Article Preview
TopThe Bert Algorithm For Multimodal Emotional Analysis Of Online Products In E-Commerce Live Broadcasting Based On Big Data
With the penetration of the internet into all aspects of people’s lives, e-commerce has developed rapidly. As a platform for customer communication and interaction, an e-commerce website plays a crucial role in its success (Zhou et al., 2021). The design of e-commerce websites increasingly incorporates various modalities, including text, images, colors, audio, and video. E-commerce platforms, short video platforms, and others have recognized the profit potential of live broadcasts and have consequently entered the live broadcast field (Wang et al., 2021). With the rapid shift in consumer behavior, the public's capacity to embrace new trends has significantly increased. Currently, e-commerce live product broadcasts are saturated. These broadcasts involve video shooting, editing, and product promotion, using live resources to drive product sales and capture market opportunities.
Sentiment analysis (SA) is an emerging technique used to analyze opinion, emotion, subjectivity, and objectivity in text. It classifies reviews containing opinions or emotions as either positive or negative (Birjali et al., 2021). Businesses seek to understand what the public or consumers think about their products or services. Additionally, potential product users are interested in the evaluations of the product or service by consumers who have already purchased it (Hu & Chaudhry, 2020). People’s perceptions and decisions are primarily influenced by others' evaluations and opinions. E-commerce live broadcast sales rely on service quality, speed, and after-sales support. Consumer reviews and feedback are crucial for merchants and platforms (Dong et al., 2022). In the field of consumer SA, the naive Bayes (NB) classification algorithm is commonly used. This algorithm classifies sentences with similar emotional expressions to analyze emotional tendencies within the text (Novendri et al., 2020).
The purpose of this paper is to explore the multimodal SA of online products based on bidirectional encoder representations from transformers (BERT) in the e-commerce live broadcast environment. With the rise of e-commerce live broadcasts, an increasing number of consumers are purchasing products by watching live programs. However, the interaction between the audience and the anchor, product displays, comments, and other multimodal information pose challenges to SA. To address this issue, a multimodal SA method based on BERT is proposed. This method uses the pre-trained BERT model to analyze the sentiment of text comments. BERT can accurately capture emotional information in comments by understanding the meaning of words through bidirectional context. By considering text, image, and video information comprehensively, the product’s sentiment can be analyzed more accurately, thus providing a more precise reference for product recommendations and user experience improvements on e-commerce live broadcast platforms (Luo et al., 2024).
In e-commerce live broadcasts, users may express their emotions in various ways, including text comments, voice, images, videos, and more. Therefore, how to integrate these multimodal data and apply the BERT model for SA is a key challenge. To address this issue, this paper first discusses the development trends of the current e-commerce live broadcast environment and examines the product quality review process conducted by the anchor online. Secondly, it analyzes the multimodal sentiment related to the product after the user's consumption. The BERT algorithm is employed to capture users' ideologies and semantic levels, classifying different emotional expressions. The goal is to enhance the algorithm for multimodal SA of online products in the e-commerce live broadcast environment. This approach aims to boost product sales and improve the reputation of merchants’ products, thereby contributing to the positive development of the e-commerce live broadcast shopping trend.