Text Mining and Social Media Sentiment Analysis: Measuring Brand Equity and Customer Satisfaction Post-Product Launch

The weeks after a product launch are when a brand’s promise meets customer experience, and social media is the only source that records that encounter as it happens. Firms increasingly turn posts and reviews into sentiment indicators, yet the link between a classifier’s accuracy on individual posts and the quality of the brand and satisfaction indicators built from them is rarely examined. This paper studies that link with a text-mining pipeline that covers spam filtering, sentiment classification, aspect-level aggregation, advocacy mining and topic discovery, evaluated on a simulated corpus of 18,560 launch-period posts over 12 weeks in which the true customer experience, including a delivery failure, a firmware fix and a price promotion, is known. A rule-based lexicon reaches only 0.534 macro-F1 on posts from weeks 4 to 12, whereas supervised linear models trained on 3,000 labelled posts from the first three weeks reach 0.795 (naive Bayes), 0.811 (logistic regression) and 0.815 (linear support vector machine). Despite that gap, the weekly net sentiment series of every method follows the true trajectory of customer satisfaction closely (correlations of 0.89 to 0.95), but the methods differ in accuracy of level: mean absolute error against the true net sentiment is 0.021 for the support vector machine and 0.050 for the lexicon, which compresses the delivery dip and the later recovery. Aspect-level tracking recovers the delivery, battery and price episodes (correlations of 0.94, 0.88 and 0.86) but is weaker for support (0.65). A brand-equity composite built from net sentiment, advocacy phrases and mention volume tracks the simulated brand equity at 0.96, whereas mention volume alone does not track it at all. Periodic relabelling brought no measurable gain in this corpus, and unsupervised topics aligned only modestly with product aspects. All data are synthetic, so the results illustrate an evaluation design and should not be read as expected performance on real platforms.

Authors

Publication Details

Journal
Mathematics for Applications
Published
2026-11-18
DOI
https://doi.org/10.66838/ma.240
Primary Topic
Sentiment Analysis and Opinion Mining
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Text Mining and Social Media Sentiment Analysis: Measuring Brand Equity and Customer Satisfaction Post-Product Launch

Mahesh Reddy Challa
Mathematics for Applications
Sentiment Analysis and Opinion Mining
article

Text Mining and Social Media Sentiment Analysis: Measuring Brand Equity and Customer Satisfaction Post-Product Launch

Mahesh Reddy Challa
article en

Abstract

The weeks after a product launch are when a brand’s promise meets customer experience, and social media is the only source that records that encounter as it happens. Firms increasingly turn posts and reviews into sentiment indicators, yet the link between a classifier’s accuracy on individual posts and the quality of the brand and satisfaction indicators built from them is rarely examined. This paper studies that link with a text-mining pipeline that covers spam filtering, sentiment classification, aspect-level aggregation, advocacy mining and topic discovery, evaluated on a simulated corpus of 18,560 launch-period posts over 12 weeks in which the true customer experience, including a delivery failure, a firmware fix and a price promotion, is known. A rule-based lexicon reaches only 0.534 macro-F1 on posts from weeks 4 to 12, whereas supervised linear models trained on 3,000 labelled posts from the first three weeks reach 0.795 (naive Bayes), 0.811 (logistic regression) and 0.815 (linear support vector machine). Despite that gap, the weekly net sentiment series of every method follows the true trajectory of customer satisfaction closely (correlations of 0.89 to 0.95), but the methods differ in accuracy of level: mean absolute error against the true net sentiment is 0.021 for the support vector machine and 0.050 for the lexicon, which compresses the delivery dip and the later recovery. Aspect-level tracking recovers the delivery, battery and price episodes (correlations of 0.94, 0.88 and 0.86) but is weaker for support (0.65). A brand-equity composite built from net sentiment, advocacy phrases and mention volume tracks the simulated brand equity at 0.96, whereas mention volume alone does not track it at all. Periodic relabelling brought no measurable gain in this corpus, and unsupervised topics aligned only modestly with product aspects. All data are synthetic, so the results illustrate an evaluation design and should not be read as expected performance on real platforms.

Mathematics for ApplicationsVol. 15(2s)
Sentiment Analysis and Opinion Mining
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon