Automatic Parameter Selection of Granual Self-organizing Map for Microblog Summarization

摘要

In this paper, a neural-network-based unsupervised classification technique is proposed for summarizing a set of tweets where informative tweets are selected based on their importance. The approach works in two stages: in the first stage, the concept of a self-organizing map (SOM) is utilized to reduce the number of tweets. In the second stage, a granular self-organizing map (GSOM), which is a 2-layer feed-forward neural network utilizing the fuzzy rough set theory for its training, is considered for clustering the reduced set of tweets. Then, a fixed length summary is generated by selecting tweets from the obtained clusters. GSOM is having a set of parameters; proper selection of these parameter values influences the performance. Therefore an evolutionary optimization technique is utilized for the selection of the optimal parameter combinations. We have evaluated the efficacy of the proposed approach on four disaster-related microblog datasets. Results obtained clearly illustrate that our proposed method outperforms the state-of-the-art methods.