Improving Amharic Information Retrieval Using Transformer Based Language Models

Information retrieval systems are essential for obtaining relevant information from large-scale digital resources. However, current IR systems are language-dependent and have challenges in under-standing the semantic meaning of queries. This study aims to develop a transformer-based semantic information retrieval model for Amharic documents. The model was developed by collecting, cleaning and normalizing 4,372,180 Amharic documents using the design science research methodology. The evaluation was per-formed on three multilingual transformer models, XLM-RoBERTa, mBERT-Cased, and mBERT-Uncased, in different training and testing configurations in Python 3.7. The experimental results show that XLM-RoBERTa achieved the best performance with an F-measure of 92.09% (97.09% precision and 87.64% recall) for an 80/20 train-test split and 91.52% F-measure (94% precision and 88.9% recall) for a 70/30 split. The results demonstrate that transformer-based models are capable of capturing semantic relation-ships in Amharic queries and improving the performance of information retrieval. Further research can investigate advanced transformer architectures and pre-training methods to improve Amharic semantic retrieval systems.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11114849/v1
Primary Topic
Information Retrieval and Search Behavior
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Improving Amharic Information Retrieval Using Transformer Based Language Models

Anbesaw Belete Mekonnen, Dereje Chinkl Kassie, Shambel Kefelegn Bereded, Andargachew Alayu Eshete
1 links
Research Square
Information Retrieval and Search Behavior
preprint

Improving Amharic Information Retrieval Using Transformer Based Language Models

Anbesaw Belete Mekonnen, Dereje Chinkl Kassie, Shambel Kefelegn Bereded, Andargachew Alayu Eshete
preprint en

Abstract

Information retrieval systems are essential for obtaining relevant information from large-scale digital resources. However, current IR systems are language-dependent and have challenges in under-standing the semantic meaning of queries. This study aims to develop a transformer-based semantic information retrieval model for Amharic documents. The model was developed by collecting, cleaning and normalizing 4,372,180 Amharic documents using the design science research methodology. The evaluation was per-formed on three multilingual transformer models, XLM-RoBERTa, mBERT-Cased, and mBERT-Uncased, in different training and testing configurations in Python 3.7. The experimental results show that XLM-RoBERTa achieved the best performance with an F-measure of 92.09% (97.09% precision and 87.64% recall) for an 80/20 train-test split and 91.52% F-measure (94% precision and 88.9% recall) for a 70/30 split. The results demonstrate that transformer-based models are capable of capturing semantic relation-ships in Amharic queries and improving the performance of information retrieval. Further research can investigate advanced transformer architectures and pre-training methods to improve Amharic semantic retrieval systems.

Research Square
Debre Berhan University (ET)
Information Retrieval and Search Behavior
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon