Sowing the Seeds of Word-of-Mouth
Uncovering the Importance of Relationship Characteristics in Social Networks: Implications for Seeding Strategies
CHEN, Xi | van der LANS, Ralf | PHAN, Tuan Q
Journal of Marketing Research, Vol. LIV, April 2017
When it comes to launching new products, major firms such as Philips and Microsoft have adopted seeding strategies that target influential customers in social networks so they can spread the word to their friends, colleagues and acquaintances. The difficulty, of course, is how to identify such people and assess their reach.
Traditionally, researchers have looked for certain characteristics and positions of seeds based on pre-conceived theories, such as the number of friends of a seed. But a new study takes a different approach. It builds a model based on the characteristics of their relationships, such as type (for instance, friend, neighbor or colleague), duration of the relationship and the interaction intensity, and weights these to assess the reach of each seed. Moreover, information is also collected during the actual diffusion process to get a more representative result.
The authors, Xi Chen, Ralf van der Lans, and Tuan Q Phan, also test out their method in two very different settings – a microfinance program across 43 villages in India and the propagation of comments on Super Bowl advertisements in a social network of more than 42,000 users. They find their method can increase the reach of seeding strategies in the first case by up to 10 per cent and in the second by up to 92 per cent.
In the microfinance study, the authors drew on social network data about the villages that had been collected previously and identified 12 types of relationships that they characterized along four dimensions: economic (borrowing or lending money or goods), social, religious and family. They also looked at the distance others lived from the seed, among other inputs. This data was then computed into two measures: weighted degree centrality, which is the weighted sum of connections of each seed and assumes only direct contact with those connections, and weighted eigenvector centrality, which takes into account not only the connections of seeds, but also the connections of connections, etc.
The eigenvector centrality is usually regarded as a more convincing measure, but the authors showed that weighted degree centrality was a much better fit. When they tested it on a hold-out sample, they had a 10 per cent wider reach. Moreover, that improvement was obtained with a relatively small proportion of seeds, which should be of interest to marketers.
In the second study involving the Super Bowl, the authors had access to anonymized data for the 42,000 users of a social network at a large American University, including age, gender, date of joining and online behavior of each student such as messaging, public posts and comments. The authors studied their response to advertisements that premiered during the 2010 Super Bowl to see how information cascaded through the network.
They identified 1,620 seeds, characterized their relationships based on the number of messages exchanged with others in the previous two months and the duration of their connection to the network, then looked at how many of their messages about the advertisement were forwarded or commented on by their friends. The number of messages exchanged was positively related to social influence in the network, with longer relationships associated with weaker influence.
A hold-out sample was tested on 50 seeds based on degree centrality and the result was striking. With weighted degree centrality, their reach was nearly 92 per cent higher than if they had been selected on degree centrality alone and 31 per cent higher than eigenvector centrality.
“Our proposed multi-network approach demonstrates that the importance of relationship characteristics substantially varies. In our empirical applications we find that recognizing these differences not only results in a better statistical fit, but also leads to better seeding strategies,” they said.
“Our methodology is flexible and suitable for any diffusion process in which social network data are available, and we believe it may be a valuable tool for managers to optimize seeding strategies in order to facilitate the diffusions of their products and services.”