Predicting Navigational Patterns in Web Applications using Machine Learning Techniques
(2024)Department of Automatic Control
- Abstract
- In large corporations, customer support is a costly service, and an area of constant optimization. Solutions to increase efficiency and decrease bottlenecks are constantly needed. One such bottleneck is support tool proficiency in a customer support organization where many different tools are used, and a potential solution is to let the tool guide the user as it is being used. This thesis explores the use of machine learning to make predictions on user behaviour, based on user web log entries, to simplify the use of a customer support tool for unfamiliar users.
From 1.8 million log entries, four different datasets are created, based on two different data processing principles. On these, three machine learning models are trained, namely... (More) - In large corporations, customer support is a costly service, and an area of constant optimization. Solutions to increase efficiency and decrease bottlenecks are constantly needed. One such bottleneck is support tool proficiency in a customer support organization where many different tools are used, and a potential solution is to let the tool guide the user as it is being used. This thesis explores the use of machine learning to make predictions on user behaviour, based on user web log entries, to simplify the use of a customer support tool for unfamiliar users.
From 1.8 million log entries, four different datasets are created, based on two different data processing principles. On these, three machine learning models are trained, namely an LSTM, a transformer, and a hybrid CNN-LSTM model. These are then compared to a naive baseline model and evaluated on overall accuracy, top-three accuracy, and accuracy based on sequence length.
The results show that all trained models perform better than the baseline model, but not significantly for certain datasets. The trained models also perform very similarly on all datasets, but for long sequences, LSTM generally outperforms the others, reaching an overall accuracy of 75 percent for its best dataset. The respective accuracy for the baseline model is 72 percent. The close results between models can mainly be attributed to the low complexity of the tool from which the log entries originated and the few features they contain. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9148922
- author
- Balan, Patric and Jönemo, Gustav
- supervisor
- organization
- year
- 2024
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6224
- other publication id
- 0280-5316
- language
- English
- id
- 9148922
- date added to LUP
- 2024-02-22 11:27:25
- date last changed
- 2024-02-22 11:27:25
@misc{9148922,
abstract = {{In large corporations, customer support is a costly service, and an area of constant optimization. Solutions to increase efficiency and decrease bottlenecks are constantly needed. One such bottleneck is support tool proficiency in a customer support organization where many different tools are used, and a potential solution is to let the tool guide the user as it is being used. This thesis explores the use of machine learning to make predictions on user behaviour, based on user web log entries, to simplify the use of a customer support tool for unfamiliar users.
From 1.8 million log entries, four different datasets are created, based on two different data processing principles. On these, three machine learning models are trained, namely an LSTM, a transformer, and a hybrid CNN-LSTM model. These are then compared to a naive baseline model and evaluated on overall accuracy, top-three accuracy, and accuracy based on sequence length.
The results show that all trained models perform better than the baseline model, but not significantly for certain datasets. The trained models also perform very similarly on all datasets, but for long sequences, LSTM generally outperforms the others, reaching an overall accuracy of 75 percent for its best dataset. The respective accuracy for the baseline model is 72 percent. The close results between models can mainly be attributed to the low complexity of the tool from which the log entries originated and the few features they contain.}},
author = {{Balan, Patric and Jönemo, Gustav}},
language = {{eng}},
note = {{Student Paper}},
title = {{Predicting Navigational Patterns in Web Applications using Machine Learning Techniques}},
year = {{2024}},
}