What's New in VICA 1.8.7

Improvements

1. Updated information on the Analytics Dashboard

Firstly, the tooltip for the Chatbot Quality score has been updated to reflect the new components.

Secondly, receiving more Thumbs Down will be displayed with a red increase arrow. (Previously showing green increase arrow). It’s a small change that corrects the logic, and hopefully becomes more intuitive for users to understand.

In the above example, users received 4 more thumbs down, hence represented with a red increase arrow.

*There is a known bug where the platform median are showing all 0 values. We are working on it.

2. Improved CSAT downloading

There is a new option to download “Table data in CSV”. As the name suggests, it allows you to download the whole table as a CSV file. This helpful in reporting/looking through the different CSAT scores and comments.

The previous functionality of downloading every single chat transcript as individual CSV file is still preserved in the “Individual transcripts in CSV” option.

Sample of CSV shown here:

3. Glossary

Abbreviations mean different things in different contexts.

For example, SVP could mean ‘Senior Vice President” in a organisational context. But when it comes to claims procedure, SVP can also mean “Small Value Purchase”. This makes it difficult for embedding models/LLMs to fully understand what users are referring to.

We’ve introduced the Glossary to help with this.

You’re able to define the abbreviations and the full expanded terms. Our systems can then perform a text replacement to increase the accuracy of the matching.

For example:

Mechanisms of the Glossary:

1. There needs to be 2 capital letters in a word to trigger the Glossary.

SVP, SVp, SvP, sVP are all valid conditions.

2. Once detected, the Glossary will then check against your inputs to find a match. In this example, there is a match for SVP. If there are no matches, then no replacement happens.

3. Glossary will then perform the text replacement from “what is SVP?” to “what is small value purchase?”

4. This gets sent to our models to interpret and perform the matching.

We are employing the 2-capital-letter mechanism as a conservative roll out. This is to prevent unintended text replacement from occurring. For example, my mom, as in mother, from being replaced to Ministry of Manpower.

Do give it a try and let us know your feedback.

Fixes 

• Stability enhancements 
• Various miscellaneous bug fixes