Social media has proliferated the online advertising space with its revenue set to grow from 5.1B USD in 2013 to 15B USD in 2018, with a YoY growth rate of about 24%  that assists advertisers to listen to, analyze, and interact with customers on social media. With this backdrop, this case-study introduces several major considerations in addressing this need and some broad areas that we can attempt to work on, in this area.
Social media marketing and advertising is an emerging and rapidly growing realm of business as reported by several market research companies [BIA/Kelsey, 2014, Business Insider, 2014, Forbes, 2014, CMO Council, 2015, IAB, 2012]. In this document, we discuss various aspects of implementing systems that can process data from social media for the aforementioned task of social media marketing and advertising (SMMA).
The primary objective of SMMA is to bring a user down a “conversion funnel” – which is illustrated in figure 1. Different SMMA activities place themselves at different parts of these funnels and attempt to drive a customer from say, creating awareness, to getting him to buy a product or use a service. The final step in the funnel is to retain brand-loyalty and to establish an ongoing relationship, so that customer churn is minimized.
Working with social media data
The three main aspects of dealing with data from social media are:
- Listening – Identification of relevant content in social media
- Engagement – Conversing and/or displaying advertisements to a user on social media in a targeted fashion. The purpose of engagement has to be clearly defined – it could be for the purposes of: brand awareness, customer support, sales, and information gathering.
- Measurement – Engagements that take place on social media, either done by the advertiser/marketer or by users have to measured for their impact with respect to a brand, competitors, and the like. Further, prior to an engagement, measurement is also necessary to identify the right users to interact with, or display advertisements to.
Products available in the market fall into the above categories and may span more than one.
Key technological challenges
The following questions have to be clearly addressed, to design systems that can work with social media data.
• Scalability Data from social media websites like Facebook, Twitter, and Google+ grow at several GBs to TBs a day. This calls for infrastructure that can process this massive scale of data.
• Data heterogeneity Each social media website has its own “theme”, user experience, and objective. Permissions to access various parts of a “social media profile” varies by the social media website too. This results in a different data “schema” for each website. Therefore, it is necessary to normalize data across sources while not removing data that is characteristic of a source..
• Low Signal-to-noise ratio Given the large quantity of data in social media, the required “signals” for an advertiser or a marketer is not always visible due to a plethora of non-relevant conversations. Therefore, the data has to be carefully sifted to identify conversations and profiles (and parts of it) that are relevant. This is akin to identifying the needle in a haystack! Keyword filtering, bot identification, sentiment and intent analysis, influencer identification, etc. are used for this purpose.
• Non-standard languages, code-mixing, etc. Each social media website carries with it, its own “lingo” that includes new words, abbreviations, code-mixing/switching (i.e. mixing of languages), and the use of special symbols to represent emotions, etc. Standard natural language processing techniques have to be revisited, to evaluate assumptions. For example, named entity recognition, entity disambiguation, POS tagging, sentiment analysis, and the like, cannot be applied in the same form as it is applied to other forms of written text.
The following steps elucidate the process of identifying areas of improvement with respect to SMMA for an organization, and some pointers to how this can be solved with state-of-art systems.
Scoping the problem
First, one has to clearly identify the people we would want to target for an activity. They could fall into different sections of the “conversion funnel”. This would require the organization to define what the outcome of the effort should be. It could be one of the following:
- Brand awareness
- Customer support
- Conflict resolution and reputation management
Additionally, the organization may also want, as outcome, intelligence gathered about the type of audience out there, how the competitors are doing, and the like. This is more of an exploratory activity. Once the outcomes have been defined, the organization has to define ways of measuring the outcomes. Table 1, as an example, briefly enlists various metrics that can be used for this.
|Brand Awareness||Reach, Cost-per-Million impressions and its variants|
|Sales||Cost-per-acquisition, Cost-per-click and its variants|
|Customer Support||Sentiment of customer support conversations, # queries
resolved, Avg. time for resolution
|Conflict Resolution||Sentiment histogram, # conflicts resolved, # conflicts identified, reputation quality of the org vis-a-vis competitors|
|Info. gathering||Sentiment, Intent, and topic distributions; influencer lists with scores, etc.|
|Table 1: Measurement of outcomes for an SMMA activity.|
Next, given the large number of sources of information available today, one has to decide the sources to use. For example, the organization, through an initial study of major social media platforms, may identify that Twitter and Facebook cover a large section of its customer base.
Once these parameters have been pinned down, we can proceed to implementing a solution; this is covered in the next section.
The most common problem buckets that one could work with are:
- Sentiment and intent analysis
- Influencer identification
- Topic identification and content clustering
- User segmentation
- Demographics and user attribute identification
- Automated campaign optimization
General guidelines to arriving at a “Social Media” solution
- Scoping the problem. The previous section can be used as a guideline for this. This will enable us to pin down specifications of the problem in a form that can evolve into a “process” or a “system”. This typically requires cross-domain expertise, from business and technological perspectives.
- Building ingestion infrastructure. Once the requirements are clear, data acquisition and ingestion are important steps that might include both non-technical (e.g. compliance) and technical issues (schema design, etc.) that require to be resolved.
- Data (Pre-)Processing. Pre-processing during/after ingestion may also require us to implement various “Data Science” models that can identify different aspects of data like demographics, topic names, sentiment, intent, etc. This may warrant the use of Machine Learning. To validate models built using Machine learning, some labelled data is required. When this is not available, we may have to collect it and also build or use existing systems to get the data labelled by human labellers when additional efforts are required.
- Analytics and measurement dashboards. With the acquisition of the right data, visualization of the data in the right form so that it can help drive decisions in an organization is key.
- Engagement and campaign management dashboards. If the solution involves interaction with users on social media, displaying advertisements, collecting labelled data, and integrating with CRMs and operations workflows within the organization forms an additional step in the overall solution.
 Source: BIA/Kelsey; May-2014
[BIA/Kelsey, 2014] BIA/Kelsey (2014). U.S. Social Media Advertising Revenues to Reach $15B by 2018. Press release, accessed on Feb 28, 2015. http://www.biakelsey.com/Company/Press-Releases/140515-U.S.-Social-Media-Advertising-Revenues-to-Reach-\$15B-by-2018.asp.
[Business Insider, 2014] Mark Hoelzel/BusinessInsider (2014). SOCIAL-MEDIA ADVERTISING: The Rush Into Social Is On, Led By Spending On Mobile And Programmatic. September 28, 2014.
[Forbes, 2014] Kyle Wong/Forbes (2014). The Explosive Growth Of Influencer Marketing And What It Means For You. September 10, 2014.
[CMO Council, 2015] CMOCouncil (2014). Marketing Spend (Facts & Stats). http://www.cmocouncil.org/facts-stats-categories.php?view=all&category=marketing-spend. February 27, 2015.
[IAB, 2012] JEGI & IAB (2012). The social media ecosystem report: Rise of users, intelligence, and operating systems. http://www.iab.net/media/file/JEGIIABSocialMediaReport.pdf.