Social sentiment tracking you can check line by line
Social sentiment tracking scores public mentions of your brand as positive, neutral or negative and follows how that mix changes. MentionRadar scores every mention from Reddit, X, review sites, forums, news and YouTube, charts the weekly trend and alerts you when negative mentions spike.
- Every mention scored as it arrives
- Weekly trend and a single sentiment index
- Correct any label with one click
- Instant alerts on negative mentions
Northwind Coffee, all sources, last 8 weeks. Sentiment index +19 this week, up from +12.
How a single mention gets its score
No black box. The scorer reads the headline and visible text of each mention and follows five rules you can check yourself.
-
1
Words carry weights
A word list gives sentiment words a weight between -3 and +3. "Excellent" and "love" are +3, "reliable" +2, "easy" +1.5. "Broken" is -2.5, "overpriced" -2, "slow" -1.5. Words that carry no opinion count for nothing.
-
2
Negation flips the next opinion word
After "not", "never", "don't" and similar words, the next sentiment word within three words changes sign and is softened to three quarters of its weight. "Not bad" becomes mildly positive, not strongly positive.
-
3
Intensifiers scale the next word
"Very" multiplies the next weight by 1.4, "extremely" by 1.6, "really" by 1.3. Softeners work the other way: "slightly" by 0.6, "barely" by 0.5.
-
4
"But" moves the weight to the second half
At "but", "however" or "although", everything said so far counts half, and opinion words after it count 1.2 times. People put their real verdict after the "but", and the score follows them. A full stop resets negation, intensity and the "but" shift.
-
5
The total is normalised and labelled
The sum is turned into a score between -1 and +1 with the formula sum divided by the square root of (sum squared plus 15). A score of 0.12 or more is positive, -0.12 or less is negative, anything in between is neutral.
More scored sentences, including where the scorer struggles
| Mention text | Score | Label | Why |
|---|---|---|---|
| “Great product, really great support” | 0.83 | Positive | Two strong positive words, the second one intensified. |
| “Not bad at all” | 0.36 | Positive | Negation flips "bad" and softens it, so the result is mildly positive. |
| “Shipped a new version today” | 0.00 | Neutral | No opinion words, so the mention stays neutral. |
| “Oh great, another outage” | 0.00 | Neutral | Sarcasm. "Great" and "outage" cancel out. A human reads this as negative and should correct the label. |
The sentiment index in one number
Individual labels are for triage. For tracking, MentionRadar condenses a period into the sentiment index: positive mentions minus negative mentions, divided by all mentions, times 100. The index runs from -100, where everything is negative, to +100, where everything is positive.
Neutral mentions are in the denominator on purpose. A brand that gets 20 positive and 10 negative mentions out of 1,000 has an index of +1, not +33, because almost nobody had an opinion. That keeps a handful of reviews from dominating a month of quiet coverage.
The index appears on the dashboard, in reports and in brand tracking next to share of voice, calculated the same way for every tracked competitor.
Positive mentions
Negative mentions
Neutral mentions
Sentiment index
(412 minus 218) divided by 1,726 mentions, times 100, equals +11.
Limits of sentiment, and how to work with them
Any automated sentiment analysis misreads some mentions. Sarcasm is the classic case: "oh great, another outage" contains a positive word and means the opposite. Comparisons are another: "faster than Brand X but still slow" is positive about you relative to a rival and negative in absolute terms. Industry words can mislead too, since "killer feature" and "crushing it" read as negative to a word list.
The scorer uses an English word list, so mentions in other languages usually come out neutral. The language filter on each keyword lets you keep the feed to the markets you care about.
These limits are why sentiment works best as a trend and as an alert trigger rather than a verdict. A shift of negative share across hundreds of mentions is real even if a few labels are wrong. When a label is wrong, change it in the feed: the corrected label is saved and counts in the index, the chart and reports. Teams that handle reviews and complaints should also read about reputation monitoring software.
The negative filter is where the work starts
Filter the feed on Negative and you get the working list for support, product and comms. Sorted by time with reach on every row, it shows what to answer first.
Instant email alerts for negative or high reach mentions are on every plan. Growth adds Slack and custom rules, so negative posts on Reddit can go to the support channel while negative press goes to comms.
The same feed covers all sources. See the full picture in the social listening tool.
Northwind Coffee subscription is overpriced now
Thread in a coffee community, 31 replies about the new price tier.
Beans arrived stale, very disappointed
Two star review mentioning a late delivery.
Northwind app keeps crashing at checkout
Post with a screen recording, 48 reposts.
Questions about social sentiment tracking
How does MentionRadar decide whether a mention is positive or negative?
Each mention is scored with a weighted word list. Words like "excellent" or "reliable" add weight, words like "broken" or "overpriced" subtract it, negations and intensifiers adjust the weights, and the total is normalised to a score between -1 and 1. Scores above a small threshold are positive, below the negative threshold negative, and the rest neutral.
How accurate is automated sentiment analysis?
It is reliable in aggregate and for routing alerts, and less reliable for single sarcastic or ambiguous comments. That is why every mention shows its text next to the label and you can change the label in the feed. Trends over hundreds of mentions are far more trustworthy than any individual score.
What is the sentiment index?
It is the number of positive mentions minus the number of negative mentions, divided by all mentions in the period, times 100. It runs from -100 to +100. A brand with 30 percent positive and 10 percent negative mentions has an index of +20.
Can I correct a sentiment label?
Yes. Each mention in the feed has a sentiment selector. A corrected label is saved on the mention and counts in the sentiment index, the trend chart and reports from then on.
Which sources are included in sentiment tracking?
Every web source MentionRadar monitors: news sites, blogs, forums, Reddit, Hacker News, review sites, YouTube and X. You can filter the feed and the chart by source to compare, for example, sentiment on Reddit with sentiment in the press.
Can I get an alert when sentiment turns negative?
Every plan sends instant email alerts for negative or high reach mentions. Growth and higher plans add Slack alerts and custom rules that combine sentiment, reach, source and keyword, for example negative mentions on Reddit with Medium reach or higher.
See the sentiment of your latest mentions
Scan your brand and get recent mentions with a sentiment label on each one, plus an AI engine check, before you create an account.