Gavagai Explorer uses a lexical approach to perform sentiment analysis. The system uses a curated list of sentiment bearing terms for each sentiment (the sentiment's lexicon), to determine the sentiment for verbatims. To ensure that the sentiment analysis performed by the system is usable across domains, the lexicon for each of the sentiments is generic, and contains non-ambiguous terms. For example, the term 'soft' has no sentiment inherently, however in the context of diapers it is a positive term and in the context of teeth, it is a negative term. To ensure that the analysis of neither data set is negatively impacted by the term 'soft' being classified as positive or negative, the term is excluded from the lexicon of both sentiment. However, for analysis of a dataset of diaper reviews, it would be appropriate to classify 'soft' as positive specifically for that dataset. This can be achieved by utilizing Sentiment Customizations.
List of Sentiment Customization
The list of your Sentiment Customizations can be accessed from the top menu by clicking the tab 'SENTIMENTS' on the top menu. Here you will be presented with a list of your Sentiment Customizations, with a possibility of deleting the Customizations. Clicking any of the Customizations will allow you to edit it. You can also choose to create a new Sentiment Customization by clicking the 'Add Customization' button.
Creating or editing Sentiment Customizations
In the edit Sentiment Customization screen you can edit the title of the Customization and add or remove terms from the lexicons of each of the 8 supported sentiments used by Gavagai Explorer. Please note that the language of the Sentiment Customization must be selected prior to adding or removing terms and once selected it cannot be changed.
Adding Sentiment Terms
You can add custom sentiment terms to the lexicon of a sentiment utilized by the Gavagai Explorer for performing sentiment analysis.
When a sentiment term is added to a sentiment, it is an indication that when the added term is encountered when performing sentiment analysis in the context of a project where the Customization is being utilized, the term is to be treated as part of the sentiment's lexicon.
For example, if the term 'soft' is added to the sentiment 'POSITIVITY' in a Sentiment Customization, and that customization is used while exploring a project, when the system encounters the term 'soft', it will be treated as a positive term when performing sentiment analysis.
Weights for Added Sentiment Terms
If you are an advanced user of Gavagai Explorer and are aware of the details of the sentiment algorithms, you may choose to include/override weights for relevant sentiment term, by adding a '::' followed by the weight after the term. For example, if you would like to include the term 'softer' with weight 1.2, you may do so by entering softer::1.2 in the text input. If the weight for any of terms is different from the default, it will be displayed when the term is hovered over.
Removed Sentiment Terms
You can also choose to exclude sentiment terms from the lexicon of a sentiment utilized by the Gavagai Explorer for performing sentiment analysis. When a term is removed from a sentiment, it is an is an indication to the system that when the term is encountered in a project where the customization has been applied, the term is not to be treated as a part of that sentiment's lexicon when performing sentiment analysis (assuming that the term was originally part of that sentiment's lexicon). If the term was not part of the sentiment's lexicon to being with, the term is no impact to adding the term.
For example, if the term 'beast' is removed from the sentiment 'NEGATIVITY' in a Sentiment Customization, and that customization is used while exploring a project, when the system encounters the term 'beast', it will not be treated as a positive term when performing sentiment analysis.
Terms belonging to other Sentiments
In some cases, you may want to the system to treat a term as part of a sentiment's lexicon that it is not generally associated with. For example, the term 'cold' is generally considered a negative term, but not in the context of ice cream, in which case it is considered positive To apply this customization to your data set, you need to create a Sentiment Customization in which 'cold' is part of the the 'Added Sentiment Terms' for POSITIVITY and also part of the 'Removed Sentiment Terms' for NEGATIVITY.
It may be hard to determine which other sentiments a term belongs to when adding it to the 'Added Sentiment Terms' of a specific sentiment. To assist with this, Gavagai Explorer will notify you if the added term belongs to the standard lexicon of any other sentiment. Based on this information, you may choose to add the term to the 'Removed Sentiment Terms' list of that sentiment.
Applying a Sentiment Customization to a project
Once your sentiment customization is ready and you want to use if for sentiment analysis of a project, you may do so by choosing the Sentiment Customization in the 'Sentiment Customizations' tab of the project in the project explore page. Using a Customization when exploring the project will cause all related Sentiment Analysis to utilize this Customization. This includes Project & Topic Level Sentiments, sentiments in Excel and CSV reports, sentiments in Examples, and in the Gavagai Analysis Dashboards.
The list of Sentiment Customizations available for a project is filtered based on the language of analysis for the project. The default Sentiment Customization selected is the 'Use Standard Sentiments' option.