Social listening analytics: WOM vs. WOM marketing

  • Facebook
  • Twitter
  • Linkedin
  • Email

May 9, 2016 - Social listening is a popular source of information about people's opinions on businesses, brands, and products. Different forms of marketing create awareness. Knowledge and experience create initial opinions. And then, opinions are spreading around the world with Word of Mouth (WOM) through (with an increasing portion of digital WOM) social networks, blogs, tweets and forums. Digital WOM can be considered from two sides:

  • It is a reflection of opinion distributions, and, at the same time
  • It is an influential driver to form and reorganize the same opinion distributions.

Thus, WOM is a self-organizing feature of people's opinions: opinions with a higher popularity have a proportionally higher number of published opinions (tweets, comments, blogs, etc.), and, therefore, attract more followers. Then, due to new followers, such opinions become even more popular and attract even more new followers. This is a cycle with positive (reinforcing) loop. In an ideal world it would lead to a single opinion. However, due to many real-life factors like different forms of marketing, diffusion of opinions, memory, opinion inertia, etc., this will never happen. For a reflection of all such factors in a math (or sociophysics) form you can read my paper [Dmitri V. Kuznetsov, “1-to-1 personalized consumer-product marketing in real-life environment with critical word-of-mouth (WOM) impacts," Model Assisted Statistics and Applications 4 (2009) 159-169].

WOM as an influential driver. Because of the two sides of WOM (reflection of opinions distributions and influential driver), there is appeared a relatively new form of marketing: WOM Marketing (WOMM). Such marketing can make corrections in the self-organizing loop above in any direction. While it is difficult to truly control WOM, there are a few common powerful ways to manage WOM by WOMM:

  • Build social opinion foundation, e.g. satisfaction, trust and commitment;
  • Create active online social interactions like referral programs, membership clubs, teaser campaigns, etc.
  • Pay WOM agents, i.e., paid opinions/ads in a form of WOM.

The last one works because people usually believe there are a “fair-play" published opinions instead paid ones, especially if they came from a major influencer, a blogger with a lot of readers/followers. At the same time, a pro-consumer WOM is a counterweight to commercially-motivated WOM.

WOM as a reflection of opinion distributions. Indeed, WOM is a valuable source of business information about people's opinions around brands and products. However, “fair-play" and paid parts of WOM are usually mixed. For business intelligence, both these parts may contain useful information. But information in these two parts may have different business meaning, even if they have almost the same content. So, a reasonable separation of the components would be beneficial. (In the appendix below we describe potential ways to separate them.)

Where is the information you are looking for? Let us consider briefly the following table to connect different parts of social listening information with the corresponding business-related factors.

Influential Driver: Almost all brand- or product-related social publications can influence readers. Maybe the only exclusion is explicit “garbage," e.g., requests to Retweet/Share/Follow, or a content with specific URL links when content and link are sometimes even not connected, etc. However, not all publications have the same influential effect. It depends on (usually proportional to) number of readers of a publication and level of trust for the publisher. Number of readers is proportional to the size of audience the publication is addressed to (public/followers, friends only, etc.)

Marketing: It naturally includes all three parts of WOMM, where paid WOM is usually not completely separated from “Fair-Play" WOM and available as uncleansed WOM content.

Opinions: “Fair-Play" WOM is a condensed form of real-people opinions that heavily mixed with paid WOM. The last (unwanted here) part can contribute over 80% in the mixture that can make social listening information strongly biased to the marketing content with forcing opinions. However, these opinions and their distribution between positive and negative polarities are not always reflecting distribution of people between positive and negative opinions, because numbers of publications from different people are significantly different and, thus, biased to the opinions of people with higher number of publications.

People by Opinions: “Fair-Play" WOM deduplicated by user can deliver the distribution of people between positive and negative opinions. In the case of many different opinions from the same user, we can consider either the most recent one or an average value for a period of time. Distribution of people between opinions can be very important for marketing (buyers and prospects), customer services (satisfaction rate), politics (voters), etc.

Thus, selection of an appropriate way for data preparation depends on particular business objectives from social listening.

Appendix: Separation of WOM types. It is not possible to completely separate “Fair-Play" WOM from Paid WOM (as a kind of spam) because in many cases they look very similar. However, we can significantly increase the portion of “Fair-Play" WOM by applying a rules-based cleansing or machine learning procedures. In particular, the rules can include recommendations to mark as Paid WOM content from: bot-friendly services, recently created users, not-verified users, users that create little content, and users with few followers; or content with the most popular hashtags or too many hashtags to attract attention. “Fair-Play" garbage can include content requesting Retweets and Follows. Moreover, unless they are re-tweets or shares, almost exactly the same messages, only with different URL addresses, look very suspicious and requires attention. They are, most likely, spam from a bot. In some instances, such suspicious tweets can reach up to 80% of the total number of tweets in a sample.

Author: Dmitri V. Kuznetsov, Ph.D. - Chief Science Officer – Specialty & Complex Analytics, Data Science & Sociophysics