Smart Blog

Read Through The Lines In Reviews With Sentiment Analysis

Davide Avella,

Read Through Lines of Reviews With Sentiment Analysis

Did you know that out of the five senses at our disposal, hearing is the only one which can switch on and off at will? We can't choose not to see or not to taste, but we are perfectly unable to listen. Or rather, we can hear without necessarily listening. And this is where we draw our conclusion: if hearing indicates a passive process, listening requires attention and cognitive effort.

The effort involved in listening to their customers is one of the things that distinguish successful businesses from those that aren't. In a world where line between online and offline are blurred, the retail sector also finds new spaces and channels to reach consumers and develop new ways to shop: e-commerce portals, social media and more.

The conversations that brands have with its customers continue to increase at a dizzying rhythm. At the same time, reputational risks are also increasing. Product descriptions may be incomplete or too detailed. Consumers' moods can be affected and a bad review may discourage a new customer from buying a product. Even a particularly negative comment on a social network has a multiplier effect and can result in a crisis.

Marketing and communication experts advocate a single strategy: listen. Listen carefully and try to communicate with your customers. It sounds simple but it's a winning strategy that requires a great deal of effort in many ways.

To address this, Business Intelligence provides new tools to retailers for text analysis - a branch of data mining focusing on textual data. The implementation of automatic learning algorithms can detect the tone of each text and translate it into a certain level of positivity or negativity, thus identifying the polarity.

The challenge is to try to capture sarcasm, irony and all other natural language features. All this takes the name of Sentiment Analysis. A valued ally in dealers' hands to improve the quality of online communication!

tablet-happy-woman.jpg

Sentiment Analysis works like any prediction model. After building a training set, the algorithm is "trained" on analyzing all the features of the texts to be assessed. These features can cover all words that make up phrases or n-grams.

An n-gram is a contiguous sequence of items within a text that can match phonemes, graphs, syllables, letters, and whatever else can provide a greater precision than the single words considered one after the other. Once the training is completed, the model is able to detect the polarity of a text, transforming itself into a 360° automated listening tool.

Polarity of the Text

The polarity of a text, being positive or negative, can be further examined in several respects:

1. Mood

The first, the most immediate one, examines the moods in a short or long text: sadness, happiness, anger, disappointment, and much more.

2. Objectivity & Subjectivity

Another type of polarity is one that sees objectivity and subjectivity. A text may contain objective information such as facts contained within a newspaper article as well as may include subjective information such as political views. Extraction of this aspect is more complex than the first, since it depends more deeply on the context of reference.

3. Attributes of products or services

A different approach, however, is based on attributes, such as a cellphone screen, restaurant service, or the lens of a camera. Taking the polarity from different points of view, since we can have in a long list of reviews on TripAdvisor or any eCommerce site, allows us to focus on the nuances and individual characteristics of the subjects studied, without having to be limited to a general consideration.

In this way, a hotel could reconstruct a more vivid image of its services to its customers. For example, by studying its reviews, it could be noted that one of its strengths is room service but at the same time many complaints about te cleanliness of the room.

Spam detection

Another less used, less publicized, text-classification technique is spam detection. More and more retailers are faced with fake reviews or comments written by fake accounts that do not match real users.

For those who manage the online communication of a brand, it's important to understand if you are facing a real disgruntled user. Text mining classification algorithms can locate a message and label it as spam, allowing anyone who manages the platform to locate it and remove it. The payoff is high. You avoid wasting energy in responding to fake messages, the reporting is timely, there's a possibility that spam will be removed and avoid future problems. Above all, it also reduces the reputational damage of a negative comment as it can be deleted after a few minutes.

To find out more about Sentiment Analysis and to find out all about Business Intelligence for Retail (methods, tools, KPIs to track etc) you can download my free guide. Download the Business Intelligence in Retail guide now!

Business Intelligence in Retail

Condividi l'articolo

   

Commenti

Condividi questo blog

   

Recent Posts