A Deep Learning-Based Innovative Points Extraction Method

Most of the research on mining online reviews now focuses on the influence of reviews on consumers and the issue of sentiment analysis for analyzing consumer reviews, but few studies how to extract innovative ideas for products from review data. To this end, we propose a deep learning-based method to extract sentences with innovative ideas from a large amount of review data. First, we select a product review dataset from the Internet, and use a stacking integrated word embedding method to generate a rich semantic representation of review sentences, and then the resulting representation of each sentence will be feature extraction by a bidirectional gated recurrent unit (BiGRU) model combined with self-attention mechanism, and finally the extracted features are classified into innovative sentences through softmax. The method proposed in this paper can efficiently and accurately extract innovative sentences from class-imbalanced review data, and our proposed method can be applied in most information extraction studies.