加权专利文本主题模型研究

Weighted Topic Model for Patent Text Analysis

【目的】解决专利文本分析中主题模型向高频词倾斜、区分度低的问题。【方法】提出基于词权重方法,形成加权专利文本主题模型,给不同的词分配不同的权重,改变生成模型生成词的概率。【结果】相较于传统的专利文本主题模型,本文提出的加权专利主题模型能够增加主题间的区分度。【局限】加权算法需要更多数据集验证,并不断优化。【结论】通过专利文本数据验证了该方法的可行性与有效性。

[Objective] This study aims to address the issues facing the topic model of patent text analysis such as the inclining to high frequency words and low discrimination rates. [Methods] First, we proposed a word weighting method for the traditional topic model. Then, the modified model assigned different weights to the words, and changed the probability of generating new words. [Results] Compared with traditional methods, the weighted patent topic model could identify the subjects more effectively. [Limitations] The weighting algorithm needs to be validated and optimized with more datasets. [Conclusions] The proposed model could effectively analyze the patent texts.