Patent term relationship extraction method based on Bi-LSTM and keyword strategies of attention mechanism

Abstract

The invention relates to a patent term relationship extraction method based on Bi-LSTM and keyword strategies of an attention mechanism. The method comprises the following steps: 1) preprocessing a patent text, identifying term features, adding position information, obtaining category keyword features through an improved TextRank algorithm, and forming a vector matrix by the category keyword features; 2), importing the vector matrix into a Bi-LSTM model, and obtaining overall features of text information by adopting an attention mechanism; 3) selecting the key feature of each sentence as a local feature by using the maximum pooling layer; 4), fusing the overall features and the local features; and 5), outputting a classification result by using a softmax classifier. On the basis of patentterm relationship extraction, the invention provides a Bi-LSTM and keyword strategy patent term relationship extraction method based on an attention mechanism in allusion to the problem of long-distance dependence in a traditional deep learning method. Through comparison of various experiments, the effect of the method is superior to that of an existing method, and the requirements of practical application can be well met.

Description

technical field [0001] The invention belongs to the technical field of patent term relation extraction, in particular to a patent term relation extraction method based on Bi-LSTM of attention mechanism and keyword strategy. Background technique [0002] With social development and scientific and technological progress, people's awareness of the protection of scientific research achievements has gradually increased, and the number of patent applications has also increased year by year. In order to more effectively analyze the relationship between patents and optimize patent retrieval, the research on automatic extraction of patent term relationships has received more and more attention. More and more scholars have paid attention to it. In the past, manual collection and extraction of unsupervised learning algorithms are far from meeting people's needs. It has become inevitable to use computers to automatically extract patent term relationships. The automatic extraction of pat...

AI Technical Summary

Problems solved by technology

[0003] At present, the main research methods of relation extraction include pattern-matching-based methods, dictionary-driven methods, statistical-based machine learning methods, and multi-method hybrid methods, but these methods require manual extraction of features, such as parts of speech, dependencies, and semantics. Roles, etc.; or rely on natural language processing tools to a certain extent, such as part-of-speech tagging, syntactic analysis, etc. However, there will be certain differences in the processing results of different tools, which will affect the final extraction results

[0004] In recent years, the use of deep learning methods for entity relationship extraction has become the mainstream, which can automatically learn and obtain effective text features. This method has been achieved in multiple natural language processing tasks without using basic natural language processing tools. However, this method is still insufficient in characterizing the local features and global features of sentences.

Images