基于专利的技术机会识别：深度学习领域的案例分析

Patent-based Technology Opportunity Identification: A Case Study Based on Deep Learning

摘要:

为及时有效地识别潜在技术机会，采用文本挖掘和异常值检测的方法，提出一种基于专利文本的技术机会识别方法。首先采用文本表示模型Doc2vec技术对专利摘要进行建模，以更深层表征文本语义信息；然后利用基于密度的离群值检测算法识别出具有潜在技术机会的专利方向；最后以深度学习领域潜在技术识别为例，构建专利检索式并收集458条专利文献作为数据集。实证结果总结出4类主题共10个潜在的技术机会，验证了该基于专利的技术机会识别方法的有效性，可为企业相应技术应用、研发和创新提供参考。

Abstract:

To identify potential technology opportunities timely and effectively, this paper proposes a technology opportunity identification method based on patent text by using the methods of text mining and outlier detection. Firstly, the paper uses Doc2vec to model the patent abstract to represent semantic information of the text. Then, the paper uses a density-based outlier detection algorithm to identify patents with potential technology opportunities. Finally, the paper takes the identification of potential technologies in the field of deep learning as an example for case analysis, constructs a patent search query and 458 patent documents are collected as the dataset. The empirical results summarize four kinds of topics and ten potential technology opportunities, which verifying the effectiveness of this patent-based technical opportunity identification method, and can provide reference for the enterprise corresponding technology application, research and development and innovation.