采用元路径时效衰减和引用模式划分的学术引文推荐

Citation Recommendation Based on Time-Efficiency Decay and Citation Pattern Partition of Meta-Path

针对现有学术引文推荐算法中元路径特征无法衡量学术文献的时效性,且对元路径特征的利用和划分粒度较粗从而导致推荐精度不高的问题,提出了一种采用元路径时效衰减和引用模式划分的学术引文推荐方法。首先,利用元路径抽取丰富的引文特征,并在计算元路径特征时加入了时效衰减因子,提升了新发表文献的推荐精度;其次,提出了融合元路径特征的主题模型MpTM,该模型利用主题特征为文献划分引用模式,并联合学习文献的主题分布、引用模式和元路径特征权重,细化了元路径特征的粒度;最后,通过所有引用模式下的元路径特征加权值,为目标文献推荐学术引文。在AAN数据集上的实验结果表明:所提方法在准确率和召回率上平均提升约41.99%和22.43%,能够提升新发表文献和非权威文献的推荐精度,并能有效缓解引文链接的稀疏性问题。

The meta-path approach in present citation recommendation algorithms cannot measure the time-efficiency of papers and its utilization and partition are relatively coarse-grained, which results in low citation recommendation performance. This paper presents a citation recommendation method considering time-efficiency decay and citation pattern partition of meta- path. We first extract abundant citation features according to the meta-path, and then calculate these features by extending Randomwalk with time-efficiency decay factor to improve the citation performance of newly published papers. Based on these meta-path features, we present a meta- path based topic model （MpTM）. This model utilizes topic model to partition citation patterns and can jointly learn topic distributions, citation patterns and meta-path feature＇s weights. The citations are finally recommended according to the fine-grained meta-path features among all citation patterns. Experimental results in AAN dataset showed that our proposed method has improved the average accuracy and recall by about 41.99% and 22.43%, respectively. It can promote the recommendation performance for newly published papers and unauthoritative papers, and effectively relieve the link sparse problem in citation dataset.