Smart Theme Explorer v/s Topic Cluster

Updated 

Topic Cluster

For generating topic clusters, we take a large sample of messages (randomly selected after applying the given dashboard/widget filters) and perform data hygiene steps like stemming and removal of stop words from these messages. We then identify phrases frequently occurring in the messages (we have an underlying text analysis engine to do that) and then cluster the messages into groups defined by these phrases. As the sample of messages generated is random, hence the topic cluster might seem different at different moments.

  • 1st Tier/Internal – The most occurring phrase in the data

  • 2nd Tier/Internal – The most occurring phrase in the individual sections of the 1st tier

Smart Theme Explorer

Smart Theme Explorer is a research tool that enables users to quickly understand the underlying themes of conversation under a Topic or any data-fetching listening entity, thereby helping them cut through the clutter of unstructured data.

It is a free tool for all Sprinklr Insights users and must not be compared with specialized tools like Product Insights which are supervised, verticalized, and highly specific.

Product scope

  • Available for limited languages currently.

  • Smart Theme Explorer can work on all data fetching listening entities like Topic, Topic Groups, Account, Account Groups, Brands, Business Locations, Products, and first-party data as well.

  • Cluster sentiment is derived from the message-level sentiments.

  • The maximum number of levels up to which you can navigate is 5. It might be possible that no clusters are formed after a particular level due to insufficient data.

Difference between Smart Theme Explorer and Topic Cluster

Parameter

Smart Theme Exploere

Topic Cluster

Sampling

  • Un-sampled

  • Same clusters for same data filters

  • Randomly selected phrases from a large sample of messages.

  • Clusters may change based on sampling.

Inappropriate Content Filtering

Mentions containing inappropriate phrases, keywords, profanity are not counted towards clustering

No Content-based filtering for inappropriate data.

Data Hygiene

Stop words are removed

Stemming of phrases

Number of Clusters

All possible clusters are present and shown and can be seen via pagination or list view.

Limited and top occurring phrases are shown only.

Volume Representation

Volumes are represented by the size (radius) of the cluster.

The size of a cluster does not represent the volume of that cluster.

Phrase Identification

AI-based but more advanced. Identifies subjects and attributes at a sentence level, takes into account nouns, adjectives, verbs, adverbs, etc.

AI-Based (NLP)

Pairing and Correlation of Clusters

  • AI-powered pairing & clustering of phrases identified in a sentence

  • It is based on Subject-Subject, Subject-Attribute, and Attribute-Attribute pairings

Phrases are clustered solely on the basis of volumes of mentions for the phrase.

Level of Clustering

Up to 5 levels

Up to 2 levels

Additional Capabilities

  • Theme Creation capabilities from the clusters

  • Top Keywords forming up the cluster is visible

  • List View/Cluster View

  • Hide Clusters for reporting

  • Theme Creation capabilities not available from UI

  • Top Keywords forming up the cluster are not direct; can be achieved via drill down by Word Cloud

Sentiment

The sentiment of clusters is shown in the UI

The sentiment of clusters is not available.