Did you know you can use Luminoso Analytics outputs with Tableau? If you’ve got Tableau Desktop, you can use Luminoso Analytics as a data source for building Tableau dashboards that fit nicely within your company’s data ecosystem. Using Luminoso as a data source for Tableau allows for augmenting Luminoso’s very powerful conceptual understanding of text analytics data with existing business metadata, grounding the insights from natural language text to real business concerns relevant to executives and non-analyst roles within an organization.
At Luminoso, we build software to help people understand and analyze large volumes of natural language text. Whether from a conversation, published content, or focused responses, Luminoso Analytics automatically discovers the major concepts and their relationships, revealing the underlying structure of what is being discussed or conveyed.
When you first import your data to Luminoso Analytics, your data is processed and presented visually as an interactive word cloud that allows you to explore the major concepts and their relationships. This process is entirely automatic, without the need to spend hours or days pre-defining complex, industry-specific ontologies or keyword queries.
Interacting with Luminoso Analytics in this way allows one to uncover new insights and raise new questions, which are then often answered in real-time within the same interface. The velocity of this loop — observing a relationship, generating a hypothesis, and then corroborating or disproving it all within the same interface — allows a single analyst to gain a better understanding of the data in an hour than a team could accomplish in a much longer time without this tool.
Yet not everyone in your organization is either an analyst or has time to explore data in this unconstrained, exploratory manner. Further, there may be other data sources that would be useful to see alongside Luminoso Analytics results. This is where it makes sense to add Luminoso Analytics data to Tableau.
Take the example of a fictional company Acme Sporting Goods that’s collected tens of thousands of customer feedback responses about their retail stores, solicited after a purchase. The data used in this example is real-world data with names and other sensitive information redacted for the purposes of this post.
The metadata associated with these responses includes the dollar value of the purchase, the location of the store, the date, and a net promoter score (“how likely are you to recommend Acme Sporting Goods to a friend?”). Upon uploading the responses to Luminoso, we begin by exploring the word cloud and discovering important concepts and their relationships. In this interactive exploration stage, we’ll create some topics of interest which we can use to assess the strength of relationship to other topics and subsets of the data. We can think of these topics as lenses that we can look through to get a unique view of our data. When we select a given topic, our view of the data set is in terms of the strength of association with that topic.
Here, based on a quick survey of the word cloud and a few verbatims, we’ve constructed several topics to get at the semantic concepts of: customer service, positive and negative sentiment, various products, and pricing. The technology behind Luminoso Analytics also takes care of the non-standard ways that people talk about these concepts, e.g. domain-specific terms or slang terms like “bucks” to refer to price, are automatically recognized as being conceptually similar to our custom topics.
Tableau provides a good balance between the open-ended nature of the word cloud and a static report. Running a simple script, we use the Luminoso API to export data from the system into a format that can be easily imported to Tableau. In this example, each response is a row in the original data set. We’ve exported the strength of relationship between each response and each of our topics, adding a new column for each topic.
Importing this data, we can construct a dashboard that allows Tableau users to explore the prevalence of our topics of interest, among subsets of the data including location, net promoter score, purchase price, etc.
To begin, we tell Tableau to shade a map (using the State metadata for each response) based on the strength of relationship between a the Luminoso topic for overall negative sentiment and the responses from each state.
Glancing at this map, we can see that South Carolina, Montana, and Minnesota stand out as having responses with the strongest relationship to our negative sentiment topic. This is useful at a high level, but we’d like to give the user a little more insight in to time trends and a few verbatim responses to make it concrete.
We’ll do this by creating a new sheet containing a list of the verbatims, sorted by strength of relationship to this negative sentiment topic, and filterable by clicking on a state. We’ll also create a new sheet with a line chart showing the time trend of the topic. While we’re at it, we’ll also add several other filters using the rest of the available metadata.
Here we’ve combined three simple Tableau sheets into a dashboard that allows filtering by clicking on a state. On the above sheet, we noticed that South Carolina was shaded to indicate relatively high negative sentiment. Now we can click on South Carolina to see the trend of negative sentiment over time and the verbatim responses to discover negative drivers that Acme Sporting Goods can act upon.
We’ve also added a drop-down in the upper right, using a parameter and a custom field in Tableau, allowing a user to select which topic they would like to display on the dashboard. So, if a user wants to see the prevalence of discussion around product returns, discounts, selection, or any other topic including more specific sentiment topics, they can choose it from the dropdown to get an entirely new view of the dashboard.
This dashboard can now be published to Tableau Server or Tableau Online and distributed across the organization.
While Tableau makes it especially easy to create portable interactive dashboards, it’s just as easy to use the Luminoso API to integrate with other solutions. If you’d like to learn more about how Luminoso Analytics can help you make sense of your natural language data, please contact us and we’ll be happy to show you.