Sentiment Compare

Sentiment Compare puts two periods of community sentiment side by side, for example the weeks after two patches or an event and the week before it. You pick the dates. BetaHub shows the numbers for both periods, the mood day by day, the topics and phrases that grew or faded, and a short AI description of what changed.

When to use

• You shipped two patches and want to see how the community took each one.
• You ran an event and want to compare it with the weeks around it.
• You want the difference between two periods without asking a question first.

To ask a question instead, such as whether a fix landed, use Sentiment Checks.

Requirements

• Pro or Enterprise plan, the same plans as Sentiment Analysis. On other plans the page explains this and you can’t start a comparison.
• A role that can view analytics.
• Sentiment Analysis collecting messages from your Discord channels or Steam Discussions for both periods. A comparison only sees messages collected there.

Comparing two periods

1. Open Sentiments and set the view you want to compare: the date range, a saved report, channels and roles.
2. Click Compare periods. The form opens with your date range as period B and the same number of days before it as period A. Period B ends yesterday at the latest, because today is still filling up. Your saved report, channels and roles carry over; the word and mood filters don’t.
3. Set both periods. For each one, type a name and pick the start and end dates, or choose a release under Fill from a release. A release’s period starts on its publication day and ends the day before the next regular or major release. Hotfixes count as part of the release they fix. The latest release runs to yesterday, the last full day.
4. Check the bars under each period. They show messages per day; red bars are days with no or almost no messages, which usually means collection stopped. Today never counts as a gap, because it is still filling up.
5. Click Compare. This uses one comparison from your monthly limit.

The result page updates by itself for up to 20 minutes; reload it after that. The numbers appear first and the description follows, usually within a minute. You can leave the page: the comparison keeps running.

Each period can be up to 90 days long and can’t end in the future. The periods can’t overlap. Dates are whole days in UTC, whatever your time zone.

From a release page

When a release has its automatic release report, click Compare with previous release under the report. The form opens with that release as period B and the regular or major release before it as period A.

Reading the result

The page leads with the description and shows the numbers it rests on.

• What changed: a short summary, the overall direction (Improved, Worsened, Mixed or No clear change) and how confident the description is. Each change is marked Better, Worse or Shifted (neither better nor worse, such as more people talking) with the numbers or phrases that show it. The note at the bottom says what limits the comparison.
• Numbers: messages and active people per day, net mood, average rating, and the share of negative, frustrated or toxic messages. Net mood runs from −100 to +100: positive and excited messages minus negative, frustrated and toxic ones, as a share of all messages. In the change column, green is better for the mood, red is worse, and grey means neither.
• Net mood by day of period: both periods start at day 1, so you can see how each one unfolded. Period A is the dashed line, period B the solid one. Dotted lines mark releases published later in a period.
• Mood mix: the share of each mood in both periods.
• Channels with the biggest change: channels with at least 20 messages in both periods, by how far their share of negative messages moved.
• Topics: mentions per day and the average rating of each topic in both periods.
• Phrases that grew and faded: key phrases mentioned more or less often in period B, with how many times they were mentioned in each period. The factor is per message, so it allows for periods of different length. new means the phrase wasn’t mentioned in period A; gone means it wasn’t mentioned in period B.
• Data coverage: messages per day in both periods, with gaps in red.

Periods often differ in length, so every number is per day or a share. Per-day numbers count only days with messages, so a gap in collection doesn’t make the community look quieter.

Checking the evidence

Every change in the description and every number can be checked against the messages behind it.

• Under each change, See messages lists what it rests on: phrases, topics, channels and moods. Click one to open its messages.
• In the numbers, click a mood in Mood mix, a channel, a topic or a phrase.

The messages open side by side: period A on the left, period B on the right, newest first. Each shows its mood, channel, date and roles, and Discord messages have Open in Discord. The lists read the messages as they are stored now; when the data changed after the comparison was computed, the list says what the comparison counted.

What a message shows depends on message storage: with Full text you see the message, with Private (the default) its key phrases. Open it in Discord to read it.

When there is no description

• Not enough messages: a description needs at least 100 messages in each period. The numbers are still shown.
• The description failed: the numbers are saved. Click Try again; it doesn’t use a comparison.

When a period has no data on more than a quarter of its days, the description’s confidence is always Low.

Gaps and changing data

A comparison is saved as it was computed: the date next to As of says when. It doesn’t change by itself, so a result you shared stays the same.

The data behind it can change, for example after you change team roles or messages are added for past days. To use the current data, click Recompute. It computes the same periods with the same filters and uses one comparison from your monthly limit. When the number of messages changed since the last computation, the result says by how much.

A comparison made from a saved report can’t be recomputed after the report is deleted.

If a comparison fails or stops updating for 15 minutes, for example during a BetaHub update, click Try again. It doesn’t use a comparison.

Your comparisons

Comparisons are saved in the project. Open the list with All comparisons on any comparison page. It shows each comparison’s periods, data, author, the date it was computed and its state, including Gap in data when a period has gaps or channels with messages in only one period.

• Who sees them: everyone in the project with a role that can view analytics.
• Who can delete one: its author, or someone who can edit the project’s settings.

Limits

• Creating a comparison and Recompute each use one. Try again doesn’t.
• The count resets on the 1st of each month (UTC). Deleting a comparison doesn’t give it back.
• Need more? Contact support.

Privacy

The description is written from the key phrases and moods BetaHub already stored, never from the text of messages, so it works with Private message storage too. The message lists behind the evidence show message text only when message storage is Full text.

See Also

• Sentiment Checks – Ask whether a fix, release or campaign landed