基于异构文献网络表示学习的个性化引文推荐技术研究

Personalized citation recommendation technology can help researchers quickly and accurately obtain the required scientific papers from massive scientific literature. Based on the study of heterogeneous bibliographic network representation learning, this project deeply integrates authors of scientific papers, contents of scientific papers, venues of scientific papers as well as the interaction relationship among these three factors, and researches on heterogeneous bibliographic network representation learning based personalized citation recommendation. Including (1) incorporates the network structure and node content information of heterogeneous bibliographic network, and researches on DeepWalk based heterogeneous bibliographic network representation for personalized global citation recommendation approach. (2) Captures the highly non-linear structure of heterogeneous bibliographic network by means of Generative Adversarial Network (GAN), and studies GAN based heterogeneous bibliographic network representation for personalized global citation recommendation approach. (3) Through introducing supervised information and defining a structure preservation function and a link prediction function, develops a predictive heterogeneous bibliographic network representation learning based personalized global citation recommendation approach. (4) Constructs a neural citation network containing encoder, attention mechanism and decoder, and puts forward a personalized local citation recommendation approach based on neural citation network representation learning. Research findings of this project can significantly improve the performance of citation recommendation system, and provide theoretical and technical support for its practical application.

个性化引文推荐技术有利于帮助科研人员从海量科技文献中快速准确地获取所需文献。本项目以异构文献网络表示学习为研究背景，深度融合文献作者、文献内容、文献来源及相互之间作用关系等多维因素，研究基于异构文献网络表示学习的个性化引文推荐技术，包括：(1)综合异构文献网络的网络结构信息和节点内容信息，研究基于DeepWalk异构文献网络表示学习的个性化全局引文推荐方法；(2)利用GAN获取异构文献网络中的高度非线性结构，研究基于GAN异构文献网络表示学习的个性化全局引文推荐方法；(3)引入有监督信息，定义结构保持函数和链接预测函数，研究基于预测式异构文献网络表示学习的个性化全局引文推荐方法；(4)构建包含编码器、注意力机制和解码器的神经引文网络，研究基于神经引文网络表示学习的个性化局部引文推荐方法。本项目的研究成果能显著提高引文推荐系统性能，并为其实际应用提供理论和技术支持。

个性化引文推荐技术有利于帮助科研人员从海量科技文献中快速准确地获取所需文献。本项目以异构文献网络表示学习为研究背景，深度融合文献作者、文献内容、文献来源及相互之间作用关系等多维因素，研究基于异构文献网络表示学习的个性化引文推荐技术，包括：(1)综合异构文献网络的网络结构信息和节点内容信息，研究基于DeepWalk异构文献网络表示学习的个性化全局引文推荐方法；(2)利用GAN获取异构文献网络中的高度非线性结构，研究基于GAN异构文献网络表示学习的个性化全局引文推荐方法；(3)引入有监督信息，定义结构保持函数和链接预测函数，研究基于预测式异构文献网络表示学习的个性化全局引文推荐方法；(4)构建包含编码器、注意力机制和解码器的神经引文网络，研究基于神经引文网络表示学习的个性化局部引文推荐方法。实验结果表明，所提方法都表现出了优于基线模型的性能。本项目的研究成果能显著提高引文推荐系统性能，并为其实际应用提供理论和技术支持。