AI Can "Find 'Em All": A Case Study on Finding Social Media Influencers for Brands

16 January 2019

The past decade has testified the rapid growth of the Internet. One can observe the drastic expansion of social networking services, where millions of users publish and consume information regularly. Built upon such growth, the social media marketing industry has correspondingly developed its capabilities of helping marketers in content personalization and deliverance. However, the growing amount of irrelevant content, such as unrelated advertisements and spam, made social media users more and more reluctant towards perceiving sponsored search results and online advertisements, such as "Google AdWords'' and "Facebook Sponsored Ads''...

The past decade has testified the rapid growth of the Internet. One can observe the drastic expansion of social networking services, where millions of users publish and consume information regularly. Built upon such growth, the social media marketing industry has correspondingly developed its capabilities of helping marketers in content personalization and deliverance. However, the growing amount of irrelevant content, such as unrelated advertisements and spam, made social media users more and more reluctant towards perceiving sponsored search results and online advertisements, such as "Google AdWords'' and "Facebook Sponsored Ads''.

To mitigate such customer skepticism, marketers often leverage human-centric content delivery channels, where Influencer Marketing clearly dominates over other marketing strategies. Indeed, it has been shown that 92% of consumers are more likely to trust brands that advertise via influencer channels rather than those who have adopted conventional marketing strategies. Unfortunately, the limited availability of influencer search platforms and the absence of audience-based and content-based influencer matching technology, result in tremendous amounts of manual work performed, the corresponding high marketing agency service costs and low efficiency of the conducted marketing campaigns.

AI Can "Find 'Em All"

In a brave attempt to bridge such a huge gap, we were selected among hundreds of candidates to showcase our new-developed technology at World's Top Multimedia AI conference - ACM MULTIMEDIA 2018. What will be shown is our online influencer discovery platform that combines advanced data analytics and behavioral science to help marketers find, understand their audience and engage the most relevant social media micro-influencers on a large scale. SoMin harvests brand-specific life social multimedia streams in a specified market domain, followed by rich analytics and semantic-based influencer search. The Individual User Profiling models extrapolate the key personal characteristics of the brand audience, while the influencer retrieval engine reveals the semantically-matching social media influencers to the platform users. The influencers are matched in terms of both their-posted content and social media audiences, while the evaluation results demonstrate the excellent performance of the proposed recommender framework.

What Does It Mean for Marketers?

Using SoMin.ai web portal, marketers will be able to specify textual and image queries. The influencer matching response will be given in a form of a ranked list of top-matching influencers, where each influencer will be listed together with a brief description of his/her audience and the corresponding audience and content relevance score. By selecting influencer from the list, users will be routed to the "Influencer Details" dashboard (e.g. see an example of @modgam above) that visualizes influencer's audience and recently-posted content in terms of Hot Topics, Named Entities, Behavioural Attributes, Image Concepts, and Sentiments of their audience. Finally, SoMin users will be able to specify various filters for tailoring Social Influencer matching engines for their specific needs. For example, the platform could be set up to output only those influencers who are active in Singapore, do not represent any brand at the moment, and did not advertise for competitive brands in the past.

Does It Work?

To evaluate the performance of our influencer recommendation engine, we have conducted a case study on real-life data collected for brands operating in the restaurant industry. We have chosen the top 20 messages posted by one of the brands on their Twitter timeline as a test query for the platform. Three independent annotators (professional marketers) were asked to annotate the recommendation results obtained for the 2$ queries to the platform as "relevant" or "non-relevant". "Relevance'' was defined as the similarity of the content in the query and the recent influencer-posted content as well as the potential match of the chosen Brand's customer segment (i.e. in our case, ``Masculine, Mature, Average Income, Non-Logical, Principal, Idealists'') to the Influencer's follower audience. To gain an insight into the quality of SoMin Influencer Recommendation, we used "Precision at K" (P@K) metric, which is the portion of relevant documents among the top K recommended documents.

The evaluation was done for different values of K and suggests that the Recommender System effectively solves the problem of Influencer Recommendation based on customer audience and intended marketing content simultaneously. We also found that the highest performance was achieved for K = 5, while for the case of K = 10, the recommendation quality is comparably lower. The finding can be explained by the relevance of the content in the performed queries: some query messages (e.g. bottles with liqueurs posted on the Twitter wall) were not directly related to the brand's market domain (i.e. Steaks and Dining), which, in turn, might bring less relevant recommendation results at the tail of the recommendation list (e.g. Alcohol Drinks-related Social Media Influencers). Overall, we would like to highlight the high recommendation quality achieved for all the examined values of K, which allows for the successful use of SoMin platform for the preparation and execution of the real-world influencer marketing campaigns.

Key Takeaways

To conclude, our main aim was to help marketers find, understand their audience and engage the most relevant social media micro-influencers at a large scale. Experimental results demonstrated that SoMin is capable to achieve the goal by showing excellent recommendation performance when matching influencers in terms of both their posted content and follower audiences. For marketers, our results are one more step towards the full understanding of the true potential that opens when they apply AI to their daily marketing routines. To us, it is one of those moments when you understand that you are crafting the future right here right now.

Original Article

The full version of the article will be released after official publication date in ACM MULTIMEDIA 2018 Proceedings:

A. Farseev, K. Lepikhin, H. Schwartz, E. K. Ang, K. Powar. SoMin.ai: Social Multimedia Influencer Discovery Marketplace ACM Multimedia Conference 2018, Oct. 22 - 26, 2018.

ABOUT THE AUTHOR

Prof. Aleks Farseev PhD

Aleks Farseev is a machine learning wizard who can teach a computer to sing "Bohemian Rhapsody" in binary code. He loves conjuring up new creations and is on a quest to figure out how machine learning can make the world a better place. When not tinkering with technology, Aleks can be found serenading his friends with his accordion skills, which he claims are only slightly less impressive than his machine learning prowess.

Want to have a chat about our products?

We're always happy to talk about how our tools can help people.

Book a Demo