Build/Test a sentiment model

• How do sentiment models work?
• Dashboard
  • Create a sentiment model
  • Add an example sentiment model
• Sentiment Models
  • Multiwords
  • Settings
  • Build/Test a model
  • Import/Export a model
  • Using a model
• Entries
  • Subentries
  • Advanced settings
• Fine tuning
  • Checking your results
  • Optimization tips

Once you have created a sentiment model and its entries you can test it with any of the services that support sentiment analysis. The Build action is a way to ensure that the most recent version of the model is the one used by any of those services.

Currently, there are two services that support sentiment models:

• Lemmatization, PoS and Parsing
• Sentiment Analysis

By default, sentiment models are tested using the Sentiment Analysis API. When you click on the Test button either in the sentiment models dashboard, in the sidebar of the different model views or in the build page, you will access the Sentiment Analysis Test Console with your license key and your sentiment model already selected.

In the image above, we are testing our example sentiment model with a restaurant review. These are the results:

In the results you will be able to see the defined terms of your own model and how they interact with the entities and concepts detected in the text. The sentiment analysis is included at different levels in the text:

• Aggregated sentiment of the whole text.
• Sentiment detected at sentence level.
• Sentiment detected in sentence divisions, at segment level.

There are three important fields in each of these analyses, the confidence of the analysis, a score_tag showing the overall sentiment of the analyzed text and the agreement between the different parts of the analyzed text.

At the segment level, a list is included with information about polarity_terms included in your model, that is the entries and subentries you have defined. Inside this list, you will have a sentimented_entity_list (or sentimented_concept_list) when one of the previous terms affect to an entity or concept detected in the text.