Industry machine learning-powered “custom” classifiers are becoming increasingly critical tools to maximize value and impact of social data in your organization. They accelerate the application of AI to your social listening data utilizing the latest technologies that often approximate, and in some case exceed, human performance. As a result, they have demonstrably proven to significantly improve data accuracy, classify even nuanced conversations by learning through human input, help bridge the gap between social listening and consumer insight and enable greater organizational adoption in areas like customer care and customer experience.

As Forrester Research says, “The cost to successfully implement (machine learning) and produce useful models can range into the hundreds of thousands or even millions of dollars due to the need to train them on large, clean data sets and the time needed to experiment with several different models before deploying into production. Using prebuilt models [such as the below] from cloud-based platforms [like Brandwatch] can be much more cost-efficient…” (AI Tech Radar 2017)

Converseon now offers a wide range of prebuilt and custom models to rapidly enhance the value of your social listening efforts. These models have been created via Conversus, the first Machine Learning as a Service Platform specifically for social text analytics. The models are immediately deployable to clients of most social listening and management platforms or can also be used directly. Simply subscribe and deploy for immediate results.

All models are designed for high precision and recall (F1 scores), enabling immediate use with confidence.  They have also been tested to ensure superior performance when “generalized to new data.

Industries currently covered include Financial Services, Insurance, Hospitality, Airlines, Telecom, Enterprise Technology, Retail, Food & Beverages.  New industries and models, including non-English, are added on an ongoing basis.

If you have a special custom classifier that is not reflected below, they can be designed, built and deployed in as little as two weeks depending on complexity (see “Custom Models” below).


Current Pre-Built Advanced Classifiers

    • Advanced Sentiment & Emotion: Raise the precision and recall of your data through industry-specific classifiers.   Language often differs by domain (“small” may be good for smartphones but not good for hotel rooms, for example).   Advanced industry sentiment and emotion provides users with the combination of high precision together with superior recall that captures every relevant expression in the data (even in mixed sentiment sentences) so that you don’t miss important signals.  These classifiers generally provide a very high .80 F1 score and available for immediate use.


    • Voice of Customer: (VoC) is used to describe the needs and requirements of the customer in social media. It is the process of capturing all of what a customer is saying about your business, product, or service. It helps you visualize the gap between your customer expectation from your brand and their experience. A VoC classifier allows you to speed time to insight understanding drivers of best vs. worst customer experience


    • Trust/Distrust: Trust is at the heart of building a brand, Brand trust reflects a consumer’s expectation that a brand’s product or service and corporate behavior reflect the promises the brand makes.  Trust in institutions is in constant flux, yet multiple studies demonstrate that brand trust is critical to business success.  Critically, this classifier captures expressions of trust when it is implicit, and specific trust-related terms may not be present.  For example: “I use the product with my baby at night because she sleeps better.”


    • Customer Care: Captures conversations that are highly actionable, legitimate customer complaints and questions expressed across social media channels.  According to one study, 84% of consumers expect companies to respond within 24 hours after posting on social media, while 72% of Twitter complainants expect a response within an hour. Currently, agents’ queues are flooded with social media noise. Productivity is lost in agents reading and deleting social media noise. Eliminating delays in your response – a Customer Care Classifier enables your organization to focus and respond immediately to customers who have a genuine need.


    • CSR: This classifier captures conversations that are specific to a company’s corporate social responsibility efforts.  This is broadly defined as a company’s efforts to improve society in some way.   According to Forbes magazine, more than 88% of consumers think companies should try to achieve their business goals while improving society and the environment.


    • Loyalty & Preference: Captures and classifies expressions of loyalty and preferences to specific brands.  Companies that can establish a strong customer base who have become faithful to their product are at a strong competitive advantage versus those who do not engender loyalty.  Importantly, this classifier captures statements of loyalty even when customers are angered, making this an excellent tool for identifying drivers of churn.  For example: “I’ve been a <insert brand> customer for 10 years, and they still haven’t responded to me about my overdraft charge.  This is the last straw!”


    • Advocacy: Captures and classifies conversations where a person recommends a brand or product and then passes on positive word-of-mouth messages about the brand to other people.  This classifier captures instances where existing customers recommend a brand or its products.


    • Detraction: The detraction classifier is the opposite of advocacy.  It captures conversations where consumers express strongly negative views about a particular brand or product.


Custom/Bespoke Models

In addition to prebuilt models, we offer services to rapidly and effectively build machine learning models specific to your requirements for rapid deployment.  These can often be completed in one to two weeks, depending on complexity. These typically include:

    • Brand Attributes: Our machine learning models can often approximate human performance in the classification of conversations for even nuanced attribute themes.   Examples include attributes like “safely,” “value,” “empathy,” “innovation,” “transparency and much more.   These social attribute classifiers have been demonstrated to be able to predict the results of traditional survey-based brand tracking results up to 12 weeks in advance and can often provide deeper insight into key underlying drivers.


    • Relevance: Separate your true brand and product signals from the noise is a significant challenge for many brands and products.  Job listings, advertising, similar names and much more often pollute the data sets when trying to isolate relevant content. Our machine learning relevancy classifiers generally attain minimum 80 per cent relevancy and high recall, meaning that we are capturing the vast majority of relevant discussion, even on the most granular level.


    • Unique Taxonomies: Not all brands see the world the same.  You may have different views on how you define “sentiment,” for example.  Or you have specific criteria for what is defined as a customer care issue, or specific key steps in your customer journey analysis.   Our custom models can modify existing models to meet your specific requirements and/or design new models from scratch that are specific to your frameworks and taxonomy.


Examples of Advanced & Custom Models in Action

    • A large hospitality brand improved the precision and recall of its data by more than 70% allowing for more confident reporting, organizational adoption and inclusion of the data into analytic/predictive models.


    • A large global food company discovered new consumer trends and was able to predict product sales six weeks in advance through custom “motive” classifiers.


    • A global software company reports using custom models to save more than $10 million annually over traditional survey-based research for buyer’s journey analysis.


    • A global Midwest based manufacturer reduced the cost of operation of their social listening by 30% and accelerated insights reducing the time needed for data cleansing and human intervention.


    • One of the world’s largest CPG company was able to predict its brand health scores 12 weeks in advance through custom attribute models.


    • A global enterprise technology company is using VoC models to predict their social net promoter scoring and customer satisfaction.


    • A large automotive company was able to raise the relevancy of the data classified as “customer care” from 20% to more than 80% through customer care machine learning, reducing latency, lowering operational costs and improving customer experience.


For further discussion and scoping of customer models, please contact us directly at