June 2022

Search & Discover Results Now Searchable 

Search & Discover users can now search within search results to help filter results even further. 

To utilize the new search bar, navigate to Search & Discover and click the Search Results report you wish to search within.

You will now see the "Search by Keyword" bar atop your search results. Type in the keyword and press Enter to perform the search within your results.

Hint: Keyword Search within a Search is only available on Search & Discover Searches executed on or after 6/21/2022.  

Machine Learning Models

Toxic Speech Model Version Update 

The Aware Data Science Team have worked tirelessly to release the latest version of our Toxic Speech Model. Users can expect improved accuracy in the classification of chat messages as healthy, inappropriate, offensive or hate speech where used in Aware, leading to decreased rates of false positive or false negative results. Scores and classification may differ for a message that was evaluated using the previous version of the model.

• Ensuring our model remains relevant and timely in the context of how employees are collaborating today - taking into consideration natural changes in language over time as well as how people communicate in the workplace on new platforms.
• Improved accuracy in predicting the likelihood of the message being in a specific toxic speech category.

• Accuracy improvements also means a decreased rate of false positive and false negative results.
• Targeted enhancements addressed certain shortcomings identified in the old model including the performance of short messages. For example messages containing only a few words like "Hey man", "hey", or "lol" were labeled Offensive when evaluated by the old model, now they return as Healthy. 

Hint: When using the Toxic Speech AI Analysis feature on a search in Search and Discover, any content surfaced prior to June 3, 2022 will  have been scored on the old model, due to messages being scored when they are created.

Spotlight

Behavior Indicator Updates

KPIs will now be replaced by Behavior Indicators. Behavior Indicators shine a light on a set of messages that have unusual behavior.

To help make things easier to understand, Sentiment and Health Behavior Indicators will be displayed as a percentile to highlight anomalies within your specific organization’s collaboration platforms.

Upcoming Changes and Examples:

Using the 1-100 scale, data will be ranked as percentiles to indicate how the displayed subset of data ranks compared to typical conversation trends at your organization. Conversations within your organization will be analyzed over time and the Standard for your organization will be set within Aware.

What is the Standard?

To determine what is Standard, Aware first takes into consideration where messages are written and how and where your organization uniquely communicates. This Standard takes into consideration the organization, scope (DM, Public Group, Private Group) and platform (Slack, Workplace, Yammer, Teams, etc) and is trained on millions of pieces of new data. A normal distribution of scored messages is determined using Aware’s Natural Language Processing models, Sentiment, and Toxic Speech.

For Example:

• A 50 would indicate that the displayed dataset falls right in the middle of what is Standard, or normal, for your organization. Meaning the 50th percentile. In this situation, half of the  conversations are more positive than this data, and half of the conversations are more negative than this data.
• A 90 is the 90th percentile, or Above Standard, for your organization. In this situation, only 10% of conversations would be considered more positive than this data (which is a big deal) and 90% of conversations would be considered less positive than the displayed data. Consider this a big win!
• A 20 is the 20th percentile, or Below Standard, for your organization. This means that 80% of conversations are more positive than the displayed data in your organization, and only 20% of conversations are considered more negative than the displayed data. This fluctuation can be an indicator to run a topic report to better understand the change in sentiment!

These Behavior Indicators will be categorized into Below Standard, Standard, and Above Standard based on the ranges below: 

Need more information on Behavior Indicators? Checkout our FAQ document