Analysis of Product Value Based on NLP and Time Sequence Model

Abstract

In most instances, necessity is the mother of invention. For example, the generation of B2C business mode enables the business company to capture data such as star ratings, reviews, and helpful votes from online customers. Therefore, the company can have an insight into the market conditions and explore prospective customer demands. In this paper, we aim to take advantage of innovative algorithms and models to conduct scientific analysis and data mining. Original data are provided by Sunshine Company, including star ratings, reviews, helpful votes, and more. Our solution is composed of three main sections. In order to reduce errors from original data, we use Python to pre-process the provided data. It will significantly improve the credibility and efficiency of the data. The data pre-processing includes three steps: Text Deduplication, Mechanical Compression and Tag Filtration. In terms of model construction, we use different algorithms in multiple-dimensions to analyze the data. For instance, Factor Analysis, Grey Relational Analysis, and Cluster Analysis are proposed to process reviews and obtain text-based product information like meaningful quantitative patterns, and relationships between star ratings and reviews. We use the LDA algorithm to visualize the word frequency. Then the Sentiment Word Analysis is performed to quantify the comments, and the user s comprehensive evaluation of the product is obtained by linear weighting of comments and ratings. In order to solve problems in multiple dimensions, we develop a brand-new kind of Time Sequence Model which can obtain the best relevant combination of the products. Specific analysis processes are: Assume each day as a measurement, Weight the customer reviews to obtain overall reviews for that day, and Visualize the date of reviews. We creatively use Time Characteristics Analysis to analyze whether customer reviews are affected by others at the end. In order to ensure the stability and credibility of the Time Sequence Model, we conduct a sensitivity analysis, and results of the test are good. Generally speaking, the results of the data analysis can basically answer the questions raised by Sunshine Company. We put forward some innovative ideas and suggestions for reference.