Author(s)

Wen Wen, Xintian Cai

Corresponding Author

Wen Wen

ABSTRACT

We process the correlation between reviews and star ratings to remove inconsistent data simplify the product's comprehensive rating model to get new expressions. And we propose a vector autoregressive moving average model (AVAMA) based on the attention mechanism to predict the reputation trend of a product through multiple measures at multiple time steps. We also established a combination of text-based measure (s) and ratings-based measures, which are used to indicate the degree of product success. We use a model for predicting time series, the exponential smoothing method. From the formula of the method, we can see that the closer the data is to the prediction point, the greater the effect on the prediction. Next, we use correlation analysis to correct the difference. The evaluation did a quantitative analysis. We selected the pacifier dataset with a large number of samples and used the negative evaluation (rating 1) as an indicator to organize the data. Earlier evaluation did indeed have a significant impact on subsequent evaluations. Result. We use the ngram language model. We take the pacifier dataset as an example, and map the title of the evaluation and the subject's sentiment rating through Text Blob to [0, 5]. The larger the number, the more positive the feeling, and the smaller the the more negative, the experimental results show that whether it is the title or body of the review, the level of correlation with the rating has stabilized over time, and the correlation level is relatively good. Finally, we give some suggestions.

KEYWORDS

Online shopping platform, time series, exponential smoothing method