TY - GEN
T1 - Attribute Value Extraction in Weapon Domain Based on Bi-LSTM and Attention
AU - Wu, Yu
AU - Miao, Lin
AU - Li, Han
N1 - Publisher Copyright:
© 2023 Copyright held by the owner/author(s)
PY - 2023/12/14
Y1 - 2023/12/14
N2 - Aiming at the problem that the traditional extraction method caused by the diversification of weapon attributes has a large amount of work to construct the label of weapon attributes, in this paper, we propose a weapon attribute value extraction method based on bidirectional long-term and short-term memory network (Bi-LSTM) and attention mechanism. The method first uses the Bi-LSTM model to extract the features of the input text and attribute names. Then, the attention mechanism focuses on the relations between words and attributes in the sentence. Afterward, the global BIO tag marks the position of the attribute values in the sentence. In this way, the method can reduce the workload during the corpus preparation period to improve the generalization ability of the model so that it can extract different weapon attribute data. Compared with Bi-LSTM, Bi-LSTM_CRF, and OpenTag from the experimental results, the F1 values of the proposed model on the weapon domain attribute dataset are increased by about 6.9%, 5.7%, and 2.5%, respectively.
AB - Aiming at the problem that the traditional extraction method caused by the diversification of weapon attributes has a large amount of work to construct the label of weapon attributes, in this paper, we propose a weapon attribute value extraction method based on bidirectional long-term and short-term memory network (Bi-LSTM) and attention mechanism. The method first uses the Bi-LSTM model to extract the features of the input text and attribute names. Then, the attention mechanism focuses on the relations between words and attributes in the sentence. Afterward, the global BIO tag marks the position of the attribute values in the sentence. In this way, the method can reduce the workload during the corpus preparation period to improve the generalization ability of the model so that it can extract different weapon attribute data. Compared with Bi-LSTM, Bi-LSTM_CRF, and OpenTag from the experimental results, the F1 values of the proposed model on the weapon domain attribute dataset are increased by about 6.9%, 5.7%, and 2.5%, respectively.
KW - Attribute Value Extraction
KW - Information Extraction
KW - Knowledge Base
KW - Natural Language Processing
UR - http://www.scopus.com/inward/record.url?scp=85192193561&partnerID=8YFLogxK
U2 - 10.1145/3638884.3638979
DO - 10.1145/3638884.3638979
M3 - Conference contribution
AN - SCOPUS:85192193561
T3 - ACM International Conference Proceeding Series
SP - 603
EP - 610
BT - ICCIP 2023 - 2023 the 9th International Conference on Communication and Information Processing
PB - Association for Computing Machinery
T2 - 9th International Conference on Communication and Information Processing, ICCIP 2023
Y2 - 14 December 2023 through 16 December 2023
ER -