PERFORMANCE IMPROVEMENT OF 7 HYBRID BLOCKS METHOD USING GATED RECURRENT UNIT FOR RATING CALCULATION AND DEMOGRAPHIC FILTERING ON RECOMMENDER SYSTEM
With the increasing number of internet users now, the use of system for buying and selling online is a very vital requirement. One of the keys to the success of the system for buying and selling online is the recommender system, because it can provide benefits for sellers, and for buyers to make it...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/36816 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | With the increasing number of internet users now, the use of system for buying and selling online is a very vital requirement. One of the keys to the success of the system for buying and selling online is the recommender system, because it can provide benefits for sellers, and for buyers to make it easier to get relevant items quickly. Collaborative filtering (CF) is the most popular recommendation system method, because it shares interests among users based on ratings so it is more dynamic. Nevertheless, CF still has several issues, such as data sparsity, cold start, gray sheep and dynamic taste.
Several studies have tried to accomplish these issues with hybrid method using combination of several techniques. One study tries to accomplish the issues by building method using 7 blocks of hybrid techniques with various approaches. However, method for calculating ratings from reviews on one block in the study has problems with reviews with different length variations. This is because the method uses accumulation of adjectives that make a long gap between the values for short reviews and long reviews. This certainly decreases the overall quality of the recommender system. This study aims to propose another method for rating calculations using gated recurrent unit (GRU). Meanwhile, the new block, demographic filtering (DF) by utilizing user social information will be added to add personalization, especially to new users. Based on the results obtained, the proposed method using GRU as rating calculation provides better results than the method in the previous study, where there is an increase of around 3-5% at the f-measure value. While the addition of new block DF does not give significant positive effect as proposed by the hypothesis. |
|---|