Article Preview
Top1. Introduction
With the rapid growth of Internet technology, the rapid promotion of mobile payment and the vigorous development of the logistics industry, online shopping has rapidly become integrated into people's lives. Analyzing online product marketing information can provide online public opinion trends for enterprises, help them improve their products, and enhance user experience (Seidel et al., 2020). In the face of massive amounts of information, it is an arduous task to obtain the required information by enterprises only manually. Big data technology can solve this problem. Using various learning algorithms to analyze the emotions and opinions expressed by people can automatically mine people's positions and feelings about something.
The rapid progress of various websites, online media and short video applications make information no longer limited to text form and develops more information forms, including image, voice, and video. To comprehensively consider the complementary relationship between various information data, Multimodal Sentiment Analysis (MSA) comes into being. The effective combination of text, images, voice, video and other modal information can better guide sentiment analysis and improve its accuracy (Aderonke et al., 2020; Ghorbanali et al., 2022). Besides, with the rapid development of information technology, large-scale data input needs to be processed with big data technology. Meanwhile, after the text data processing is completed, the data are sent to the Convolutional Neural Network (CNN) for text feature extraction, which can make the sentiment analysis of the multimodal model more accurate.
Based on the above background, this work starts from the text mining of online product marketing information, discusses its text preprocessing methods in text mining, and analyzes the CNN variant Temporal Convolution Network (TCN) model. The Attention Mechanism (AM) and Transformer structure are adopted to design an MSA model of TCN Based on AM (AM-TCN). The online marketing information of a mobile phone on a shopping platform was the experimental object. The model's performance was verified through the accuracy rate and the harmonic mean F1 value. The main research innovation was to design the AM-TCN model through text data mining and use more Transformer structures to optimize the model's performance. The primary logical structure of the work is as follows. In Section 1, the background of the Internet and big data is introduced. In Section 2, the random forest classifier is used to classify the data through the literature research on text analysis methods and emotion models. In Section 3, the MSA model of online product marketing information is established through text mining and text preprocessing. In Section 4, the performance of the model designed in the experiment is evaluated. In Section 5, the final research conclusion is drawn through the comparison and discussion of the experimental data. The work has significant reference value for promoting the efficiency of model data processing (Lu et al., 2022).