4 Key Technologies Behind the Genius of Social Media Listening & Monitoring Software

The term artificial intelligence for processes and software is often overworked and overused. It has become a marketing ploy that sometimes refers to algorithms, which are extremely well-designed but far from being intelligent enough to derive judgment and free will from processing data with no prior programming by humans. AI mainly describes the process of a machine, mimicking certain cognitive functions such as learning and problem solving.

And yet, the statistical methods and algorithms that are mistakenly referred to as artificial intelligence can be utilized to optimize and perfect monitoring and social media listening tools. At Digimind, we talk more about Machine Learning algorithms, also known as Deep Learning, but also about other technologies used in monitoring tools like Natural Language Processing (NLP).

Here are 4 popularized examples of technologies used by the best social media monitoring, listening and analysis tools, making it easier and faster for you to use them, and the results more relevant.

📌

I. Automatic Language Processing

II. Sentiment Analysis

III. Intelligent Automatic Classification

IV. Visual Listening

I. Automatic language processing

Automatic natural language processing (NLP) will make it possible to qualify and group together millions of pieces of information and messages collected on the web, chat rooms, and social networks.
This technology works with hundreds of languages and dialects, and with statistical and semantic analysis models to make data analysis easier.

Example: Clustering of Data

NLP can be used to automatically extract and recognize key concepts and group them together in data visualizations so you can see, at a glance, what we are talking about by analyzing hundreds or even thousands of messages about your brand. There are many possible "dataviz" but we can mention for example data clustering (grouping in coherent clusters by theme), hashtag analysis, the wheel, or even the more classic tag cloud.

According to the clustering view below, based on the analysis of all types of Chanel brand messages, the NLP will group more than 2500 messages into large themes (interactive and clickable to display the relevant messages) concerning the luxury fashion brand.

• the brand,
• a perfume,
• market research,
• other market players,
• other Chanel products, etc...

NLP to better group data by theme

Example: Qualification of Posts and Mentions

To save you the tedious task of reading hundreds of mentions, automatic language processing aided by machine learning will make it possible to label any post, tweet or comment on social media. For example, on the Instagram post below, the algorithm "tags" the mention according to criteria of brands, products, characteristics... It also automatically determines the language, country or city of origin.

The real advantage is the flexibility of setting up your own labels that perfectly match your sector, your brands. And then? Machine Learning takes over and learns from the mentions collected to understand new words and concepts that correspond to your labels and will know, for example, how to spot a new skincare product or make-up routine.

Machine learning to better qualify digital mentions

II. Sentiment Analysis

Sentiment analysis (or tone analysis) applied to social media listening (your mentions are qualified as positive, negative or neutral) has been talked about for a long time. And there has often been criticism that the automatic qualification of a mention’s sentiment by the algorithms is too low, especially for a one-sentence tweet! But here's the thing: with the strengthening of NLP algorithms combined with machine learning, silence (unqualified mentions) and errors are now greatly reduced.

Indeed, machine learning will be fed initially by the examples that you submit to the machine, which will then learn from all the other messages it collects, depending on the context of the keyword (other keywords, tone, associations of keywords, etc.). Over time, the algorithm will build a library of posts and expressions to enrich its lexicon with positive or negative expressions. Result: a 30 to 75% increase in the relevance of the sentiment analysis of your messages.

Example: in the message below, NLP combined with machine learning can determine a positive sentiment for a short tweet containing both positive and negative terms.

Machine learning to better analyze feelings

III. Intelligent Automatic Classification

Machine learning also makes it possible to tag and organize the data collected for more effective categorization and classification.

For example, machine learning will understand from your usage that when you monitor "Target", it’s the US retail chain that you are interested in but not the "targets" at all. Similarly, it will understand that the messages relevant to you are about the "Boulanger" chain of shops and not about artisanal bakers. And this, without having to list all the keywords to specify your query such as shop, electronics, internet, phone- NOT target NOT sport NOT bakery, etc.

After a learning phase with you, the user, machine learning combined with NLP allows you to better understand the true nature of the mentions associated with each tag and to automatically tag all the mentions collected, for a quick and relevant organization of the data which will then be classified. This classification will be used to select in one click the relevant data for your analysis graphs, reports and dashboards.

Machine learning to categorize data

IV. Visual Listening

Deep learning allows images to be analyzed in detail. Deep learning algorithms also require data to learn and solve problems. It is indeed used to solve complex problems where the data is large, diverse, and less structured. It learns progressively from raw data and previous experiences and corrects itself without explicit and prior programming as in the case of Machine learning.

In the context of social media analysis, deep learning recognizes logos and images to reveal the presence of a brand not mentioned in a text but also a type of object, person, or environment useful for understanding consumer behavior.

Deep learning for visual listening

In the first example above, using a classic keyword approach, the algorithm would not have been able to detect the presence of the brand "Nike" which is not mentioned in the text of this Instagram post. Deep learning, on the other hand, detects the wing of the Nike logo on the shoe.

In the other two examples, deep learning recognizes the environment (mountain) or people in a situation (woman doing gymnastics).

Image recognition via deep learning (visual listening) therefore offers multiple possibilities to brands, agencies, and advertisers:

• recognition of logos that are present alone, without any text mention of the brand (Nike's Woosh, Lacoste's crocodile, Carrefour's "C", Shell's shell, Starbucks' siren, Apple's apple...).
• recognition of the consumer's environment (sea, mountain, city, home, family, work)
• recognition of gender, practices, and activities