跨类型的学术资源优质推荐算法研究

Research on Cross-Type Excellent Recommendation Algorithm for Academic Resources

提出一种新的融合内容特征和非内容特征以及用户行为的推荐算法ER(Excellent Recommendation),快速准确地为用户推荐感兴趣的、高质量的5类学术资源,以实现学术资源的优质推荐。ER算法从资源类型、学科分布、关键词分布和LDA(Latent Dirichlet Allocation)主题分布共4个内容特征对5类学术资源建模,融合用户行为后进行用户兴趣偏好建模,根据权威度、社区热度和时新度等3个非内容特征对学术资源的质量值进行评估,最终根据学术资源的兴趣值和质量值进行Top-N推荐。通过预测准确度对推荐结果进行评估,实验表明ER算法的推荐效果最佳,证明了ER算法的有效性。

Excellent recommendation （ER）, a new recommendation algorithm that fuses content ann non-content fea- tures and user behaviors, is proposed to enable users to obtain high quality and multiple types of academic resources of interest quickly and accurately. In the ER method, academic resources will be modeled using four content features including resource type, subject distribution, keyword distribution, and LDA （Latent Dirichlet Allocation） topic dis- tribution. Based on the user behaviors and resource model, user interest will be modeled. The quality of academic resources is calculated using three non-content features including authority, community popularity, and freshness. Finally, top-N recommendations are provided according to the recommendation degree, which is calculated by the interest value and quality of academic resources. The prediction accuracy （precision） is applied to evaluate the per- formance of the ER method and the experimental results show that our approach is the best, thus verifying the validity of the ER method.