Audience Sentiment and Mood Analysis

Parrot Analytics utilizes sophisticated, large-scale sentiment and mood analysis pipelines to uncover overall audience feelings, as well as specific moods, towards particular episodes, seasons, or overall TV show brands. The sentiment and mood analysis provides a deeper layer of understanding into the relationship between content and its consumers, adding a contextual layer of insight to the industry’s leading global content demand measurement system.

These industry-leading advancements are yielding previously unseen results with empirical correlations between the sentiment for a TV show and its overall demand (see below).

 

Audience Sentiment Analysis

Using text-based data from social channels, forums, fan and critic ratings platforms and community-based sites, we are able to carry out episode-specific sentiment analysis that can be drilled into as well as aggregated to track overall seasonal changes in sentiment. The Parrot Analytics “opinion index” is used to measure the overall sentiment for a TV show:

  • The opinion index is positive if audiences are, on average, expressing positive opinions about a show.
  • The opinion index is negative if audience opinions are, on average, negative.

The following screen shows an example of Parrot's episode-by-episode sentiment analysis:

sentiment-analysis.jpg

The level of positive or negative sentiment can be established prior to the start of a season, during the season, as well as post-season. Season-on-season sentiment analysis is also possible.

The level of positive or negative sentiment can be established prior to the start of a season, during the season, as well as post-season. Season-on-season sentiment analysis is also possible.

 

Audience Mood Analysis

Sentiment can also be expressed in terms of primary audience moods, expressed as a % contribution. Parrot Analytics’ mood analysis is currently based on a combination of AI and computational linguistics techniques. By incorporating social tagging, episode labels and other contextual metadata via the company’s Content Genome, this approach avoids the issue of "social noise" to ensure accurate sentiment and mood analysis.

The mood breakdown utilizes the eight primary, behavior-driving, emotions:

  • Anger
  • Anticipation
  • Disgust
  • Fear
  • Joy
  • Sadness
  • Surprise
  • Trust

The underlying data for these 8 emotions is generated by applying a custom lexicon, based on word–emotion associations, which has been specifically setup to handle social TV data analysis.

The following screen shows an example of Parrot's mood analysis for a TV show over multiple season life-cycle stages:

The primary moods surrounding a title can be established episode-by-episode, or season-on-season as above, to understand if fans are looking forward to new plot developments, for example.

Audience moods can obviously change during the season, as can the overall opinion of a series. For example, during the post-season period audiences may regain some trust if this has decreased during the season.

 

Demand Correlation Analysis

It is also possible to correlate mood with both our opinion index, as well as our demand data. For example, we have found that both demand and sentiment increase with positive moods like anticipation, and decrease if negative moods like anger are prevalent.

We have also discovered that high arousal emotions like anticipation and surprise have the greatest positive impact on the opinion index, whilst low arousal emotions like fear and sadness have the greatest negative impact.

The following screen demonstrates the inverse relationships between how certain moods correlate with the overall audience opinion, as well as demand data.

mood_correlation_analysis.png

Pipeline Processing Methodology

Parrot Analytics’ processing methodology leverages textual (e.g. social) data for comment pre-processing and subsequently applies a custom text classification machine learning system that has been trained on (and continues to learn from) hundreds of millions of TV fans’ inputs globally.

At the highest level, the system establishes the mood pipeline using a lemmatized word-recognition engine based on a crowd-sourced word-emotion association lexicon, custom-trained for content-related comments:

  1. Word-category disambiguation: The process of marking up a word in a text (corpus) as corresponding to a particular part of speech, based on both its definition and its context.

  2. Lemmatization: The process of grouping together the inflected forms of a word so they can be analysed as a single item, as demonstrated in the table below:



  3. Word Recognition: The process of applying a bespoke word–emotion association lexicon, which has been customized for textual TV data analysis. 

From there, a final layer of sentiment and mood allocation takes place for each comment according to the previously laid out approach, resulting in a highly accurate set of moods that can be tracked over time to ensure data-driven decision making can take place to maximize the content’s success.

Have more questions? Submit a request

0 Comments

Article is closed for comments.