Graph-Based Representation of Customer Reviews for Online Stores

Abstract

The purpose of this paper is to investigate the graph-based representation of the data required for the vector space model (VSM) and PMI (pointwise mutual information)-enriched VSM used for text mining (e.g. text classification). The transformation of a dataset containing free text reviews for online stores in a graph-based form is described and its format that allows to be used by Neo4j graph database management system is proposed. Queries for retrieving the data required for training text mining models are considered; the steps and the actions for their modification when receiving new data are specified. The advantages of the proposed graph-based representation in regard to the maintenance and the extraction of current data needed for retraining data mining models are summarized, in order to prevent loss of performance of results from the execution of the respective data mining task.