발간년도 : [2023]
| 논문정보 |
|
| 논문명(한글) |
[Vol.18, No.5] A Comparative Research on CNN Model Performance for Product Recommendation Based on Time Series Image Data |
|
| 논문투고자 |
Jun-Hyeon Sim, Chul-Jin Kim |
|
| 논문내용 |
In the modern world, advances in information technology have led to the expansion of e-commerce, making it important for automated recommendation systems to efficiently gather the flood of information and data to present consumers with their favorite products and services. Various techniques are used to improve the accuracy of product recommendation in existing e-commerce. Among them, there are chronic problems that use RNN, a multi classification-based product recommendation model. RNN is a deep learning model suitable for time series classification tasks, but it suffers from issues such as gradient vanishing and gradient exploding. To Compensate for these issues, CNN models are often used to effectively detect local patterns through kernels. In this study, we compare the performance of recommendation models based on an architecture that generates product recommendation models by training CNN models with time series data through three different imaging encodings: GAF, MTF and RP. In our experiments, we split the 540,000 published transaction dataset into train and test. The splitted data is constructed as time series data and zero-padded to equalize the size of the model’s input image. We train AlexNet, VGG16, ResNet50, and MobileNet models on images generated by the three imaging algorithms and compare their product recommendation accuracy with the performance of existing RNN recommendation models. We can see that the CNN models perform better than the LSTM. When imaged with the GAF algorithm and trained on the MobileNet model, the highest recommendation accuracy was achieved, and the learning time was also shortened, improving efficiency. Future research will include the advancement of imaging algorithms to improve the performance of product recommendation models and the development of CNN models optimized for time series image data. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
