Sentiment analysis is a valuable tool for understanding how your audience is feeling about a given topic. However, measuring audience sentiment is a nuanced endeavor and often requires that brands apply their unique context and perspective to ensure an accurate assessment.
This article contains the following sections:
For example, if a gaming publisher was listening for messages around a game their company produces called “War Zone,” they might find their sentiment results to be overwhelmingly negative because the term “war zone,” is inherently negative. This can lead to skewed results and an inaccurate analysis on what the sentiment of the conversation on social actually is. Sometimes, sentiment analysis can be unintentionally inaccurate.
The good news? You can use Sentiment Reclassification and Sentiment Rulesets when creating your Listening Topics to help you increase the accuracy of sentiment.
With Sentiment Reclassification, you can reclassify the sentiment rating that was applied to a message. The second, Sentiment Rulesets, helps with the “war zone” example above, by changing how Sprout’s sentiment classifier interprets words or phrases.
In the example above, you could make “war zone” a neutral term since it is tied to a name of a game. By changing the rating on individual messages or words and phrases, you can improve the sentiment accuracy and improve the ratings the algorithm applies to messages.
Sentiment reclassification is great for your Brand Health or custom Listening Topics that are smaller and don’t contain thousands of messages. You can quickly update the sentiment on a message-by-message basis.
Reclassifying message sentiment
To change the sentiment on a message:
- Navigate to your Topic Insights.
- Open the Messages tab.
- Click the Sentiment dropdown.
- Select Positive, Neutral or Negative.
While Sentiment Reclassification is a great way to change sentiment on individual messages, Sentiment Rulesets let you decide how the sentiment classifier interprets and classifies specific keywords. Rulesets help guide the sentiment classifier to make a more informed decision on how to treat words in a message.
If a message is fairly straightforward, the message will likely get classified per your ruleset. For example, “I have back pain.”
The word “pain” in that message is interpreted as negative, so the overall message is negative. If you added a rule to make “pain” interpreted as neutral, the message rating likely becomes neutral.
It’s important to know that sometimes the rest of a message can influence the overall sentiment classification. For example if the message said, “I have bad back pain. These new pills are awful and don’t help.” The message is interpreted as negative. Whereas, “I have bad back pain. These new pills I got are great.” This message is interpreted as positive.
In both of the situations, other keywords and phrases influenced the classification of the message beyond the rule. But, in either case, you’re ensuring the word “pain” remains neutral and leaving the rest of the message context to determine the true sentiment.
Creating a ruleset
You can create a ruleset from the Settings page and view which Topics are linked to it. You must have manager permissions in order to create or edit rulesets.
To create a sentiment ruleset:
- Navigate to Account and settings > Settings.
- Click Sentiment Rules under Listening. Your Sentiment Keyword Rulesets appear.
- Click Create new. The Create New Sentiment Keyword Ruleset popup appears.
- Enter a Title for your ruleset.
- Enter a Description.
- Enter keywords and phrases (up to three words) in the Positive, Neutral and Negative Sentiment fields, separating each keyword and phrase with a comma.
- Note: The system doesn’t take into account variations of words, so you need to enter each variation if you want it classified in a particular way. (Think: win, winning and wins)
- Additionally, you can’t use the same keyword or phrase in multiple sentiments. (Think: You can’t classify “win” as negative and positive)
Applying a ruleset
You can only apply one ruleset to a particular Topic, but you can reuse your ruleset for multiple Topics. For instance, if you wanted to create a ruleset that ensures all your brand-specific keywords are classified as neutral, you could apply this ruleset to Topics that focus on your Brand Health and Campaign Analysis.
To apply a ruleset to a Listening Topic:
- Navigate to the Topic you want to apply the ruleset to or create a new Topic.
- Select a keyword ruleset from the Apply Sentiment Keywords (Optional) dropdown.
Rulesets only apply to data going forward. You can’t edit an existing topic and apply a ruleset.
Editing a ruleset
From the Topic Builder, you can also edit a particular ruleset or manage all keyword rulesets. This impacts everyone in your group.
If you click from the Sentiment dropdown in the Query Builder you're taken to your Settings to make additional edits to all of your rulesets. Note that your edits affect any Topic the ruleset is applied to. Click Save to save your changes.
If I add a new ruleset to an existing Topic, does it apply to the previously collected data?
No. Any ruleset you apply to a Topic only applies to messages from that point forward. If you want to apply the ruleset to historical data, you must backfill the Topic. If you recently created the Topic, duplicate the Topic and let it automatically backfill historical data.
If I edit a ruleset to change the rules, how does it affect Topic data?
Any rule changes apply to new data from that point forward.
Can I put the same rule in multiple sentiment categories?
No. A keyword or phrase can only be in one category.
Can I create a default ruleset to apply to all of my Topics?
Not at this time.