The internet allows us to really listen to people, to understand their thoughts and feelings in their own words. Marketers can seek to learn more about their customers from a variety of online sources – forums, product reviews, and social media posts.
Unfortunately, the methods used to analyze these conversations only give us a vague “sense” of what customers are thinking and feeling – like the widely used sentiment analysis.
Let’s look at the definition of sentiment:
‘A thought, opinion, or idea based on a feeling about a situation, or a way of thinking about something.’ (Cambridge dictionary definition)
The widespread use of sentiment analysis in various forms can be associated with the beginning of the digital era. Brands needed a way to capture customer feedback, no matter how vague it was. At the same time, computer-driven analysis, called Natural Language Processing (NLP), was still evolving, and we could only perform ‘word-based sentiment analysis.
Let’s take a closer look at the most common methods to capture customer feedback and measure sentiment:
This is based on the familiar number or ‘star’ rating customers give a product or service, usually in e-commerce reviews. There is little guidance on how these are defined, so this becomes highly subjective; one person’s definition of 5 can be someone else’s 3 ratings.
This approach aims to measure what customers wrote in their reviews. It originates from the early days of NLP and leverages the ‘tokenization’ of textual data.
Tokenization:
This means splitting a phrase, sentence, paragraph, or entire text document into smaller units called ‘tokens.' Tokens can be words, numbers, or punctuation marks.
The word-based sentiment approach searches for positive and negative words in the review text, which are referenced against a list (databases are shared amongst the NLP community) where the positive words and negative words have been defined, such as
The frequency of positive and negative words is counted, and everything else gets placed in the bucket of ‘neutral.' This feedback is probably not neutral; it’s just not been pre-defined in the list.
This approach is referred to as a ‘deep learning-based transformer approach’ – where the context of the words is preserved. ‘Transformer-based architecture’ (a machine learning model) classifies the review text word-by-word and focuses on the positional context of words, similar to a human-level approach.
This and the arrival of the ‘attention’ mechanism, which mimics cognitive awareness, have been significant developments in pushing NLP to its recent cutting-edge capabilities. This approach isn’t within everyone’s reach; even if it is, it requires a dedicated focus on training models with relevant and carefully ‘labeled’ data.
Let’s have a look at some sample reviews and compare the sentiment analysis approaches:
The first review tells us that the food prescribed by the vet has solved the cat’s problem with UTIs. However, the review has been given a numeric rating of ‘2’, which suggests a negative sentiment (but because the customer’s problem was solved, this low rating could be a human error). Similarly, a word-based approach suggests a negative sentiment because of words such as ‘suffered’ and ‘unfortunately.'
But, if we consider the whole context, the product solved the problem, so it’s a positive review.
The second review shows that although the pet likes the product, the pet owner doesn’t. The average rating of ‘3’ is classed as ‘positive,' so it would be labeled as ‘positive’ from a ratings-based approach. The word-based approach also gives a positive sentiment because of the words ‘good’ and ‘like.'
However, if we consider the context, the pet owner isn’t happy with the product and its smell, so the context-based approach reveals a negative sentiment.
In the third review, the pet owner is happy with the taste and price but finds the kibble too big.
The review has a numeric rating of ‘5’ (obviously positive), with a positive word-based approach because of ‘loves’ and ‘good.' Contextually it is positive; however, there are negative ideas that can be helpful to understand too.
In the early days of the digital and big data era, it was understandable to see how getting just a ‘sense’ was better than having nothing. Given the available capacity, we can and should expect more.
At PetThinQ, we have developed the best contextual sentiment model for the pet category. It’s the basis for understanding what customers think, from their interests and concerns to the language they use to express themselves. This is really what brands need to develop specific content and enhance customer experience.
If you’d like to learn more about how we go further, please read our article, “Taken out of context, I must seem so strange."
The rich text element allows you to create and format headings, paragraphs, blockquotes, images, and video all in one place instead of having to add and format them individually. Just double-click and easily create content.
A rich text element can be used with static or dynamic content. For static content, just drop it into any page and begin editing. For dynamic content, add a rich text field to any collection and then connect a rich text element to that field in the settings panel. Voila!
Headings, paragraphs, blockquotes, figures, images, and figure captions can all be styled after a class is added to the rich text element using the "When inside of" nested selector system.
Headings, paragraphs, blockquotes, figures, images, and figure captions can all be styled after a class is added to the rich text element using the "When inside of" nested selector system.
