基于PPIN的社交网络推荐系统

Social network recommendation system based on PPIN

为了提升海量数据下社交网络推荐系统的性能,将传统聚类方法与蛋白质网络的新特性相结合,提出了一种竞争-抑制节点模型(CINM).该模型将数据的整个处理流程分为节点重构、膜外聚类、膜内聚类及内容推荐4个部分,分别完成数据预处理、数据清洗、精度匹配与数据输出.在数据预处理过程中,通过矩阵运算,将复杂多维数据集构成的用户信息转换成结构化定量数据,并产生数据摘要.数据清理通过判断竞争值来获取用户的特征数据.在精度匹配阶段,基于蛋白质相互作用网络的相似性匹配原理获取相似性最大的一组值,并结合与用户相关联的数据项进行最终内容或关系的推荐.实验结果表明,CINM模型可以通过数据预处理和特征值竞争抑制机制较好地完成数据过滤,从而提高数据处理效率并提升最终推荐结果的精确性.

To improve the performance of the social network recommendation system on massive data,a competition-inhibition node model(CINM)is proposed by combing the traditional clustering methods with the new features of the protein networks.The whole processing flow is divided into four parts including node reconstruction,out-of-band clustering,intra-film clustering and content recommendation,in which data preprocessing,data cleaning,precision matching and data output are performed,respectively.In data preprocessing,the user information with the complex cube is converted into the structured quantitative data by the matrix operation,and the data summary is generated.In data cleaning,the user^characteristic data are obtained by judging the competition value.During the precision matching phase,a set of values with the greatest similarity are acquired by the similarity matching principle of the protein-protein interaction network.The final content or the relationship can be recommended by the user-association data items.The experimental results show that the CINM model can complete data filtering by data preprocessing and eigenvalue competition prefabrication mechanism to improve the efficiency of data processing and the accuracy of the final recommendation results.