Opinion based intelligent recommender system
Recommender system (RS) is one of area of machine learning research. Building an accurate and useful RS has become important for both research and commercial field. One of the most recurrent problem in RS is data sparsity. It is difficult to get sufficient explicit ratings data, however textual data...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/77274 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Recommender system (RS) is one of area of machine learning research. Building an accurate and useful RS has become important for both research and commercial field. One of the most recurrent problem in RS is data sparsity. It is difficult to get sufficient explicit ratings data, however textual data such as social media posts on Twitter, Facebook, etc. are plentiful and easily retrievable. These social media data should be utilized properly, thus it can help relieve the RS data sparsity problem. Considering the importance and opportunity given to improve the current RS technology, this project aims to build an intelligent RS that utilize textual data to represent user’s preference toward a brand or product.
This project consists of two tasks, first, the development of Twitter Sentiment Visualizer, a web application that retrieve and visualize live public opinion toward certain brand names, second, integration of sentiment information on existing recommendation algorithm. The first task is done by applying sentiment analysis process on Twitter Tweets data. The system is built on Dash by Plotly.js, a Python framework to develop a single web page application meant for data visualization. The second task is done by applying sentiment analysis process on Amazon 5-core reviews dataset then feeding it as an input to 3 machine learning based recommendation algorithms.
The Twitter Sentiment Visualizer web application successfully shown that public opinion toward certain brands can be extracted from social media – in this case Twitter. The experiment on Amazon 5-core dataset shown improvement of performance, in terms of root mean square error (RMSE) and mean absolute error (MAE), of all algorithms when sentiment information is considered.
In conclusion, this project successfully shown the possibility of utilizing plentiful data on social media to increase the performance of current RS. |
|---|