AI: An Ongoing Journey at Mention 

12 Years at the Forefront of Artificial Intelligence

Since our launch 12 years ago, Mention has consistently integrated AI to both enhance the user experience and simplify the resolution of our customer’s pain points.

For us, artificial intelligence has been the cornerstone of our product for years, not just a marketing trend to surf on. 

Let’s look back together.

Sentiment Analysis – 2013

It was only a year after launching our service that we realized the importance of Sentiment monitoring. It wasn’t just about collecting mentions, it was about making them meaningful and translating their feelings (positive/negative/neutral) into valuable insights for customers.

We represented the sentiment of a mention with a green/grey/red color code, as you can see below.

At the time, we were one of the few players in the market to offer this functionality. Since then, we’ve published no less than four versions of Sentiment Analysis, the latest of which was released in June 2024. 

But let’s have Bryan explain the latest changes!
The architecture of Sentiment Analysis dated back to 2017, and we felt we needed to bring it up to date in light of technical advances in NLP (Natural Language Processing) and machine computing capabilities. 

The model we built is based on the Transformers architecture (similar to ChatGPT).

Our first step was to pre-train it on 2.5TO of data and on 100 different languages so that it understands the grammar and spelling of all the different languages. But that was not enough! Our model needed to work on sentiment classification. 

We fine-tuned the model by gathering a pool of 500k mentions and asking it to classify between positive/negative/or neutral. 

The result lived up to my expectations, with an accuracy rate of 82% on the 5 main languages. 

My next step will be to focus my energy on a new project: emotion analysis where I’m going to classify mentions based on Anger, Joy, Sadness…
Bryan Sichoix – Machine Learning Engineer @Mention

Pulse Email – 2015

Given the success of the Sentiment analysis feature, our technical team continued their investigations in the AI field.

Collecting mentions, being able to categorize them automatically, and knowing the sentiment was already a good foundation. Still, the sheer volume of data meant that we could offer customers even more, especially those facing crisis, by alerting them in real time to unexpected changes in data volume. Thus, a new use-case was born: Crisis Management. 

“We’re really talking about basic artificial intelligence here, but artificial intelligence nonetheless. In practice, it’s pretty straightforward: by creating alerts, customers collect mentions. The more you collect, the more you have. The more data you have, the better you are able to spot trends and anomalies in those trends. 

That’s Pulse: the ability to alert our customers by email when an anomaly in the volume of mentions is detected.”

Andrey Virodov – Lead Dev @Mention

Want to know more about the Pulse email (now ‘Spike’ email)? Check out this article.

If there’s one thing we can be sure of, it’s that since these early developments, technological advances have enabled AI to be widely adopted. And that’s precisely what we’re building for our customers right now. 

So, let’s go over the latest developments.

A few months of focus, a few months of success 

Yes, the title may seem a little presumptuous, but objectively speaking, we have to admit that our product and tech teams have spoiled us recently with the recent releases:

• AI Alert Overview: a daily AI-generated summary about all your followed topics
• AI Digest Emails
• AI Sentiment Analysis V4: an enhanced AI Sentiment classification with an 82% accuracy score
• AI Report Highlights: ​​AI tracking of significant variations in the subjects tracked

I haven’t lied to you. These are truly impressive successes, and I invite you to explore them with me.

AI Alert Overview – July 2024

By leveraging our dataset and centering customer feedback, we arrived at a few insights:

• On average, only 6% of collected mentions are read by our customers.
• Our customers want to spend less time going through mentions. 

How do we maximize the comprehension of trends? How do you stop guessing about tendencies randomly? 

It was on this basis that Alert Overview was developed. 

With the Alert Overview, users can access key KPIs for their alerts—such as volume, reach, and sentiment trends—while also receiving a detailed daily AI summary of their mentions. This helps users quickly identify the main topics to prioritize.

The result is clear: 

• The overview offers a complete analysis of all customer mentions and helps you remain aware of vital information 
• Our customers can now save an average of 6 hours per month thanks to this feature. Almost the length of a working day!
• Our customers now have all the tools they need to identify passing trends and isolate perennial ones to capitalize on

AI digest emails – July 2024

In the wake of the launch of Alert Overview we have also redesigned our communications so you can readily access your alert’s most important information 

• Summary of your latest mentions and topics associated
• Identification of trends and related mentions
• Top Mentions sorted by influence score
• Qualitative KPIs such as: Sentiment, Reach, Volume

And as usual, we leave it up to our customers to define the recurrence (daily, weekly) and the scope of analysis (whether to analyze all of your alerts together, or each alert separately). 

But let Audrey, our PM in charge of this new feature, explain its scope.

“With digest emails, we’ve improved the design for faster reading, enhanced the statistics section to properly explain what happened during the specific period and also implemented AI to provide better insights. The aim of this digest redesign is to offer you a quick-to-read summary and give you a global overview of the latest activity on your alerts!” 

Audrey David – Product Manager @Mention

AI Sentiment Analysis – June 2024

Everyone wants to know how their audience feels about their brand, their competitor, this or that technology… And there’s a simple reason for this: understanding the sentiment around a subject is a virtual guarantee of being able to perceive its value. 

For this, we have to consider that there are between 5,000 and 10,000 languages. Some are much more widely used than others since 80% of humans use 80 languages, and only 20% of the thousands of others.

Given the complexity of human languages and the rapid evolution of technology, regularly updating our sentiment models is crucial. Relying solely on keywords is insufficient; effective sentiment analysis must support multiple languages to capture global perspectives.

But how do you make sentiment analysis operational with so many variables, and above all, how do you achieve a score of 82% accuracy?

Wanna know more about Sentiment Analysis? Check this article.

AI Highlights – May 2024 

Highlights is our first foray into AI in 2024.
It started with our Reporting functionality. 

We designed this feature to help you quickly identify the most relevant conversations in your reports, making it easier for you to stay informed and act promptly on key insights.

Indeed, our reports were particularly effective in quantitative terms but lacked the qualitative elements that would make them easier for users to understand. 

With Highlights, our customers can now : 

• Identify trend variations at a glance 
• Detect abnormal changes around their alert without scaremongering
• But above all, understand the reason for these changes without having to analyze every single mention. 

Once again, our AI developments have led to two major benefits for our customers:

• Considerable time savings 
• Increased value in qualitative analysis 

Want to know more about Highlights? Click here.

Cloud Words Feature – October 2024

We’ve made a major upgrade to our Cloud Words feature! By switching to a more advanced model, we’ve significantly improved how keywords are extracted, especially in non-English languages.

• 🎯 More Accurate Cloud Words: With better keyword identification, your Cloud Words will now more accurately reflect the most important topics and trends in your mentions.
• 🌍 Enhanced Multilingual Support: The upgrade provides more reliable keyword extraction across multiple languages, making your Cloud Words more insightful no matter the language of your mentions.
• ⚡️ Faster Processing: We’ve optimized our system to handle large datasets efficiently, ensuring you get your Cloud Words quickly, even with extensive data.

These enhancements mean your Cloud Words will be more relevant and precise, giving you clearer insights into the key themes and conversations around your brand, regardless of the language.

AI Emotion Analysis – October 2024

That’s it! We’ve enhanced the analysis of your mentions and Sentiment by adding Emotions.  

Emotion Analysis goes beyond traditional sentiment analysis by identifying the specific emotions behind each comment, post, or review. 

Our proprietary AI model now recognizes up to 7 feelings: joy, sadness, anger, surprise, disgust, fear, and neutral. 

Thanks to emotion analysis, our users can now: 

• Gain a better understanding of your audience: Understand not just what your audience is saying, but how they feel about it. Emotion Analysis provides granular insights into customer experiences.
• Create stronger connections with your audience: Respond to your audience with empathy by understanding their emotional states, leading to stronger connections and loyalty.
• Build clever strategies: Use emotional data to tailor your marketing and communication strategies, ensuring they resonate on a deeper level with your audience and customers. 

As you can see in this example, those two mentions have Negative Sentiment but not the same Emotion. 

How is it possible? Let’s ask Bryan:

Emotion Analysis is the logical continuation of our team’s efforts on Sentiment Analysis. We started with the Open AI GPT 4o mini model, then created a data set of one million mentions to train our own model and let magic happen.

Those two examples speak for themselves, by reading the mention, you understand why they have been classified this way. In short, we’re leveraging the absolute best in NLP technology to give you a tool that’s both highly intelligent and accurate. With this algorithm, you’ll get insights that are not only actionable but also deeply reliable.
Bryan Sichoix – Machine Learning Engineer @Mention

Stay tuned for more updates as we continue to innovate and enhance your experience with Mention. If you haven’t already done so, start your 14-day free trial!