Research And Application Of Graph Indexing Technology Based On Subgraph Division

The data relation structure based on graph index is the basic component of application fields such as recommendation system,data mining,and information retrieval.As the process of cultural resource intelligence continues to accelerate,cultural resource data is fragmented and unstructured,which makes it very difficult to define and mine the relationship between public cultural resource entities,which makes it difficult for users to obtain the required cultural resources.The approximate nearest neighbor search algorithm can quickly search for data similar to the query object in the candidate set.The approximate nearest neighbor search algorithm based on graph index has attracted much attention because of its fast search speed and high search accuracy.However,the search algorithm based on graph index has high complexity in the index construction stage,and requires high computing resources,and cannot handle massive,scattered and unstructured cultural resources.At the same time,the use of greedy search on the graph index has a probability of falling into a local optimal solution,which in turn leads to a decrease in the recall rate.Therefore,this thesis proposes a graph index structure suitable for the field of cultural resources,which not only optimizes the index construction complexity and retrieval recall rate,but also has been verified by the national public cultural cloud experimental platform and achieved good results.Specifically,the following research contents are included:1.The graph index-based approximate nearest neighbor search algorithm needs to compute and store more neighborhood information,which leads to a high resource consumption in its construction process.To address the problem of large computational resource consumption in index construction,this thesis proposes a Graph Partition-Navigating Spreading-out Graph index construction process.In the index construction stage,the computational resource consumption is effectively reduced by reducing the number of composition nodes;in the index search stage,a search method based on dynamic pruning strategy is proposed to navigate the query nodes to the closer subgraphs,which reduces the probability of falling into local optimal solutions during greedy search.The experimental results show that the memory consumption for index construction decreases by 61% and 34% on SIFT1 M and GIST1 M datasets,respectively,and the retrieval recall rate is effectively improved by setting the cluster selection factor.2.In this thesis,we design and implement a semantic search platform for cultural resources based on graph indexing.Based on the guided dispersion graph based on subgraph division,the semantic analysis model is used to intelligently understand the semantics of user needs and establish the correlation between user needs and cultural resources.The feature extraction model is used to obtain the feature vectors of resources,and the approximate nearest neighbor search algorithm is used to construct the index and provide vector search service,which is combined with the traditional keyword matching search function to realize the accurate matching between user needs and cultural resources.By processing cultural resources,the number of resources that will be able to be retrieved is increased in the case of the same resource base.