With the intense competition of global intellectual property, the number of authorized patents is increasing. However, the patent conversion rate is low and the patent valuation is hard. The realization of patent valuation faces some basic challenges including: (1) how to develop a patent valuation model in consideration of technical factors; (2) how to train/test the patent valuation model with the insufficient standard value data. To solve the above issues, we assume that the realization of patent valuation begins with selecting the realistic value-paths: defense, implementation or transfer. We explore a Bayesian neural network-based model to predict the paths toward the realization of patent valuation. In the model, a function-effect-based patent representation is proposed, from which some technical features are extracted. Given the patent features, we use Bayesian neural network to predict the value-paths toward the realization of patent valuation. The model is evaluated by precision, recall, F-measure. The results show our method can improve evaluation measurements significantly after the addition of technical features.

随着全球知识产权的激烈竞争，授权专利的数量正在增加。但是，专利转换率低，专利估价困难。专利估价的实现面临一些基本挑战，包括：（1）如何考虑技术因素发展专利估价模型；（2）如何利用标准价值数据不足来训练/测试专利估价模型。为了解决上述问题，我们假设专利估价的实现始于选择现实的价值路径：辩护，实施或转让。我们探索基于贝叶斯神经网络的模型，以预测实现专利估价的途径。在该模型中，提出了一种基于功能效果的专利表示，并从中提取了一些技术特征。给定专利特征，我们使用贝叶斯神经网络预测实现专利估值的价值路径。通过精度，召回率，F-措施。结果表明，在添加技术功能后，我们的方法可以显着改善评估指标。