Abstract:
Relation extraction is a very useful task for several natural language processing applications,
such as automatic summarization, knowledge graph development and question answering. An
entity relation defined as a semantic interaction that holds between Named Entities. The entity
relation extracting system developed for English or any other language in some specific
xiii
domain cannot work for other languages of the same domain. Previously, there is research
conducted on entity relation extraction from Amharic language free texts for a sentence have
only one relation. When we study the Amharic sentence behaviors, it has two and more entities
that exist so the relation between those named entities is also triple and multiple relations it
has. This research attempted to design entity relation extraction from Amharic language free
text with a sentence having multiple relations. Due to the number of relation s between the
named entities and the existence of triple and more relations of a sentence, to address the issues
it is essential to develop a multi-label relation classification. Task-related entity indicators
designed to enable a deep neural network to concentrate on the task-relevant information. By
implanting entity indicators into a relation instance, the neural network is effective for
encoding syntactic and semantic information about a relation instance.
We conduct the experiments using logistic regression classical machine learning algorithm in
addition to encoder decoder models. These are LSTM, Bi-LSTM and CNN-BiLSTM. To
conduct our experiments, we have used 2,500 sentences; it carries both triple and multiple
relations. In order to extract our relation data, we have used entity indicators. To propose an
optimal model with best relation classification from Amharic language entities unit we
consider efficiency (training time, memory usage, and accuracy score). Finally, we have
proposed multi-label relation classification using BiLSTM model with an accuracy score of
0.55. The major weakness of the study is unavailability of enough dataset to conduct an
extensive experiment. As a result, there is a need to prepare corpora for conducting similar
research.
Keywords: Named Entity, Relation Extraction, Entity Indicator Based Relation Extraction,
and Deep learning approach, LSTM, BiLSTM and CNN-BiLSTM.