Previously we discussed how Artificial Intelligence helps media enterprises understand their video content better. We introduced ourselves to the concept of ‘content entities’ and explored their utility in the present-day context where video consumption is becoming increasingly non-linear. 

Our changing media habits and the rise of video-on-demand (VOD) platforms have now set us free from our TV appointments and schedules. Today media enterprises are competing for audiences on different platforms and are not restricted to just TV time-slots.

As AI continues its foray into the News, Entertainment & Sports industry, it is dramatically changing how audience engagement can be measured. AI-generated content entities and their analysis are the keys to holding the audience’s attention for longer.

 But what is the need to understand content better?

Today, content is the most significant competitive differentiator there is, and a granular understanding of what works with audiences is bound to create winners. AI-generated Content entities hold the key to this change.

To recap –

Content Entities refers to people, objects, location, dialogues, keywords, or any other useful concept that can be used to describe what constitutes the content meaningfully.

This applies to any kind of content we consume, whether it be Entertainment, Sports, or News.

In this piece, we would discuss a few useful applications of these content entities across three different industry segments – Entertainment, News, and Sports.

Applications of AI – generated ‘content entities’ for sports & Media

There could be different contexts for analyzing content entities for multiple use-cases across different industry segments. For example, some use-cases might facilitate the creation of useful editorial content,while others are important from an ROI measurement perspective. In the forthcoming sections, we discuss three critical applications of content entities and bring some anecdotes from our own engagements with clients. 

  1. Audience Engagement Analysis for General Entertainment and News
  2. Examining editorial transparency of News
  3. Brand Logo Visibility analysis for Sports

Audience Engagement Analysis for General Entertainment 

Traditionally, TV Viewership Analysis at a target-market level is done via media-consumption data collected by the audience measurement bodies. While it tells which TV Show had high/low viewership, explaining the underlying factors is mostly creative speculation or, at best -some guesswork.

As discussed in the last article, Artificial Intelligence breaks video content into meaningful entities. These entities, when mapped with viewership metrics, facilitate a richer understanding of audience behaviour.

For both linear and on-demand content, this is achieved by calculating the correlation coefficient between: 

– Share of duration (SOD) of different entities like relationship tracks, storylines, and locations

– Content level success metrics like Rating points, Likes, Views, Time spent per user to name a few

The analyses used to track engagement would be:

  • Point-in-time analysis: This shows which  particular plot or character in a sitcom or a serial creates engagement.
  • Time-bound analysis: The results of this analysis could be around exploring the charm of a specific track in a TV program that starts to wane off, and the viewership reduces considerably over a period of time.

For example, when analyzing a TV series of a client, we observed that a marriage plot that was central to the narrative kept the audience hooked onto their screens. However, after a while, we noticed a dip in viewer engagement. This was an indication that the plot was dragging, and it was time for a course correction in the storyline. For another client, we observed an increase in viewership, when the character of the lead actress suddenly took on a negative arc.

The indication of an approaching popularity spike or slump can help the creative team make the necessary tweaks required to change the track of the story and keep the show’s buzz alive among the viewers. 

Audience Engagement Analysis for News

The viewership analysis framework for News is slightly different from General Entertainment since similar content entities repeat throughout the day across various programs. Therefore instead of analyzing these entities’ impact on specific programs, it is more relevant to study their effects overall. This brings us to the concept of Yield.

Yield (within a given time frame) = SoV/SoD


SoV = % of viewership within a time frame.

SoD = % coverage of Content Entity within that time frame.

A high yield tips the balance in favour of the entity and guides editorial teams for curating similar News. A low yield indicates a need to address content gaps and think of ways to make News engaging.

A comparison of the Yield of two different entities requires adjusting for time slots like prime-time, which generally have high SOV and picking entities with statistically significant coverage.

Target Group/Market-level personalization of insights

Content entity based analysis helps in better personalization of News and entertainment & sports content. Specific target groups and market segments may have different content affinities, and a one-size-fits-all approach may not work.

Looking at content entity based audience insights through the TG/M lens can help in tailoring different strategies around promos, trailers, and highlights depending on the audience demographic.

For example, the affinity of an ongoing character track may vary for two different TGs. Specific couple chemistry may be a hit for GenZ audiences in metros, but not so much for millennials.

But there is a caveat here. Analyzing AI-generated entities at a TG/M level can lead to a proliferation of findings. Hence, they need a subject matter expert to surmise their meaning – based on their statistical and business relevance.

Examining editorial transparency of News

For News broadcasters, maintaining editorial transparency involves finding if due coverage is being done to ‘newsworthy’ content. 

Auditing for transparency involves analyzing stories, personalities, and celebrities present in News and analyzing their coverage with respect to other news channels.

The right way to track it is through Share of duration(SoD). It represents % coverage duration of a news content entity over a certain period. Here content could be a specific news topic, story, genre, or personality. Comparing the SoD for popular stories across different news broadcasters can uncover gaps and biases in their editorial choices.

A recent report by Knight Foundation revealed that most US adults expressed a lack of media news, and over 71% attributed it to transparency in coverage.

News Analysis by Stanford Cable Tv News Analyzer
The prime time coverage of ‘Donald Trump’ by CNN & Fox this Year (Analysis Source: Stanford Cable TV News Analyzer)

Brand Logo Visibility analysis for sports content

Brand visibility analysis involves using AI-generated content entities to measure on-screen brand occurrences on players’ attires and stadium properties in sporting leagues or during major sporting events. The analysis is done based on the number of brand logo appearances as well as the duration of each occurrence. A comparative analysis can also measure the visibility of different brands during a single match.

However, brand exposure and duration is only a quantitative treatment to brand visibility, and often businesses require a higher granularity in the analysis to position their brands for maximum return on investment. For this, brand visibility can also be bolstered based on the match-context in which the brand logo is displayed. 

For example, to find out if the brand logo was displayed during specific high points within a football event, we need to loop in event-based content entities like goals, and penalties. This kind of analysis will give businesses an in-depth report for specific parts of the match.

FIFA 18 Match Analysis, created by AthenasOWl
Caption: FIFA Match Analysis (AthenasOwl)
Identifying the top placement spots like jersey, runner boards on sidelines, behind the goal post.
Determining the visibility during a match high-points and creating a timeline comparison.

Implementing AI-powered Workflows to analyze Content Entities

At AthenasOwl, we work with Media & sports clients to deploy specialized workflows as containerized services in their cloud environment. These workflows, equipped with different computer vision and speech processing models, can be scheduled to get triggered using other event-based triggers. These workflows can also be accessed via REST-API end-points.

The scalable Kubernetes architecture allows the generation of metadata (content entities) for different workloads. Once the metadata is generated and curated, it is pushed into the client’s big data systems to power their analytical workloads. The various content entities are either analyzed standalone for use-cases related to editorial transparency or brand visibility. Or they are fused with the audience data like live viewership, time-shifted viewership, drop-out rate, and positive likes to understand their impact on audience engagement.

In the next article on this series, we would discuss this implementation in further detail. We would also cover some cutting- edge and state-of-the-art techniques for generating these content entities.


Now, we have looked at a decent spectrum of applications across the media and entertainment industry and also discussed the technological feasibility of these applications on a surface level. We can now agree that using ‘content entities’ to analyze audience engagement is a scalable approach i.e. applicable across different genres, platforms, and applications.

At present, viewership analysis procedures involve quantitative surveys & polls followed by macro-approximations of observed audience behaviour. Useful in their own regard, they give the creative and production team very few opportunities to tweak the content for better engagement. 

AI-generated content entities can facilitate micro-decisions. When coupled with modern analytical techniques, it can operate at scale and provide reliable insights into furthermore latent factors.

By doing so, it pushes the boundaries of creative decision making and personalization

Contribution: Sankalp Chaudhary