TY - GEN
T1 - A Framework for Adaptive Deep Reinforcement Semantic Parsing of Unstructured Data
AU - Jain, Shubham
AU - De Buitleir, Amy
AU - Fallon, Enda
N1 - Publisher Copyright:
© 2021 IEEE.
PY - 2021
Y1 - 2021
N2 - Semantic parsing of telecommunications data from nodes is an important topic of research in network monitoring and diagnostics. This data contains insights to achieve higher data rates, low latency and reliability. The current machine learning algorithm relies heavily on iterative meta-heuristics tools for data mining. Such tools often fail to capture the consistency, and complementary information inadequately which results in loss of high-level semantic information that can be used to extract features inside unstructured text data. The features learned by the current Machine learning algorithm can perform classification of unstructured data with high accuracy. However, the same methods fail to extract entities from data with high variation in structure and uneven density of words. Manual steps to parse and extract relevant information have failed to provide higher efficiency in extracting useful information which delays the process of network analysis. In this paper, we present an Adaptive Deep Reinforcement (ADR) framework fused with a Deep Q Network (DQN) for feature extraction and LSTM to extract relational dependency of words in unstructured data. A novel dynamic optimizer component based on LSTM and fully connected layers is adopted for dynamic state reformulation and Q values greatly enhance the parsing accuracy by extracting robust features to measure variation in data and identifying relationships between entities in raw data to parse them to a structured format. We compare ADR with state-of-the-art Drain parser. The complexity comparison of parsed files with state-of-the-art and ADR demonstrates the effectiveness of the proposed framework for semantic parsing of unstructured data.
AB - Semantic parsing of telecommunications data from nodes is an important topic of research in network monitoring and diagnostics. This data contains insights to achieve higher data rates, low latency and reliability. The current machine learning algorithm relies heavily on iterative meta-heuristics tools for data mining. Such tools often fail to capture the consistency, and complementary information inadequately which results in loss of high-level semantic information that can be used to extract features inside unstructured text data. The features learned by the current Machine learning algorithm can perform classification of unstructured data with high accuracy. However, the same methods fail to extract entities from data with high variation in structure and uneven density of words. Manual steps to parse and extract relevant information have failed to provide higher efficiency in extracting useful information which delays the process of network analysis. In this paper, we present an Adaptive Deep Reinforcement (ADR) framework fused with a Deep Q Network (DQN) for feature extraction and LSTM to extract relational dependency of words in unstructured data. A novel dynamic optimizer component based on LSTM and fully connected layers is adopted for dynamic state reformulation and Q values greatly enhance the parsing accuracy by extracting robust features to measure variation in data and identifying relationships between entities in raw data to parse them to a structured format. We compare ADR with state-of-the-art Drain parser. The complexity comparison of parsed files with state-of-the-art and ADR demonstrates the effectiveness of the proposed framework for semantic parsing of unstructured data.
KW - Deep Learning
KW - Reinforcement Learning
KW - Semantic Data Extraction
UR - http://www.scopus.com/inward/record.url?scp=85122940786&partnerID=8YFLogxK
U2 - 10.1109/ICTC52510.2021.9620904
DO - 10.1109/ICTC52510.2021.9620904
M3 - Conference contribution
AN - SCOPUS:85122940786
SN - 9781665423830
T3 - International Conference on ICT Convergence
SP - 1055
EP - 1060
BT - ICTC 2021 - 12th International Conference on ICT Convergence
PB - IEEE Computer Society
T2 - 12th International Conference on Information and Communication Technology Convergence, ICTC 2021
Y2 - 20 October 2021 through 22 October 2021
ER -