Project summary

• Project name: Understanding Changes in a Brand’s Core Positioning and Customer Engagement: A Sentiment Analysis of a Brand-owned Facebook Site

• Authors: Zhenning (Jimmy) Xu, PhD, Colin Vail, MBA, Amarpreet Kohli, PhD, Saeed Tajdini, PhD

• Keywords: relationship communication, meaning creation, NLP, Social Media, Sentiment Analysis, KPI, Word Frequency Analysis, Word Association Analysis, Cluster Analysis

Drawing new insights from unstructured data

• “Data is the new oil”
• The data Dilemma - Data rich but insight poor (IBM,2019; Tableau, 2020)
• Macro-level data vs. micro-level data (e.g., textual data)
• 80% of data will be unstructured in five years (King 2019)

• Main challenge: drawing new insights from unstructured data

• Pop question: Can you give an example of unstructured data? How is it different from structured data? Feel free to share your answers in the chat.

Reference: Data Rich but Information Poor: Building a Data-Driven Culture in Manufacturing https://www.tableau.com/learn/webinars/data-rich-information-poor-building-data-driven-culture-manufacturing

Research Objective

• Observation - While the number of total Facebook posts in 2016 is 37% greater than the previous year (236 vs. 172) in our data, the number of likes, comments, and shares dropped significantly from 2015 to 2016.

• Objective - to identify the main topics of communications on brand-owned social media and their evolutions that might be revealing of a brand’s positioning on social media and the level of customer engagement, as measured by the number of likes, comments, and shares the brand’s posts receive on the brand’s social media

The application of sentiment analysis has never been more accessible than now

• API (Application Programming Interface) -Facebook Graph API, Twitter API, Web APIs, service APIs (Uber, Lyft, Here, Mapbox, DoorDash Tracking API, etc.)
• R allows us to scrap unstructured data using the Facebook Graph API
• The Facebook Graph API - the major problem behind Cambridge Analytica’s scandal in 2016 (Albright 2018)
• R offers a varieties of open source libraries or packages for us to analyze unstructured data
• This presentation (html format) is made possible due to the awesome packages and platforms contributed by the R open source data science community (Xie et at., 2018)

Reference:

CRAN now has 10,000 R packages https://blog.revolutionanalytics.com/2017/01/cran-10000.html#:~:text=10%2C000%20R%20packages.-,Here’s%20how%20to%20find%20the%20ones%20you%20need.,R%20packages%20available%20for%20download*.

Xie, Y., Allaire, J. J., & Grolemund, G. (2018). R markdown: The definitive guide. CRC Press. https://bookdown.org/yihui/rmarkdown/

Collecting Data with the Facebook API and Rfacebook (an R package)

Rfacebook is an R library that allows us to create an app on Facebook’s developer website (https://developers.facebook.com) and collect data

Refernce: https://cran.r-project.org/web/packages/Rfacebook/Rfacebook.pdf

A graphical representation of unstructured social media data analysis

A feedback mechanism linking analytics, communications, and brand positioning and customer engagement on social media

An Integrated Relationship Communication Model

An integrated relationship communication model that connects situational factors to meaning creation via analytics (adapted from Finne and Grönroos 2009, 2017)

• Situational factors are the aspects that often influence the consumer’s meaning creation process.

• The effectiveness of these situational factors in brand communication could manifest itself in the way how customers interact with the virtual touchpoints, as reflected by the number of likes, comments, and shares of the social media posts in the context of this study.

• Since the place is named “Thompson’s Point,” the most important situational factor for both years is “point.”

Method - Text analytics workflow

Text analytics workflow for evaluating brand-owned social media textual data)

• Different sentiment analysis methods and R packages for text analytics used in this study might open new doors for marketers to derive micro-level data-driven insights from brand-owned social media

Word Cloud Analysis

Comparison of word clouds featuring Facebook post content in 2015 and 2016

Word Frequency Analysis

Top situational factors identified using the word frequency analysis

Word Assoication Analysis

Text analytics workflow for evaluating brand-owned social media textual data

• There was a slight fading of the brand’s core brand position.

• The communication language used in 2016 is focused on a broader context of business activities, thus less engaging and interactive with consumers.

Cluster analysis for year 2015

Text analytics workflow for evaluating brand-owned social media textual data)

• The brand’s situational factors are captured in clusters 1, 2, and 3 as these clusters contain words related to the company’s service offerings, location information and brand-related features.
• Core brand messaging (i.e., hashtag ‘gettothepoint’) and other situational factors (e.g., ‘theater’) can be communicated and tracked over time

Cluster analysis for year 2016

Text analytics workflow for evaluating brand-owned social media textual data)

• The left side of the dendrogram are composed of two terms only (“services” and “company”).

• These words are associated with a broader context of the business vs. the venue’s specific offerings.

Web Analytics is complex, but it is also highly rewarding

• Skill Diversity - Like cats and dogs, marketers and data analysts tend to have different responsibilities and mindsets. In general, data analysts love pattern recognition, experimentation, and forecasting, while marketers live for creativity, storytelling, and visual design.

• Micro-level big data - The current literature has been more occupied with the volume of data or how “big” the data is (Wedel and Kannan 2016) than discovering the deeper meanings in textual and micro-level big data such as brand-owned social media data (Baesens et al. 2016; Iacobucci et al. 2019).

• A recursive feedback loop - According to the traditional IMC model, the corporate content is usually “pushed” without actionable data for a recursive communication feedback loop.

• The comparative word clouds, word frequency analysis, word association, and cluster analyses are incorporated in the proposed analytical workflow to visually illustrate how brand- owned content, in the form of social media textual data, can be used to evaluate brand positioning and customer engagement over time.

• Insights from this form of analysis become more effective when a relationship communication strategy exists, and where core brand messaging (i.e., hashtag ‘gettothepoint’) and other situational factors (e.g., ‘theater’) can be communicated and tracked over time.

• References upon request