Sentiment analysis MBK Fincom ProduceShop

Sentiment analysis in a data-driven company8 min read

Choosing your business path is a step that precedes the entire settlement of your business. Data-driven companies already know that among the vast amount of data to be analysed, the most important to be taken into consideration are those relating to user sentiment. MBK Fincom, which since its foundation has been involved in data analysis and the creation of solutions, explains with the help of some of its experts what this approach consists of.

Foreword: sentiment analysis is only one of the possible (and necessary) operations in a data-driven business perspective; moving through the mass of data that the web today allows us to collect, in fact, requires careful discernment based on sources, types and, above all, targets.

Analysing the sentiment relating to your brand, product, or even specific niches or keywords, can help in a number of strategies. Starting from a more customisable marketing design, we move on to finance and procurement strategies that are more in line with the ‘real’ demands of the market.

So let’s try to understand what this analysis is all about, and how its correct implementation can provide significant benefits to a data-driven company.

What is Sentiment Analysis (SA)?

As with any concept about digital, we can’t define it in one simple way; in general, it’s a set of activities and practices designed to listen to, analyse and exploit the emotional weight of opinions expressed by users on all kinds of online platforms.

Also called Opinion Mining, the two branches of study that have given material and birth to the concept are computational linguistics and textual analysis. The concept of opinion is central; in fact, the definition includes the verbal response, expressed in writing, and often on the spur of the moment and with great sincerity, of the emotional situation of the user (called in this context “opinion holder“) at any point of contact with the brand/product/service/situation.

Obviously, the relevant data should be sought in the most capacious ‘containers‘: social media and online reviews. In these types of platforms and modalities, in fact, users freely and directly express all their moods, emotions and opinions; both the search and the selection of suitable material, however, must be carried out and processed according to specific actions.

What are the steps that constitute the activity of SA?

The best practices associated with the process (carried out by a machine, let’s be precise), are normally set to certain standards

  • keywords scoring: this involves finding in a source all the simple words linked to the emotional sphere (beautiful, ugly, bad, sad, happy, angry). Based on the machine’s knowledge or instructions, the words are given a positive, negative or neutral score;
  • lexical affinity: widening the spectrum of the first step, this methodology not only detects emotional KWs, but also assigns other words, chosen in an arbitrary manner, a probable affinity to certain emotions, so as to refine the primary selection;
  • statistical methods: based on artificial intelligence (AI) and the machine learning process. They are not immediate and spontaneous, but require the construction of associative models and grammatical instructions; in this process, in fact, the relationships between the words used are analysed through an in-depth analysis of the latent semantics, i.e. the ‘rules‘ of relationship between the parts of a text that the machine can interpret according to patterns that are in any case unknown, and thus give them a value. The process is even more complex and meticulous, because it must imply the ability of the system to give a bias not only to simple sentences, but to entire topics and complex constructs;
  • techniques at conceptual level: this is the analysis of subtle, more complex semantic structures, and is the part that most risks generating problems for the machine in the attribution of a scoring.
Sentiment analysis: satisfaction metrics
Sentiment analysis: satisfaction metrics

Levels and methods of analysis

Each element taken into analysis must be controlled according to different levels, since it is not sufficient to work only on the meaning of the words themselves.

The first parameter to consider in a comment, review or content is the Tone of Voice; although this is often quite clear and obvious (especially if we are talking about an escalating/angry tone in the case of complaints), sometimes elements such as sarcasm or irony could be a point of confusion for the machine.

The intensity of the comment, i.e. the strength (verbal and paraverbalfont, font size, punctuation) of the data, helps to better define the ToV; it also helps to measure the ‘urgency‘ of the eventual comment/review.

Linked to the lexicon, to the presence of possible additions such as emoji or emoticons, and to other paraverbal elements, is emotionality, i.e. the emotional categorization of the content.

The relevance of the text, on the other hand, is the relationship of coherence with the main theme being analysed; for example, a comment on ‘weather‘, even if negative, would be of little relevance in the case of the analysis of a linguistic text.

Finally, the semantic analysis; thanks also to the most recent implementations of Google’s algorithm in this sense, we are talking about one of the most complex activities to perform. In all languages, in fact, it happens quite often that the same word takes on different meanings or nuances depending on the context, the position in the sentence, the way it is written.

Since these analyses are carried out on large amounts of data, there are of course specific tools for sentiment analysis; however, everything can be implemented by human input. The better the program is able to navigate by itself among the various nuances of human language (and therefore the better NLP (Natural Language Processing) mechanisms are implemented in its AI), the more valid it will be as a tool for sentiment analysis (and the more, plausibly, it will cost the relative licence).

Limits of SA

The limits of this kind of analysis, however, are the limits of the machine itself; a software, however perfect, will hardly be able to understand specific nuances and shades of human language. We are talking about sarcasm (“a broken product is what I was looking for, really great! “), irony (‘huge congratulations!’, following a negative trade or a non-response), slang (‘what a slam!’) or misleading or ambiguous context (‘I’m not going out because of the weather‘, where it can mean both ‘time-lapse‘ and ‘weather‘).

In addition, at the current development of AI technology, machine learning processes still require time frames that are far from immediate, and often implementations/corrections of a human nature.

Data-driven companies, success linked to data analysis
Data-driven companies, success linked to data analysis

How can this be exploited in a data-driven company?

We posed the same question to Ronny Soana, chief marketing officer and COO of ProduceShop, who explained the different applications of sentiment analysis across industries, as well as established in their company.

“Although many people often think that Data Analysis, in general, is for the exclusive use of the marketing sector – Ronny replies – in reality an entire company can easily benefit from it (if properly trained and structured).

We are obviously talking about data-driven-businesses, right? So we are talking about situations in which the “fishing” of data and their analysis, “sifting” and graphic representation are the daily bread of every manager and departmental analyst.

Let’s think about MKTG, for example; we will see that, by analysing the sentiment of customers and target users over a specific period of time, and in relation to keywords linked to a product, a category, a fashion, it is possible to design and draw up appropriate and customized campaigns; moreover, it is an indispensable tool for many other activities, such as monitoring the perception of the brand by users, the management (and prevention) of possible crises in a sudden manner, the in-depth analysis of competitors (with all the necessary benchmarking operations); finally, the effective measurement of certain KPIs.

If we take a look at other departments, we can see that SA can help all of them:

  • in finance it is an excellent tool for assessing markets;
  • it helps the purchasing department to monitor both what users feel about specific potential products and the desirability of certain markets;
  • Evaluating customer feedback is crucial for customer care, as is evaluating media and trends for IT.”

In conclusion

Having weighed up the methods and benefits, the question remains: is it really necessary to rely on this kind of analysis?

Our opinion on this is favourable; years of best practice geared towards this modus operandi have meant that the results have been measured positively. Moreover, the number of companies that have decided to implement their strategies with an adequate sentiment analysis is easily visible. They are the ones we interact with every day.

Sources:
  • Corporate PR
  • Finance Dept. ProduceShop
  • IT Dept. ProduceShop
  • Marketing Dept. ProduceShop (https://mbkfincom.com)
  • Aruba
  • TalkWalker
  • InsideMarketing
  • Monkey Learn
  • Denodo
  • KPMG
  • Digital 4Biz
Share
Copyright 2024 © All rights Reserved - MBK Fincom Sa | Privacy Policy