Our project could be useful in determining how fake reviews could be generated. Therefore, we chose to use Kaggle datasets that contain customer review data to train our models. Our focus was on generating product reviews, which hasn’t been studied as intensely as other areas in text generating. We directly compared the LSTM character-based model to n-gram models trained on the same wine review dataset and found that the trigram model generated more realistic reviews. While creating the LSTM models, we found that the character-based model generated more realistic reviews than the word-based LSTM, so we focused on the character-based LSTM model. We created a character-based LSTM model, a word-based LSTM model, and a word-based n-gram model. Our main contribution was comparing different text generation methods and applying them to product review datasets. We continued tuning these models until we were satisfied with the reviews that they generated. We created LSTM models and n-gram models for each of the data sets, but found that the wine dataset generated the most realistic reviews. We began by removing all punctuation except periods, commas, exclamation marks, apostrophes, and question marks.
We preprocessed three different datasets: reviews of technical Amazon products, women’s clothing, and wine.
Throughout this project, we wanted to generate realistic product reviews using different datasets and different models. We want to apply machine learning to generate fake product reviews and see if we can pass them off as real product reviews. As a general rule, reviews aren’t closely monitored, and consumers are often allowed to post reviews regardless of their veracity. Someone with malicious intent can damage the reputation of a company’s product with a few fake negative reviews. The number of fake reviews on popular websites, such as Amazon, has increased in recent years in an attempt to influence consumer buying decisions. Online reviews are easily accessible and easy to post and share with others. Percentage of Online Consumers Affected by Product Reviews
0 kommentar(er)
