Product Evaluation Prediction Model Based on Multi-Level Deep Feature Fusion

Traditional product evaluation research is to collect data through questionnaires or interviews optimize design, but the whole process takes a long time deploy and cannot fully reflect market situation. Aiming at this problem, we propose prediction model based on multi-level deep feature fusion of online reviews. It mines satisfaction from massive reviews published by users e-commerce websites, uses analyze relationship between design attributes customer satisfaction, products satisfaction. Our proposed can be divided into following four parts: First, DSCNN (Depthwise Separable Convolutions) layer pooling are used combine extracting shallow features primordial data. Secondly, CBAM (Convolutional Block Attention Module) realize dimension separation features, enhance expressive ability key in two dimensions space channel, suppress influence redundant information. Thirdly, BiLSTM (Bidirectional Long Short-Term Memory) overcome complexity nonlinearity prediction, output predicted result connected layer. Finally, using global optimization capability genetic algorithm, hyperparameter constructed above carried out. The final forecasting consists series decision rules that avoid redundancy achieve best effect. has been verified method paper better than above-mentioned models five indicators such as MSE, MAE, RMSE, MAPE SMAPE, compared with Support Vector Regression (SVR), DSCNN, DSCNN-BiLSTM. By predicting emotional it provide accurate decision-making suggestions for enterprises new products.