Strategic Brand Analysis of Big Data Using Latent Dirichlet Allocation

Strategic Brand Analysis of Big Data Using Latent Dirichlet Allocation

Online chatter can be an extremely helpful in source analyzing the thoughts of consumer concerning for example quality, and it is very well possible to analyze possible changes. To do so you need a way to extract this information from the Internet. The article says that it is thus critical to extract key latent dimensions about consumer satisfaction with the quality.

Latent dimensions: variables that are not directly observed but are distracted through mathematical model from other variables that are being observed.

Latent Dirichlet Allocation

This can be done using the Latent Dirichlet Allocation. This model can extract and analyze big data, without too much intervention of humans, and thus saving a lot of time. It is not only very efficient in handling big data, in can also be used to separate time periods with very thin data.

Advantages

• Because it analyzes very specific levels over time, and it thus can show differences in the dynamics over time.

• It also can calculate the importance of the dimensions by the intensity of those dimensions out of the conversations.

• Many steps in the process can be taken without human intervention, thus using as it is called unsupervised methods.

• Because it mostly need no human interaction, less human errors are made and thus it can be done with relatively minimal bias or errors.

• The researcher does not necessarily need to know the dimensions up front.

• The model does also find contextual connections from the dimensions, (since e.g. large is sometimes a negative, and sometimes a negative feature).

The LDA test on online chatter concerning Motorola at the mobile phone market resulted in to six dimensions that were found the most. Of these six 2 dimensions were seen as negative ones, and 4 as being positive ones.

• Instability (negative)

• Discomfort (negative)

• Portability (positive)

• Receptivity (positive)

• Compatibility (positive)

• Secondary features (positive)

These results can be used for brand mapping. Brand mapping illustrates the results, in making specific graphs. It scales the results on two dimensions of quality, thus their relations comes clear.

The study clearly showed that dimensions of quality are very different for different markets, and the contextual connections did as well.

Additionally it showed that multidimensional scaling could be used to capture a brand’s positions on the dimensions. This position changes over time as well as the dimensions themselves, which can increase a brand’ s awareness and is thus very important.

Some differences are found between vertically differentiated markets and horizontally differentiated markets. These are shown in the figure below. Two examples of vertically differentiated markets are mobile phones and computer. Horizontally differentiated markets are for example shoes and toys. Zie bijlage tabel 1.