Author(s)

Qiang Lu, Longan Xiong, Yaoyao Liu

Corresponding Author

Qiang Lu

ABSTRACT

In this era when artificial intelligence is very popular, natural language processing has naturally attracted people's attention and has been widely used in many fields, including text analysis. In our article, we set up a model that can extract and quantify specific texts, an analysis model based on Rolling Time Windows, successfully extract and analyze various information contained in comments in online shopping, and prove that this analysis method is accurate enough and can be applied in reality by comparing with other models. As a creative way, we preprocessed the data mining, removed the incomplete parts from the given data, and then analyzed the user's comments using Term Frequency-Inverse Document Frequency (hereinafter called the “TF- IDF”) algorithm and Python-based Latent Dirichlet Allocation (hereinafter called the “LDA”) to classify and quantify them. In order to change the obtained data from discrete to continuous, we skillfully use Cosine Similarity Algorithm and linear weighting method to obtain very ideal numbers as comprehensive evaluation values. In order to further analyze the influence of user comments in it, we use the dynamic collaborative filtering recommendation model based on rolling time window, and use the model of the first point to process quantitative comments according to time series, and successfully analyze and obtain the curve of user satisfaction changing with time. From the above three graphs, we find that the comprehensive evaluation value of the hair dryer has been showing an upward trend, and the evaluation value is higher, while the comprehensive evaluation value of the microwave is lower, and shows a downward trend. So we think that the hair dryer is the potentially most successful product, and the microwave is the potentially most failed product.

KEYWORDS

TF-IDF; LDA; Cosine Similarity Algorithm; python