Combining Behaviors and Demographics to Segment Online Audiences:Experiments with a YouTube Channel
1 like535 views
The document describes research focused on simplifying audience segmentation for social media channels, specifically analyzing a YouTube news channel called AJ+. It details methodologies for behavioral and demographic profiling, leading to a reduction of potential audience segments from 150 to 42 through various heuristics, achieving a 72% simplification. The results support automatic persona generation for improved strategic planning and content creation.
![Data structure
Type Description
Demographic
attributes
ageGroup: YouTube viewers are classified into multiple age
categories (13-17, 18-24, 25-34, 35-44, 45-54, 55-64, and 65
years and older); 7 possible age categories for a customer.
gender: YouTube viewers are classified as either male or
female, so there are 2 possible categories.
country: YouTube uses the two-letter ISO-3166-1 country
code index to classify where viewers are from, with 249
current officially assigned country codes at the time of this
study.
Behavioral
attribute
viewCount: YouTube provides the number of views per
video for a given [country] by [gender, ageGroup].](https://image.slidesharecdn.com/insci18customersegmentation-181024164649/85/Combining-Behaviors-and-Demographics-to-Segment-Online-Audiences-Experiments-with-a-YouTube-Channel-6-320.jpg)
![Matrix decomposition
For each content piece and
aggregated user group, we
retrieve a value of the chosen
interaction metric. Then we
decompose this matrix V into W
and H in regards to p, which is
the number of segments [1].
[1] An, J., Kwak, H., & Jansen, B. J. (2017). Personas for
Content Creators via Decomposed Aggregate Audience
Statistics. In Proceedings of Advances in Social Network
Analysis and Mining (ASONAM 2017). Sydney, Australia.](https://image.slidesharecdn.com/insci18customersegmentation-181024164649/85/Combining-Behaviors-and-Demographics-to-Segment-Online-Audiences-Experiments-with-a-YouTube-Channel-7-320.jpg)
Combining Behaviors and Demographics to Segment Online Audiences:Experiments with a YouTube Channel
- 1. Bernard J. Jansen1, Soon-gyo Jung1, Joni Salminen1,2, Jisun An1, Haewoon Kwak1 1Hamad Bin Khalifa University, Doha, Qatar 2 University of Turku, Turku, Finland bjansen@hbku.edu.qa, sjung@hbku.edu.qa, jsalminen@hbku.edu.qa, jan@hbku.edu.qa, hkwak@hbku.edu.qa 5th International conference ‘Internet Science’ (INSCI'2018) St. Petersburg, Russia, October 24 to 26, 2018. Combining Behaviors and Demographics to Segment Online Audiences: Experiments with a YouTube Channel
- 2. Problem Read also: Salminen, J., Milenković, M. and Jansen, B.J. (2017), “Problems of Data Science in Organizations: An Explorative Qualitative Analysis of Business Professionals’ Concerns”, Proceedings of International Conference on Electronic Business (ICEB 2017), Dubai.
- 3. Problem • Social media channels with audiences in the millions are increasingly common. • Efforts at segmenting audiences for populations of these sizes can result in hundreds of audience segments, as the compositions of the overall audiences tend to be complex. • Although understanding audience segments is important for strategic planning, tactical decision making, and content creation, it is unrealistic for human decision makers to effectively utilize hundreds of audience segments in these tasks.
- 4. Problem • In this research, we present efforts at simplifying the segmentation of audience populations to increase their practical utility. • Using millions of interactions with hundreds of thousands of viewers with an organization’s online content collection, we first conduct behavioral profiling, and then demographic profiling. …the outcomes are implemented in real online system for automatic persona generation.
- 5. Data collection • Our data source is the online news channel AJ+ , which was designed from its founding to serve news in the medium of the viewer, with no redirect to a website or other platform. • For this research, we use the AJ+ YouTube Channel. However, the technique presented here is generalizable to any social media channel providing aggregated audience statistics. • We collect data on 4,320 YouTube videos produced from June 13, 2014 to July 27, 2016. • Collectively, these videos have had more than 30 million views from people in 190 countries at the time of the data collection. • The YouTube analytics platform provides, for each video, view counts by user group (age, gender, country). • We access the data via the YouTube application programming interface (API), through which data can be collected automatically if permission is granted by the channel owner.
- 6. Data structure Type Description Demographic attributes ageGroup: YouTube viewers are classified into multiple age categories (13-17, 18-24, 25-34, 35-44, 45-54, 55-64, and 65 years and older); 7 possible age categories for a customer. gender: YouTube viewers are classified as either male or female, so there are 2 possible categories. country: YouTube uses the two-letter ISO-3166-1 country code index to classify where viewers are from, with 249 current officially assigned country codes at the time of this study. Behavioral attribute viewCount: YouTube provides the number of views per video for a given [country] by [gender, ageGroup].
- 7. Matrix decomposition For each content piece and aggregated user group, we retrieve a value of the chosen interaction metric. Then we decompose this matrix V into W and H in regards to p, which is the number of segments [1]. [1] An, J., Kwak, H., & Jansen, B. J. (2017). Personas for Content Creators via Decomposed Aggregate Audience Statistics. In Proceedings of Advances in Social Network Analysis and Mining (ASONAM 2017). Sydney, Australia.
- 8. Overview of segmentation • We decompose (i.e., separate into simpler components) the matrix V into two matrices: W and H. • Matrix H contains the audience behaviors and matrix W contains the user demographics. • Matrix H encodes an association between the behavioral segments and the individual videos. Once we have the matrix H, we discover, via NMF, the latent patterns describing the user segment interactions with the videos. • The matrix W encodes an association between user demographic groups and behavioral user segments with the set of corresponding user demographic segments each with an associated weight indicating how strongly each demographic segment is associated with the given behavioral segment. • Each row in W represents how each demographic segment can be characterized by different behavioral patterns. The columns in W show how strongly each latent behavioral pattern is associated with different demographic segments.
- 9. Example User Behavioral Segment #1 (B1) Seg. Country Age Gender Weight z-score D1 US 25 male 415.73 2.330935 D2 US 35 male 313.00 1.215947 D3 US 45 male 221.56 0.223497 D4 CA 25 male 216.42 0.167709 D5 CA 35 male 176.54 -0.26513 D6 US 55 male 165.83 -0.38137 D7 GB 25 male 147.47 -0.58064 D8 CA 45 male 124.21 -0.8331 D9 US 65 male 114.93 -0.93382 D10 GB 35 male 113.99 -0.94402
- 10. Simplification heuristics • For this research, we use 15 behavioral segments, each associated with 10 demographic segments. Therefore, we have a total of 150 potential segments, which is the number we aim to reduce. • Using elbow and z-score approaches, we reduce our entire data set of 150 audience segments to 63 segments (a 58% reduction).
- 11. Simplification heuristics • For this research, we use 15 behavioral segments, each associated with 10 demographic segments. Therefore, we have a total of 150 potential segments, which is the number we aim to reduce. • Using elbow and z-score approaches, we reduce our entire data set of 150 audience segments to 63 segments (a 58% reduction). • Country heuristic: If two or more demographic segments are from the same country, then those audience segments are candidates for consolidation • Gender heuristic: If demographic segments are of the same gender from the same country, those segments are candidates for consolidation • Age heuristic: (a) if the age bracket is bounded by corresponding age brackets from the same country and gender, then the age brackets are aggregated, or (b) if the age bracket is adjacent to a corresponding age bracket from the same country and gender, then the two age brackets are aggregated.
- 12. Simplification heuristics • For this research, we use 15 behavioral segments, each associated with 10 demographic segments. Therefore, we have a total of 150 potential segments, which is the number we aim to reduce. • Using elbow and z-score approaches, we reduce our entire data set of 150 audience segments to 63 segments (a 58% reduction). • Country heuristic: If two or more demographic segments are from the same country, then those audience segments are candidates for consolidation • Gender heuristic: If demographic segments are of the same gender from the same country, those segments are candidates for consolidation • Age heuristic: (a) if the age bracket is bounded by corresponding age brackets from the same country and gender, then the age brackets are aggregated, or (b) if the age bracket is adjacent to a corresponding age bracket from the same country and gender, then the two age brackets are aggregated. • When applying the country, gender, and age consolidations to each of the 15 behavioral segments, our original 150 integrated audience segments are reduced to 42 segments, a 72% reduction.
- 13. Advantages 1. Simplifies behaviorally and demographically complex customer base: With the example dataset, we achieved a 72% simplification of the segments 2. Demographic groups can belong to many behavioral groups (unlike in clustering) 3. Behavior is the basis for segmentation (as recommended in state-of-the-art marketing theory) 4. The approach is generalizable to any customer data with age, gender, location and interaction metric (e.g., social media data, in-house data)
- 14. How do we actually utilize the results of this research?
- 15. Automatic Persona Generation (APG)
- 16. See the personas Persona Listing based on selections. Clicking on persona image displays persona profile.
- 17. Learn about the personas Persona Profile, with image, name, attributes, topics, most interested content, audience size, etc.
- 18. View changes in personas over time Shows the similarities and changes in the persona set over time.
- 19. View how personas interact with content Analyze how well your content interests different segments.
- 20. Identify market opportunities Identify gaps in current and potential audience. The share of our audience is like this The share of total audience in the world is like this
- 21. Configuration Collection Generation A matrix of content interaction patterns Automatically generated personas Collection/Generation/API information See: http://persona.qcri.org Read: Jung, S., Salminen, J., An, J., Kwak, H. and Jansen, B.J. (2018), “Automatically Conceptualizing Social Media Analytics Data via Personas”, presented at the International AAAI Conference on Web and Social Media (ICWSM 2018), San Francisco, California, USA.
- 22. Thank you!