Social Media Analytics: Chapter 10

• Social Media Analytics is based on Chapter 10 of a SMM book and additional material.

Social Media Research

• Social media research uses scientific marketing research principles to collect and analyze social media data for valid and reliable results.
• Social media research uses data derived from social media sources.
• Secondary research is information already available internally, publicly, or via syndicated sources, including market, industry, competitor, and brand history.
• Primary research collects data for research purposes at hand to understand consumers' psychology, spending, media consumption, and response to appeals.
• Primary research methods include exploratory, qualitative methods (observation, focus groups, interviews), descriptive techniques (surveys), or experimental techniques (simulations).

Social Monitoring and Social Listening

• Social monitoring tracks mentions of specific words/phrases on social media and triggers a company response, and is reactive.
• Social listening identifies and collects social data for analysis to inform strategic marketing decisions, and is proactive.

Top Applications of Listening

• Social Customer Care: Brand mentions identify service satisfaction issues, aiding social CRM tactics and detecting "disservice" experiences.
• Market Research: Informs decisions for marketing strategists, ranging from new product development ideas to audience insights for campaigns.
• Campaign Assessment: Provides feedback on how a campaign or brand communication was received by others.
• Social Media Listening involves appropriate research design and setting research protocols, ensuring data collected are appropriate for research uses.
• It uses software to systematically searches and scrapes data from social media channels.
• Data includes qualitative (verbatim comments) and quantitative (source, volume, unit characteristics) data.
• Common applications include sentiment analysis and content analysis.

Sentiment Analysis

• Sentiment analysis determines the writer's attitude or emotion (positive, negative, or neutral).
• The process involves:
  • Fetching, crawling, scraping, and cleansing data from sources using web crawlers and word-phrase dictionaries.
  • Extracting relevant posts and filtering data to tag entities of interest.
  • Extracting sentiment using sentiment indicators and a sentiment dictionary.
  • Aggregating the data into a sentiment summary.
• The Bag-of-Words Model proposes lists of positive and negative words and counts their occurrences.

Sentiment Analysis Challenges

• Accuracy of automated tools is a challenge as they struggle with coding meaning; systems combining human analysis, keyword meaning, and natural language processing have better accuracy.
• Cultural and linguistic nuances and differing contexts make it hard to code text into categories, for instance, describing a chocolate torte as "wickedly sinful" has a positive intent, not negative.
• Multiple word meanings can complicate sentiment dictionaries when words of same spelling have different meanings.
• Uncertainty about the source occurs as it is difficult to assess who is commenting based on demographic and geographic descriptors.

Social Intelligence

• Social Intelligence involves capturing, managing, and analyzing social data to identify and apply insights to business goals.
• Predictive Analytics uses data, algorithms, and machine learning to predict future outcomes based on historical data.

Social Intelligence Model

• The steps are:
  • Social Listening and data capture (source data).
  • Data management (classify and organize data).
  • Data analytics (analyze data using advanced analytics and predictive models).
  • Distribution of intelligence (translate data into insights and deliver to appropriate departments).

Data Analytics: Machine Learning

• Unsupervised Learning: Group and interpret data based on input data.
• Supervised Learning: Develop a predictive model based on both input and output data.
• Reinforcement Learning: Train machines through trial and error.

Unsupervised Learning - Clustering

• It learns from unlabeled data with the objective to find the hidden structure.
• A program looks for patterns in unlabeled data, identifying trends people aren't explicitly seeking.

Supervised Learning - Classification

• It learns from labeled data, and aims to investigate how inputs affect classification and predict classifications for new inputs.

Supervised Learning - Regression

• It learns from labeled data and seeks to investigate how inputs affect the output level and predict output levels for new inputs.
• Supervised machine learning models are trained with labeled data sets that allows a model to learn and become more accurate over time.
• Linear regression analysis predicts a dependent variable's value based on the value of an independent variable, and estimates the coefficients of a linear equation.

Reinforcement Learning

• It involves learning through trial and error with the objective to take the best action by establishing a reward system.
• Reinforcement learning trains models to play games or train autonomous vehicles by telling the agent (machine) when it makes the right decisions.

Google Trends

• Interest over time shows relative popularity, where 100 is the highest point and other points are scaled accordingly, and excludes absolute numbers.
• Interest by subregion shows a higher proportion of queries with 100 representing the location of highest popularity.
• Related Queries/Topics:
  • Top shows the most popular search queries/topics, with the most commonly searched having a value of 100.
  • Rising indicates the highest increase in search queries/topics since the last time period, including breakouts with tremendous increase.

Social Listening via Google Trends

• Keyword Selection involves avoiding ambiguity and considering language issues and how selected keywords relate to research questions.
• Location includes worldwide, U.S., and interest by region/subregion.
• Time Range is since 2004.
• Category is the type of search such as web, image, news, Google Shopping, and YouTube search.
• Related Topics / Queries shows users searching for certain terms.
• Comparison allows comparing a maximum of 5 terms across time range and locations.

Applications

• Find opportunities for new product/service launches, why, when, and where, and what are the challenges?
• Analyze trends in a product/service category, comparing competitors, assessing differences across locations, and determining improvements.
• Investigate the impact of seasons, holidays, days of the week/month on an industry.
• Determine if calendar effects are present, if they vary by location, if firms are using them, and if there are business opportunities driven by calendar effects.