,a text summarization generation system based on deep learning is designed and implemented.The system can generate summaries based on the input articles or files.And two methods are supported,one is based on feature extraction and semantic enhancement fusion,the other is based on multi-attention fusion.