GRAPH REDUCED SUMMARIZATIONFOR CLUSTER ABSTRACTIONOF KLUSTREEBASED KEYWORD SEARCH RESULT

• Main Author: Muzdalifa, Annisa
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55853
• Institution: Institut Teknologi Bandung

This thesis discusses keyword search using the KlusTree method. The interesting thing about the KlusTree method is the use of language models to clustergraph structure data so that the search results are displayed in clusters. This research proposes a form of representation that increases information from the cluster by adding a labelinto each cluster.Labels are constructed usingabstraction ofa simplified graph from a collection of trees in the cluster. Abstraction in KlusTree is done by graph reduced summarization method, which is a method used for frequent pattern mining in knowledge graphs. The advantage of this method is that there isa bi-criteria function that assesses the informativeness and difference of various patterns so that it can be used as a summary of the knowledge graph.The thesis experiment was carried out using DBLP data which stores information on scientific works in the field of computer science. From the research conducted, the graph reduced summarization method can produce cluster abstractions and can be modified to calculate the quality of information generated from the KlusTree method. With each experiment entering100 queries, the best average value was 0.81 with ?0.1.Labeling by doing cluster abstraction can add information to the user. The proposed solution for constructing the abstraction of the cluster is feasible and gives good results. The graph reduced summarization method produces labels that match the contents of the cluster. In addition, the adapted bi-criteria function can evaluate the quality of KlusTree. The evaluation function using bi-criteria can be improved for further research so as to provide more precise values for Klustree.