Sentiment Analysis in Sales Estimation: An Econometric Analysis of Product Listings and Reviews in a Chinese Cross-Border E-Commerce Context
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Business |
Bachelor's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
2021
Department
Major/Subject
Mcode
Degree programme
Tieto- ja palvelujohtaminen
Language
en
Pages
23 + 9
Series
Abstract
Since the advent of electronic word-of-mouth communication, particularly in the form of user-generated reviews on e-commerce platforms, research has been undertaken to quantify and draw insights from this growing wealth of data. Coinciding developments in machine learning and natural language processing have enabled the systemic analysis of these texts, heightening the role of user feedback from a simple information channel between users to an indispensable source of “big data” information regarding consumer sentiment and behaviour. While otherwise extensive, contemporary research into the role of consumer sentiment, and, in particular, its effect on sales outcomes, is largely built around data gathered from Western e-commerce platforms, most notably Amazon. This has potentially limited its generalizability to wider contexts. In addition, many studies simplify the role of feedback valence by interpreting the sentiment polarity of a written review as equivalent to its corresponding numerical rating – a conflation that seems to go against existing research into rating inflation and other biases. This study seeks to further this field of e-commerce research by accounting for these issues. Us-ing cross-sectional data gathered from an industry-leading Chinese cross-border e-commerce platform, this study analyses the relationships between user-generated review sentiments and order amounts in a new context. By applying three different sentiment analysis tools to a total of 451,375 product reviews, overall sentiment polarity and subjectivity metrics were calculated for 8,319 product listings. Using these values, alongside other control variables (including numerical ratings, separate from sentiment polarities) from the listings, econometric regression models de-scribing the relationships were estimated and interpreted. The findings of this study demonstrate that, on a broad level, the notion of review sentiment polarity being positively related to sales outcomes is generalizable beyond the Western context. The role of a more nuanced aspect of review sentiments, namely the subjectivity of reviews, is found to be seemingly different from existing research into Western platforms, albeit somewhat inconclusively. The findings also support the notion that review sentiment polarity is not directly represented by its corresponding numerical rating, and that future studies should continue to differentiate between these two metrics. This study leaves open the exact causal nature of these relationships, requiring future research using time series data over multiple years. In addition, a greater variety of product categories could be studied in order to confirm the overall generalizability of these findings.Description
Thesis advisor
Seppälä, TomiKeywords
sentiment analysis, big data, e-commerce, online reviews, business analytics