Knowledge Powered Cooperative Semantic Fusion for Patent Classification
Zhe Zhang,Tong Xu,Le Zhang,Yichao Du,Hui Xiong,Enhong Chen
DOI: https://doi.org/10.1007/978-3-030-93046-2_10
IF: 14.4
2021-01-01
Artificial Intelligence
Abstract:Patent classification is beneficial for many patent applications, such as patent quality valuation, retrieval, and litigation analysis. Recently, many automatic patent classification methods have been proposed to save labor costs, which usually formulate this task as a multi-label text classification problem. In reality, patent language is highly terminological, full of scientific entities and domain knowledge. However, existing works seldom consider such unique property of patents, which reduces the classification performance. To this end, we propose a novel framework named Knowledge Powered Cooperative Semantic Fusion to capture deeper knowledge semantics for patent classification. Specifically, we first exploit knowledge graphs to enrich the patent with related entities. Then we design a mutual attention mechanism between entities and original texts to emphasize the crucial semantics of entities with the guide of texts, and vice versa. Finally, we introduce the graph convolutional network further to enhance the fusion representation of entities and texts. Extensive experiments on large-scale patent data demonstrate the superior performance of our model on the patent classification task.
computer science, artificial intelligence