Concept Lattice Parallel Generation Algorithm And Its Application In Recommendation System

In the period of Internet with rapidly increasing information,a large amount of data has been created continually.There are various difficulties when retrieving information such as we couldn't find what we need exactly.Thus,a totally new method should be developed to obtain precise knowledge.Concept lattice,as a tool for data mining,could mine the underlying value from huge amount of data which resolved the problem of obtaining message.In terms of this aspect,the theory and application of concept lattice have been researched including constructing concept lattice and its application.And concept lattice is also hard to be stored and computed with respect to larger volume data.Therefore,it's a problem to be done.We realize the storage and computing of big data via the cloud-computing architecture by parallel computing and distributed storage for Formal Concept Analysis(FCA).It works on the Hadoop's distributed file system(HDFS)and Map-Reduce module.We develop the algorithm for generating concept lattice at the first step,and then realize it under the cloud-computing architecture.The experimental result shows that this algorithm is feasible and efficient.We have also studied the theoretic knowledge and construction of concept lattice,the main job of this article is distributed storage and parallel computing for concept lattice.Hadoop's distributed file system and Map-Reduce module are the technical support.Concept lattice is an efficient tool for data analysis.It can describe the underlying relationship between object and attribute,among concept,with which we can find the relationship intertwined within people and substance.Thus a recommendation system which based on this kind of association rules emerged.Nowadays,however,a lot of recommendation system faced two problems: one is cold start and the other is data sparsity.For the second problem,we have proposed a new scenario.A matrix completion scenario was proposed to solve the problem of matrix sparsity in FCA collaborative filtering recommendation in our article.After analyzing various technology of matrix completion,the inexact augmented Lagrange multiplier was selected to recover the matrix.We use singular value decomposition collaborative filtering algorithm on the recovered matrix.Evaluating recommended data utility the measure of Mean Absolute Error and Root Mean Squared Error.The experiment shows that the accuracy is higher when using the recovered matrix.