Research On Self-classification Model Of Medical Patents Based On Deep Learning

The patent contains the vast majority of inventions and creations in the world.It is an important standard for measuring the innovation capacity of the country and enterprises,and an important way to obtain advanced knowledge.It is also an irreplaceable role for the pharmaceutical industry.On the one hand,medical patents can provide legal protection for pharmaceutical intellectual property rights,and on the other hand,they can provide researchers with current research information on various types of drugs and the latest scientific research technology information.Through research on existing pharmaceutical patents,not only can avoid useless investment but also obtain the latest research direction.A well-classified medical patent search system can effectively reduce the time for researchers to collect medical pa tent documents.In the past,patent classification models mostly used machine learning methods.The classification speed of the model is fast but the accuracy is not good.The main reason is that machine learning is not good at processing complex text.Recently,deep learning has received a lot of research because of its ability to mine the semantic and emotional information in complex texts,and it has performed well on tasks such as public opinion processing,sentiment classification,and translation.However,deep learning is rarely used in patent text classification tasks,and further research is needed.This paper focuses on the four issues of multi-label classification,large sample data processing,medical text representation,and data imbalance in medical patent text classification.Using the convolutional neural network and recurrent neural network in the deep learning method,a two-dimensional deep neural network model for feature extraction from sentences to words is established.The main contents of the study are as follows:(1)According to the characteristics of complicated medical patent texts and more professional vocabulary.Design a text processing module that conforms to the medical patent text,including text cleaning,Chinese word segmentation,and word vectorization.(2)According to the characteristics of deep learning,it is easier to deal with the characteristics of large sample data sets.This paper studies convolutional neural networks and recurrent neural networks,and establishes a two-dimensional deep neural network model for feature extraction from sentences to words.And add word-based attention mechanism to the network to further improve the classification effect of the model.(3)For multi-label problems and imbalanced sample distribution.The method of changing the activation function and loss function is adopted to solve.In order to verify the effectiveness of the medical patent text classification model proposed in this paper.The text compares the proposed model with other patent classification models.And using 6 evaluation indicators for evaluation,the results show that the classification effect of the medical patent text classification model proposed in this paper is superior to other patent classification models.The classification accuracy rate and accuracy rate can reach more than 90%.