In these times of data overload, businesses are facing a data explosion as enormous volumes of consumer-generated content vie for organizational attention. Concepts of knowledge discovery and datamining, when applied to this voluminous data, can harness it for business intelligence, thus contributing to agile intelligent enterprises.
I experiment with discovering actionable knowledge from a Corporate Blog, to aid decision making. I treat the ‘social’ aspect of a blog as an attribute which enables it to serve as an application which can form a bridge between the consumer and the organization.
Customer Intelligence Tool for CRM
Further to my article on Blogs as marketing campaigns, where I had considered comments as sets of opinionated text, with the assumption that the text (each set of comments on a single post) is related to a single issue or item, I propose a system of allowing consumers who comment on an organizational blog, to tag their comments.
Tags are created with the intent of signifying or suggesting concepts that are potentially or accompanying or associated with possible content ontologies and are inherently connotative, social and possibly democratic. If consumers can be allowed to tag their comments, a better aggregation of user generated information can be achieved to serve as an aid in response modeling. This requires consumers to associate keywords with content signifying their intent.
The best judge of classification of one’s comment under a blog post would be the consumer himself, who can tag his comment with the closest possible option available from an organization specified list. Tags are inexpensive, scalable and very near the language and mental mode of the users. If consumers tag the comments themselves, relevance and consistency of the relation of the content in the comment to the tag will be greater. Problems pertaining to lack of terminological control in the tags can be omitted by allowing consumers to choose from a predefined list only. This way, we could have a tag to represent each separate comment typology.
The content categorization is taken further, by performing a factor analysis (based on responses from a set of consumers) on types of comments made by consumers on several blog posts. This helped identify the various comment typologies, which loaded onto the factors of Consumer “Liking,” “Satisfaction” and “Involvement I, Involvement II, III and IV.” This can help form groups of tags thereby creating folksonomies related to consumer liking, satisfaction and involvement. As it is in the organizational interest to keep track of respective pipelines generated and assimilate feedback, while also taking care of consumer redressal, data redirection to respective CRM functionality can be initiated.
For each campaign the dominant consumer viewpoint can further be assessed by analyzing the frequency of consumer comments under each tag. For the purpose of this discussion, we term this tag frequency (TF). The tag frequencies of individual campaigns across the separate comment typologies can be determined. Further, a quantitative tag analysis aids campaign optimization.
Substantial organizational decisions are based on experience. When faced with new situations, companies are guided by memories of similar situations they have experienced in the past. Using this as basis, I make use of a set of mining techniques.
I try to identify situations similar to each other by calculating similarity of consumer responses between different campaigns.
The TF measure can be used to determine similarity of results (consumer thought process as represented by the comments under the respective tags) between two blog posts, which can be considered as individual campaigns.
If we compare the Tag Frequency vectors across individual campaigns (posts) by using a similarity measure, we can evaluate the number of comments (representing volume of customers) in a particular tag category. Hence, by comparing volumes of consumers depicting certain kinds of sentiments across various campaigns, prediction of consumer behavior can be done. By using the tag frequency vectors of two campaigns, similarity methods can be used to normalize the no. of comments in a post.Determining similarity between the responses to two campaigns can help organizations predict consumer responses and improve targeting of future campaigns. If consumer responses to my current campaign resemble consumer responses to a campaign organized an year back, i sure have some interesting information to fall back on and plan my strategies accordingly.
Data Redirection to respective CRM functionalities
All comments tagged under the tag cluster visible under the folksonomies of liking and satisfaction can be diverted for sentiment mining to aid consumer segmentation. All comments under the tag cluster representing the folksonomy of Involvement I and II are routed to the ‘Customer Service and Support’ functionality. All comments under the tag representing the folksonomy of Involvement III are directed to the ‘Marketing Communication ‘function and all comments under the tags clustered under Involvement IV are sent to the ‘Product development’ function.
Corporates can invest in creating more organization friendly retrieval measures like developing interfaces for respective campaigns for retrieval of individual comments viz. comments with sales leads, product suggestion etc.
Consumers can be clustered on the basis of sentiment scores. (Calculation of Sentiment scores has been discussed in the earlier post on Sentiment mining). To demonstrate an example, the sentiment scores of a set of consumers from a blog post of an airline company, were subjected to cluster analysis, whereby using the nearest neighbor technique, consumer clusters can be formed. The resulting dendrogram brought out clusters of consumers.
While Aaron, Bill and Bob belong to one cluster, all other remaining members except Drew fell under one cluster. Drew formed the third cluster.
This can form the basis of consumer profiling as and when a new consumer enters the system. Consumer profiling can help in customer response prediction. The new customer joining a cluster is likely to respond to an offer in a particular way that is probably similar to the previous customers who have responded.
By allocating the consumer to the respective cluster, and hence identifying the next likely customer to behave in a particular way, behavioral traits and prospective consumer targeting strategies can be formulated for better success rates from the viewpoint of Marketing and CRM.