On a recent webinar co-hosted by Radian6 and OpenAmplify, our team had the chance to sit in on an active and detailed discussion about the role of automated sentiment analysis in monitoring social media streams. OpenAmplify’s CEO did a great job explaining the pros/cons of sentiment analysis in a candid and open manner – and kudos to Radian6 for organizing open online discussions like this.
That said, it was clear from the comments shared on the board that the attendees who had tried to use sentiment and keywords to detect leads were less than impressed with the results.
I believe there are three good reasons for this under performance, and most of it is common sense.
Reason #1: Keywords & Sentiment Aren’t Needs
Millions of people are searching for potential leads in social media streams right now. They are looking for prospective customers they can connect with – perhaps by answering a question or by offering some useful content. Most of them are filtering social media streams using keywords. Some are using sentiment signals to enhance their filters.
When trying to detect potential leads in social media, it’s really important to understand what your tool is actually searching for. Because keywords, sentiment and needs are very different things:
Keywords are “word markers” used to identify specific places, people, concepts, brands, products, etc. A keyword is nothing more than a single spelling of a noun or a phrase. A simple topic like “sports car” has many dozens of relevant keywords that identify it online. By itself, a keyword isn’t an expression of need.
Sentiment analysis aims to determine the attitude of the author with respect to some topic or for the overall post.
A statement of need is typically a question, a complaint, or an explicit request for information. A need is often expressed as a sentence or a question (although it doesn’t have to be). A need usually involves an actor and a verb phrase plus perhaps an object of action or intent. A single need is a complex thing, linguistically speaking – it may be represented by hundreds or even thousands of meaningful word combinations.
Because of these differences, Keyword Filters, Sentiment Analysis Tools and Need Detection Tools differ in their ability to spot a lead:
Keyword & topic filters give you a stream of posts containing specific words and phrases. That’s it. There’s no particular meaning or intent guiding the results. There’s no room for interpretation. Spam is included if it contains a keyword. If you don’t type the right keywords into your filter, then you won’t get good results. Most SMMS tools allow you to slice and dice your stream with “tags and flags”, but the bottom line is that while keyword filters are easy to use, they generate streams that omit a lot of needs and contain a lot of noise.
Sentiment classification is typically applied to a keyword-filtered stream. Sentiment doesn’t help you detect more potential leads, but the messages you do find are tagged with positive and negative sentiment. This is useful information when lead-hunting – complaints tend to be negative, for example. But many other expressions of need are often neutral. Bottom-line, sentiment analysis provides a little better signal, but it doesn’t overcome the limitations of keyword filters – it just adds a little more information.
Need detection platforms like NeedTagger yield a stream of posts containing human expressions of explicit needs, implied needs, feelings, opinions, questions and frustrations related to your content, your brand, your target audience, your products and your industry. Need detection produces a highly filtered stream – the number of posts (yield) tends to be much less than keyword-filtered streams. For the most part, spam is also filtered out of the stream.
If you want to detect potential leads, which of these three methods should you be using?
Reason #2: Keyword-based Tools Require You To Be a Language Expert
How good are you at generating keyword lists? Before you answer, consider the following two posts, both of which are decent leads:
“I need an affordable new car with great gas mileage”
“I need to replace my VW”
If you search for brand names like “VW” or “Volkswagen”, then you won’t ever see the first lead.
This illustrates a common but important limitation of all keyword filters: if the user doesn’t understand and load all of the most important phrase and keyword combinations into their query – every possible combination that matters – then s/he will miss many potential leads.
Problem is, there are often thousands of valid word combinations that can accurately describe a specific need (lead) in the language of social media. Which ones matter most?
In the real world, most business professionals have neither the time nor the skill to generate & prioritize huge lists of relevant word combinations. They aren’t language experts, and they won’t be any time soon. This issue is a major stumbling block preventing SMMS tools from detecting leads with broad coverage and precision. This is one of the reasons we built NeedTagger.
Reason #3: Keyword & Sentiment Filters Miss Implied Needs
Implied needs are statements that represent precursors, symptoms or root causes of commercial intent. Expressions of implied needs often contain no recognizable keywords, so they are tough to detect.
For example, consider the following tweet:
@johntaylor: ”I just totalled my ride, can’t tow the boat any more!”
- John probably needs a new truck. (implied need)
- what keyword would you have used to detect this need?
- if you searched for any of the common keywords contained in this post, you would have introduced off-topic noise into your stream.
Social media monitoring tools are simply not designed to help people detect implied needs with broad coverage and precision, because they lean so heavily upon keywords and a limited set of very explicit phrases.
Why Not Just Add Action Phrases to My Keyword List?
Some SMMS providers tell customers they can find leads by searching for action phrases in addition to keywords. While this is technically correct (you will find some needs this way), if you do this then it is highly likely you will miss a lot of needs out there, and you may introduce new types of noise into your filtered stream. The reality of doing this well – creating exhaustive lists of 2- 3- 4- and n- word phrases – is a very time consuming exercise. Also, most SMMS tools limit the number of keywords you use in a filter.
Keyword Filters & Sentiment Analysis Are Great… for Certain Things
We are not arguing about the general utility of keyword and sentiment filters – keywords are fundamental to NeedTagger’s platform. We use keywords a LOT. We use sentiment classification in our platform, as well.
Keyword filters and sentiment analytics definitely have their place in social media monitoring: they work well for tracking brand sentiment and measuring explicit product / brand mentions over time. They are also great for monitoring popular channels, for example by using hashtags.
But keywords and sentiment are not needs; so, relying upon keyword and sentiment filters to detect leads is not enough. Detecting needs in social media requires a different approach.
To Detect Leads In Social Media, You Need To Analyze More Than Text
Human expressions of needs are complex and nuanced things. And complex/nuanced things require a lot of data to identify exactly what they mean.
Keyword filters and sentiment analysis tools typically only analyze the text contained in the body of a message. On Twitter, that’s at most 140 characters. Not a lot. Blogs, discussion forums and many other social networks also consist primarily of short text messages.
Unfortunately, a short text message is usually not enough data to precisely classify a post as a need you can meet… from a person you care to help… with content, products and services you offer… right now. You need more data to do this with precision. Unfortunately, the vast majority of SMMS tools today still only look at the body of the message when classifying a post for intent or action.
Thankfully, there are a LOT more signals available in social networks than the body of a post. Modern “need detection” platforms like NeedTagger are built to take advantage of multiple signals to do their job.
NeedTagger’s platform considers multiple signals when detecting a need for your campaign, including:
- What is the context of the post? Is this a need within this specific industry and campaign? Is this post part of a conversation? or is it a request for help? a news-sharing event?
- What is the profile of the person discussing the need? What is their job title? What is their posting history, social graph, and interest graph?
- What is their conversation history? do they regularly talk about topics you care about? have they expressed a need in the recent past?
- Is the person influential in your industry? how do they rank – within the scope of your campaign?
- Historically, what signals have best served your campaign? what types of posts & people tend to deliver the best results for your organization and content?
- and of course, what does the post say? are there important keywords in it? what meaningful part of speech patterns exist? what actors, verb phrases and objects are involved? what links exist?
- does this person want to speak with you? has this person asked you to stop sending posts in the past?
NeedTagger’s platform is designed to leverage all available signals to detect relevant needs for your campaign. We do this with a high degree of sensitivity to the unique terminology used in your industry and in the social networking platforms that matter to your campaign. In fact, everything we do is optimized for each campaign – the keywords, speech patterns, hashtags, author profiling, etc.
So far, the engagement levels our customer messages have been able to produce with our vertically-sensitive, multi-variable approach are an order of magnitude better than most other forms of marketing currently deliver on Twitter. Plenty of work left to do, but our customers seem to appreciate the results so far.
We hope this post explains why keyword and sentiment filters struggle to detect human needs for content, products and services that you can meet – right now. That said, keywords and sentiment remain excellent tools for tracking mentions and brand sentiment.
NeedTagger was designed to detect needs in social media that you can meet – right now.