Intelligent Question Answering using Knowledge Representation for Managed Services

M. Saravanan, Satheesh K. Perepu

Abstract: Field service operations (FSO) personnel play a critical role to handle the issues in Managed Services for the telecommunication industry. It is required to support the FSO’s by giving the necessary resources and make them to solve the problem in the most efficient way. Currently, in managed services related to different industries, the FSO has been provided with documentation of the product/s that they are dealing with, which span to hundreds and thousands of pages. However, solving a problem efficiently requires the information in a concise manner by considering FSO’s legitimate time. In this paper, we propose two methods to extract necessary information from the thousands of pages of documents based on the issue the FSO is currently handling. The first method presents a hybrid methodology that combines effectively the results of both the models (term categorization and topic modelling) through the modified loss function to efficiently search through the knowledge graph representation for picking the relevant set of documents. As a second method, we have introduced a new fine-tuned Bidirectional Encoder Representations (BERT) model for Question and Answering (QA) purpose and it has been modelled as a classification problem. The proposed fine-tuned BERT approach automatically generates an answer to the FSO’s questions during emergencies from the selected documents using the first method. We found both these methods are working effectively with >90% accuracy and efficiency in most of the cases relate to managed services.