Convolutional-NN and Word Embedding for Making an Effective Product Recommendation Based on Enhanced Contextual Understanding of a Product Review

E-commerce is essential application in world wide. In everyday live, we cannot escape from e-commerce transaction. E-commerce requires intelligent machine to deliver product information to customer. Intelligent machine popular called recommender system developed by matrix factorization. Rating is representation of customer expression for satisfied product or service. Unfortunately, number of ratings is too sparse due to majority customer lazy to give rating for e-commerce product. Number of sparse rating matrix have impact in matrix factorization in rating prediction. Moreover, extreme sparse rating has impact degrade performance significantly. Many researchers consider to enhance matrix factorization using customer and product information such as customer demographics information, customer testimony, product review, product description, and etc. In this research, we consider to incorporating Stack Denoising Auto Encoder (SDAE), attention mechanism aims to enhance product review document understanding representation and matrix factorization based on Probabilistic Matric Factorization (PMF) to produce rating prediction. According to experiment report, our model superior over previous work based on hybrid model between PMF, Convolutional Neural Network (CNN) and SDAE that popular called PHD in 1%, and superior over hybrid model between PMF, Long Short Term Memory (LSTM) and SDAE that popular called DDL-PMF in 0.9%, and achieved significantly in 8% over traditional PMF based on RMSE evaluation metrics.

Section: Methodsmentioning

confidence: 99%

Enhance Rating Prediction for E-commerce Recommender System Using Hybridization of SDAE, Attention Mechanism and Probabilistic Matrix Factorization

2022

“…At the time, deep learning achieved phenomenal performance in the image processing classification task. Additionally, deep learning has established itself as the industry standard for a variety of computer science-related problems, including image processing [15], speech recognition, recommender system using contextual document based on CNN [16][17][18], LSTM [19,20], SDAE [21], Word embedding and LSTM [22].…”

Section: Related Workmentioning

confidence: 99%

Enhance Intrusion Detection (IDS) System Using Deep SDAE to Increase Effectiveness of Dimensional Reduction in Machine Learning and Deep Learning

2022

The intrusion detection system (IDS) is very essential tools to detect malicious network. IDS is a hardware or software approach to observe the internet for malicious attacks. It has ability to screening an internet or network that possibility dangerous activity or security threats. IDS application responsible to defend network territory in accordance with the network-based intrusion detection system (NIDS) or host-based intrusion detection system (HIDS). Using known normal network activity signatures, IDS applications perform tasks by comparing them to known attack activity signatures. In this study, a dimensional reduction and feature selection mechanism known as the stack denoising auto encoder (SDAE) was found to be effective in increasing the effectiveness of naive bayes, KNN, decision tree, and SVM classification algorithms. The researchers evaluated the performance using evaluation metrics such as a confusion matrix, accuracy, recall, and the F1-score, among other measures of success. When compared to the results of previous studies in the IDS field, our model using statistical pre-processing, dimensional reduction based on SDAE success to increase the effectiveness of KNN, naive bayes, decision tree, SVM and deep learning using LSTM. We applied our experiment in the NSL-KDD Dataset. According to evaluation metrics using confusion matrix, accuracy, recall, and f1 that the effectiveness of our model achieve more than 2% over several previous work without statistical pre-processing and dimensional reduction based on SDAE. Furthermore, the use of statistical approaches and SDAE improved the accuracy of traditional machine learning and modern deep learning based on LSTM. Aims to improve the effectiveness of IDS detection in the future, it may be possible to integrate SDAE with another deep learning model such as MLP, CNN, attention, and GAN.

“…The hybrid of dual deep learning is called PHD-PMF [26]. The model inspired three state-of-the-art past models including SDAE [27], CNN [28] [29], and PMF [6]. PHD-PMF, implemented on popular real datasets, including MovieLens, and its movie review documents were obtained by Amazon.…”

Section: Introductionmentioning

confidence: 99%

Handling Sparse Rating Matrix for E-commerce Recommender System Using Hybrid Deep Learning Based on LSTM, SDAE and Latent Factor

2022

E-commerce is the most essential application for conducting business transactions. Delivering product information to customers require an essential machine called recommender system. Recommender systems have been adopted in many large e-commerce companies such as Amazon, e-Bay, Alibaba, YouTube, iTunes, and so on. Ratings have become an essential factor to calculate product information. They are users' expressions about their satisfaction regarding a product or service. Unfortunately, the number of ratings is extremely sparse. Generating rating prediction is a major issue in the recommender system research field. The most popular model using latent factor or matrix factorization to generate rating prediction faced the problem in accuracy performance. This research aimed to develop a novel model to generate rating prediction using two deep learning variants based on Stack Denoising Auto Encoder (SDAE), Long Short Term Memory (LSTM), and combining with a latent factor model based on Probabilistic Matrix Factorization (PMF). This study considered integrated information resources, including user information and document product information. Following the experiment report involved in Movielens and Amazon Information Video dataset, our model outperformed previous works using PMF, Collaborative Deep Learning (CDL), Probabilistic Hybrid Deep Learning (PHD) and LSTM-PMF model, with more than 5% in average using Root Mean Square Error (RMSE) evaluation metrices.