Ship early and ship often, yet make sure that what you’re shipping is better than existing alternatives.”
Interview with Vidar Brekke, Chief Product Officer at Converseon. You can follow Vidar on twitter: @iVidar
Tell us more about Converseon
Converseon is a 10-year old, independent, full service social media agency and consultancy, with offices in the US and Europe, serving primarily Fortune 1000 clients. We have been recognized by Forrester and Gartner as a leader in Social Intelligence, due, in large part, to our proprietary text analytics technology and social insights practice.
What API(s) does Converseon offer today?
We just released ConveyAPI, a powerful text analytics engine that yields near-human accuracy around sentiment, emotion, intensity and relevancy of social media text, from short tweets to long blog posts. It offers analysis at a document, sentence or named entity level.
The initial set of features was designed to attack the two biggest challenges (or should we say, dirty secrets) in the social text analytics field:
The belief that Boolean logic offers an effective approach to filtering relevant content from social networks, or other large datasets.
We know that this is not the case, as your queries either end up being too broad, resulting in more noise than signal OR your criteria are too exclusive, resulting in missing relevant content.
For example, imagine being a brand manager at Ford Motor Company. What would the Boolean algebra look like for tracking twitter sentiment around the various car models? You’d want to exclude Ford Models (the modeling agency) as well a Betty Ford, Gerald Ford or Harrison Ford, except in the event the actor is reported driving a Ford, or when there is trouble at the Ford dealership in Harrison, NY.
We’re solving this problem by taking a machine learning approach to determine relevancy that is both effective and flexible, by allowing our clients to train their own relevancy models and access them via our API. With very little training data and training time, the results are astounding.
The second dirty secret is that automated sentiment analysis is as accurate as coin flip.
This is true for the majority of first generation tools that rely on what’s known as “words spotting,” a technique that fails to detect sarcasm, slang or understand that people dislike small hotel rooms but love small cell phones. Furthermore, this approach doesn’t work for the many instances where text lacks both positive and negative words. For example: “I stood in line for most of my lunch hour at my local bank today.” Humans get that the author is not a happy camper.
This is where only machine learning approaches, like Convey API, will work. Our sentiment classifiers are trained on actual human analysis, yielding well over 90% accuracy of analysis out of the box for most domains, compared to humans tasked with the same analysis.
When did Converseon realized that it needed an API?
As social data is starting to prove relevant beyond traditional marketing and PR use cases, we found that not all needs could be met with our vertically integrated, listening and monitoring platform.
A new category of clients approached us with the need to integrate social text analytics into existing BI and CRM tools, recommendation engines, ad optimization tools, search applications, as well as innovative consumer applications. In order to serve these markets, we choose to break out key text analytics capabilities via an API.
What recommendations and tips would you give to a company planning to launch an API?
Ship early and ship often, yet make sure that what you’re shipping is better than existing alternatives.
Why and how are you using 3scale API Management Platform today?
We are using 3scale to handle authentication, usage, portal tools, documentation and even site hosting. 3scale API Management platform let us quickly build out authentication and tracking mechanisms for our API. It saved weeks if not months of development on that front alone. The portal tools and documentation were also time savers. We did not have to build a lot of the generic plumbing that goes along with all APIs. It has been a great service and we look forward to growing with 3scale as we onboard new customers
What benefits have you seen/are you expecting from your API program?
Immediately after launch we saw developers creating apps from ‘starter kits’ we put up on Github, which is exactly the kind of engagement we were hoping for. We believe that further innovation will come from the community, and we’re curious to see what kind of new applications developers will build with ConveyAPI.
What is your vision for your API?
Our vision is to provide developers with classifiers that offer near human-level quality textual analysis, answering ‘universal’ business critical questions such as content relevance and sentiment. At the same time, we want to give developers the power to train our classifiers to their domain or even train our system to answer completely new business questions for their companies and customers. We are far along on both those fronts and hope to be able to roll some of these advanced features out this year.
ConveyAPI™ is a simple to use REST web service that provides programmatic access to a powerful text analytics engine. It utilizes natural language processing, statistical modeling and machine learning techniques to return content-specific annotations, including Sentiment, Emotion, Intensity & Relevance. ConveyAPI’s annotations yield near-human accuracy and can scale as wide as you need with respect to API calls.