This is the first part of a special two-part post from the Jugnoo algorithm and language scientists, looking at the complexity of natural language and lead gen opportunities in social media.
With the massive number of conversations and buzz on the internet and social networks, businesses of all sizes are asking the same question on how to tap into the social web to discover opportunities and leads, and nurture prospects into loyal customers.
Machine learning has been at the forefront of the technology to extrapolate tweets, posts, blogs and user conversations into actionable information for businesses to act upon and drive successful results.
What is Machine Learning?
So what is machine learning and how does it change the landscape of social business? If you look at the Jugnoo Visualyzer image below, you will see buzz on “machine”, “learning”, “social”, “data”, “algorithm”, “mining”, and the conversations going on with people on these specific words.
Related Resource from B2CWebcast: PR Hacking: How Ideas Spread And What Marketers Need to Know
Machine learning is a discipline of Artificial Intelligence using algorithms and computational techniques, to extrapolate answers and predict the potential outcome based on gained experience and knowledge from learning [see Wikipedia definition].
Online consumer conversations and reviews have given Machine Learning the wealth of information needed to acquire knowledge and evolve to adaptive learning, in contrast to manually building a training set for learning.
Machine learning is further fueled by “Big Data” to manage billions of social conversations in the cloud and computational algorithms, such as collaborative filtering for recommendations, made popular by Amazon (see the infographic at the end of this post for more).
But what can Machine Learning do for the social web and business?
How Machine Learning Applies to Social
By discovering a consumer’s intent when he or she converse with their friends, or with brands using Machine learning technology, businesses can use that actionable information to reach out to the consumers as prospects, either by introducing relevant products or resolving any issues or problems.
Because this outreach happens on social channels, the resulting awareness and positive sentiment (if the outreach is done properly, of course) becomes a win-win for all parties involved.
There are various ways for machine learning to build up the knowledge base so the actionable information is relevant to the goals of the consumers:
- First, the business can decide if the monitoring of the conversation is specific to a local region or it has a wider reach.
- Once a specific goal is identified, such as launching a new product or supporting an existing service, these goals can be identified in the input data stream for monitoring. A competitive goal can also be established where a negative sentiment on competitors, products or services translate to opportunities for your business.
- The machine learning can then analyze a thread of conversations, using the top down approach to uncover the intent and business goal. Once the intent and goals are extrapolated, the inferred outcome for leads or opportunities can be predicted by correlating the sentiment and mood of the conversation.
The predicted outcome can range from sales leads, opportunity, customer service requests and much, much more.
Based on the predicted outcomes, actions can then be recommended for the business to follow up with the consumers – either a friendly introduction or promotion for a prospect, or private direct message to provide necessary customer care to resolve issues and complaints.
Machine Learning for Intent Recognition in Social Media
In online conversations, consumers can express various intents: inquire, complain, sell, buy and opine, for example. By recognizing the intent in those conversations, you can create a closer connection with your customers, improve product ratings, and increase the sales.
The popular intent in social media is the intent to opine, where users express the sentiment toward a subject. Depending on whether they are complaining about you or a competitor’s product, for instance, you can reach out to them to help or advise, improving their satisfaction or creating a sale opportunity.
1. Intent to buy: social sales leads
Probably more important, but not understood enough, is the recognition of intent to buy.
While search engines like Google have significantly advanced their technology of recognizing customer’s intent based on their search, businesses may not have tapped into social media enough to recognize that users announce valuable pieces of information about themselves all the time.
These can lead into great business opportunities. The sticky part, if you like, is that the opportunity in hand isn’t always obvious.
The search engines, for example, are identifying and helping users already interested in an item, so when they type “laptop” in the search field, you can display appropriate ads that will introduce them to a computer store . Notice that users themselves create a search query with the intent to get as accurate results as possible which means that their query has already be cleaned of any noise.
On the other hand, social media users will sometimes be more explicit when stating their intentions, by using natural language expressions that are not easy to process. Additionally, there is a big amount of implicitly stated intents, which require way more complex analysis in order to be found.
2. Explicit intent recognition – the introduction of Machine Learning
Recognizing explicitly stated intent to buy can provide leads like click-through ads in search queries. If someone tweets “I am going to buy a laptop.”, we can be almost certain about having a determined customer at hand.
On the other hand, if we notice a “laptop” query submitted to an engine, we still don’t know whether the user already has the computer and needs more info about the features, or maybe is just trying to help a friend to buy.
In order to recognize an explicit intent, a robust engine for natural language processing is part of the machine learning technology stack. Using Bayesian classifiers messages allows social media conversations to be cleaned of spam or noise.
Then, by applying text analysis, tweets or posts are broken down to better understand the topic of conversation and locate intent – which can then be analyzed using machine learning methods like Maximum entropy classifiers or Support Vector Machines (SVM).
Once the analysis is performed, the subject of the conversation is uncovered and the attitude of the user toward the subject is pinpointed.
3. Implicit intent recognition – furthering the application of Machine Learning
There are hundreds of millions (billions?) of messages broadcast daily into the social media space, and there are many non-obvious hints that users intend to make a purchase. Sometimes, the users themselves are not even aware about their intentions!
With a thorough analysis of conversation threads, combined with a social user graph or browsing behavior, those hidden signs of intent can be discovered.
For instance, announcing “I just bought a laptop” gives a reason to assign a high probability that this user might be interested in buying a printer, various software packages, or laptop accessories.
Topic recognition can be employed to understand certain state in the message such as buying a laptop, or feeling bored. Then those states can lead to co-sell or upsell opportunities.
Further, sequences of conversations can be analyzed to extract buying intent.
If a user enters various discussions about smart phones, vigorously defending the iPhone against others, this user is a much better opportunity than a random person to be an Apple customer, even though the purchase intent is mentioned. Association rules for learning can be used to identify this type of implicit intent.
To handle more complex situations, analyzing user connections can identify interests and opportunity graphs using clustering in machine learning.
The interest graph can provide insight on demographics and user social behaviors, leading to specific qualities for a certain set of users who can have similar interest and may react similarly when faced with same offers. This means these clusters of users may have the same intent to buy certain items.
This process of creating useful clusters using machine learning then allows the human intent to be discovered. Which is where the next part comes in – how to approach these opportunities…
Make sure you come back tomorrow for part two, where we take the legwork of what we do behind the scenes and show you the best practices for your success in front of it. Ready to use our tools? Sign up here!