Sentiment Analyzer Control Panel

The Sentiment Analyzer Control Panel is a stand alone application that can be used to test & train the ThinkAutomation Sentiment Analyzer. It can be started by clicking the Open Sentiment Analyzer Control Panel button on the Score Sentiment, Train Sentiment & Classify Sentiment actions or by running the ThinkAutomationSentimentControlPanel.exe application.

You must first login with your ThinkAutomation User Name/Password.

The current Class Names will be listed. Select a Class Name. You can add a new class using the Add button. To clear all training data for a class click the Reset button.

Test Or Train With Text

Select this tab to test or train the currently selected class name with manually entered text.

Enter or paste text into the box. Click the Test button to analyze the text. The sentiment score will be displayed along with Tokens Used and Tokens Scored.

Click Train Positive or Train Negative to add positive or negative training data for the entered text. Click Add Ignore Words to add all tokens from the entered text as Ignore Words. Click Tokenize to just display the extracted tokens without adding training data.

Test With Files

Select this tab to run multiple tests using local files. Select the Positive Files Path for the folder containing files that you expect to score positive and Negative Files Path for the folder containing files that you expect to score negative.

Enter the file Mask (eg: *.txt) or enter *.* for all supported file types. You can use txt, html, csv, eml or msg (Outlook Messages).

Click Run Test to start.

Each file will be scored - and the overall results displayed along with the overall accuracy percentage.

Train With Files

Select this tab to add training data using multiple local files. Select the Positive Files Path for the folder containing files that will be trained as positive and Negative Files Path for the folder containing files that will be trained as negative.

Enter the file Mask (eg: *.txt) or enter *.* for all supported file types. You can use txt, html, csv, eml or msg (Outlook Messages).

Click Run Training to start.

Each file contained in the selected folders will be read, tokenized and the tokens added as either positive or negative training data.

You can use the Test options to run tests after each training operation.

Training Strategy

For best results each Class Name should be trained with roughly the same number of positive and negative messages. Accuracy will improve with more messages trained.

For example, suppose we want to use the Sentiment Analyzer to flag up incoming messages as 'Sales Enquiry'. You would create a 'SalesEnquiry' class name. You then need to obtain roughly the same number of messages that ARE sales enquires and that are NOT sales enquiries. Save these to separate folders on your file system. You can save Outlook messages to these folders. Several hundred at least of each if possible.

Then use the Train With Files option to add the training data.

Now use the Test With Text option and enter some new text that you would consider a Sales Enquiry. The scored result should be POSITIVE. Enter some new text that you would consider NOT a Sales Enquiry. The scored result should be NEGATIVE.

You can also use the Train Sentiment automation action to add training data. The benefit of the automation action is that you can link it to a message source to read folders from an email source. For example: You could create email folders for positive/negative messages and move respective email items to those folders. Then create ThinkAutomation message sources to read the folders and execute the Train Sentiment action. You can then add new email items to these folders whenever you want to update the training data.

Classification

Classification is the process of finding the most relevant Class Name for any text. For example: Suppose you have two Class Names: "SalesEnquiry" and "SupportEnquiry" and training data has been added for both. Classification can be used on some new text to indicate which Class Name is the most relevant. You can use the Classify button on the Test With Text tab to perform classification tests.