DEVELOPMENT OF A SEARCH ENGINE MODEL FOR E-COMMERCE WITH A SEMANTIC APPROACH AND FEATURE EXPANSION
E-commerce is projected to reach $7.96 trillion trillion in global sales by 2027, driven by increasing adoption of mobile technology. Search engines on e-commerce platforms play a crucial role in the shopping experience, and this study explores the implementation of semantic search using transfor...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/85139 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | E-commerce is projected to reach $7.96 trillion trillion in global sales by 2027, driven by increasing
adoption of mobile technology. Search engines on e-commerce platforms play a crucial role in the
shopping experience, and this study explores the implementation of semantic search using
transformer models for electronic products. Using the CRISP-DM methodology, model evaluation
shows that 'ft_all-MiniLM-L12-v2' excels with a micro F1 score of 0.4052 and a macro F1 score of
0.3826, while 'ft_all-mpnet-base-v2' achieved the highest weighted F1 score of 0.3443. However,
these models also have longer search times, particularly 'ft_all-mpnet-base-v2' at 1.1096 seconds.
The 'all-MiniLM-L12-v2' model has the fastest search time at 0.4887 seconds but with lower
accuracy. 'TF-IDF' showed the lowest performance across all metrics. In conclusion, feature
expansion improves accuracy and relevance, despite increasing search time, making 'ft_all-
MiniLM-L12-v2' and 'ft_all-mpnet-base-v2' balanced options between speed and accuracy. |
|---|