발간년도 : [2022]
| 논문정보 |
|
| 논문명(한글) |
[Vol.17, No.1] A Research on the Application of Word Embedding for Product Recommendation Based on Multi-Classification |
|
| 논문투고자 |
Chul-Jin Kim |
|
| 논문내용 |
In order to implement a multi-classification based product recommendation model in e-commerce, identifiable product information is expressed in a one-hot encoding method. However, the data representation of the one-hot encoding method consists of sparse data, which wastes memory and reduces the efficiency of learning. As an alternative to this issue, the embedding method is applied, and in this study, a processing procedure for converting and restoring a high-level dimensional one-hot encoded product code to a low-level dimensional word embedding product code is proposed to be applied to the product recommendation system of e-commerce. In addition, a word embedding encoder and a word embedding decoder for transformation and restoration are implemented and applied. The word embedding encoder receives a one-hot encoded product code as an input and derives a word embedding product code as an output. The optimal value derived through backpropagation of the hidden layer of the word embedding encoder becomes the word embedding product code. The word embedding decoder restores the word embedding product code derived through learning to the original one-hot encoded product code. In the experiment, on the basis of 540,000 open e-commerce transaction data, the suitability is verified by experimenting with the implementation of converting and restoring one-hot encoded product codes to 10-dimensional word embedding product codes for 2719 product data. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
