EXPERT FINDING SYSTEM IMPROVEMENT WITH MULTILABEL CLASSIFICATION IN INDONESIAâS ACADEMIC INSTITUTION
The need for a credible expert in his field is always needed both in academia and industry. The weakness of some of the mainstream expert search systems at this time is the lack of validation of the track records of these expert candidates. The expert finding system that was built previously can...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/62137 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The need for a credible expert in his field is always needed both in academia and
industry. The weakness of some of the mainstream expert search systems at this
time is the lack of validation of the track records of these expert candidates. The
expert finding system that was built previously can overcome this problem, but it
creates another weakness, namely that it can only map a problem description into
one field, even though one problem description may consist of multiple fields.
This expert finding system was built using lecturer’s data, taken from the PD Dikti
Indonesia page. In general, the system consists of two subcomponents, namely field
classification and expert candidate recommendations. The construction of the
classification model uses TF-IDF feature extraction and produces multi-label
classification results, to accommodate the need for more than 1 field for a problem.
Calculation of expert scores uses the weighted sum model method. Aspects that
become the value of expertise are aspects of education, teaching, research,
community service, and positions held.
The system is built in the form of web-based application with the Django
framework. The test consists of testing the system functionality, testing the results
of model development, and testing the quality of expert recommendations. The
classification model is quite good with an accuracy of 71% with the SVM algorithm
for 10 skill classes.
The system that is built is expected to be used for wider use in academic institutions
or the government to find the right experts, especially for research needs or
community service. In the future, it is recommended to multiply the data so that the
model can learn to classify more specifically into subfields, so that it can deepen
the scope of use.
|
|---|