I like to eat. More specifically, I like to dine out. I dine out so often that even my wife complains about the cost. And because I go to restaurants a lot, I began writing Google reviews a couple of years back. And like me, there are thousands of consumers every day, happy as well as frustrated ones, who go online and share their views and opinions across various social media platforms. The question is, are these reviews just a vent where people share their experience, or can these reviews be used in order to improve the business and attract more patrons. Let’s dig deeper into identifying the best sentiment analysis approach that is used to do just that.
There are few social activities that people discuss and debate more than their choice of where to dine. Eating out is emotional, as it engages almost all the senses while representing personal taste and status. This is why so many people read restaurant reviews, and others choose to write them. These types of reviews are jammed full of emotionally descriptive words like “love”, “hate”, “amazing”, “ok”, “horrible”, “best”, “worst”, etc. Making sense of them is a challenge as many reviews possess more than one emotion making it tough to know whether a review is in fact positive or negative.
Unfortunately, all those cute star ratings don’t always help either, as often what is written and the star ratings are given do not always match up one to one. 5 Star Rating systems are highly subjective and often people just don’t really understand the scoring scale. The other challenge is the actual review volume. Some restaurants have thousands of reviews across different social media platforms, how do you analyze and summarize them quickly enough to be useful.? This is where sentiment analysis is most helpful. Sentiment analysis is a process usually used by marketers to better understand their customers, or what is often referred to as the voice of customer. They usually do this by conducting social media listening. Understanding customer feelings, opinions, and motivations are important to advertising better to target audiences.
Sentiment analysis is the exercise of identifying, scoring, and classifying peoples’ feelings expressed through any form of text data as positive (+1), neutral (0), or negative (-1). This helps marketers make smarter, better choices in the tactics, channels, and creatives they use to message consumers. When it comes to examining restaurant reviews, the complexities and challenges of analyzing the everyday language used in social media, mobile food apps, and hospitality websites, quickly comes to the surface. That is the reason that automation, through machine learning, is needed. Understanding people’s written emotions isn’t easy, especially at large scale, that’s why there are various ways that natural language processing is used to make sense of the judgments people make about food, customer service, drinks, and other related topics within hospitality.
There are three basic ways that sentiment analysis can be conducted to make sense of the emotions and opinions people make when writing restaurant reviews. They are called Document-level, Topic driven, and Aspect-based sentiment analysis. The easiest way to understand them is by taking apart some example reviews.
Document level sentiment analysis is done by conducting sentiment analysis on a statement as whole. It is simple as long as there is only one sentiment in the complete document but doing the same on a bigger and more complicate statement can give conflicting results while diminishing the effectiveness of sentiment analysis.
Let’s begin with a simple review:
Went to Bar Chef last night and loved their drinks.
The above example can easily be analyzed and classified as positive, on a document-level as it is a short composite sentence with one expressed, obvious sentiment. In basic text analytics, most short phrases or sentences can be considered documents. Now, this is easy enough because the review is brief and possesses only one sentiment. But this approach quickly begins to show cracks and weaknesses once a document becomes longer and more complex with multiple emotions being conveyed about various topics, subjects, or things.
Topic based sentiment is the application of sentiment analysis on a list of specific topics that are known ahead of time and used for easy keyword identification and aggregate scoring. Topics within the restaurant industry can include things like drinks, food, reservations, customer service, cleanliness, accessibility, etc. Now consider how things change once the review above is expanded:
Went to Bar Chef last night and loved their drinks, but the food was horrible.
This review can be analyzed at a topic level with a positive sentiment for the topic “drinks”, and a negative sentiment for the topic of “food”. By restricting the scope to only specific topics and the insights that document-level analysis fails to accomplish can be achieved. Yet even topic analysis can fail to truly capture the most granular insights that lay hidden in restaurant reviews. This is made obvious with reviews that are even longer, complicated, and more detailed.
Topic based sentiment analysis isolates the sentiment for each topic and ensures that no nuance is lost. For each specific topic found in the text, Topic-based sentiment calculates the sentiment and returns the numerical sentiment score. When done across a large dataset, insights quickly surface, and sentiments are unlocked for each topic of interest.
Aspect-based sentiment analysis goes one step further than typical sentiment analysis that automatically assigns sentiment to predefined categories or aspects. With aspect-based sentiment analysis, it can be distinguished which features of a product or service offering are liked and which ones can be improved, without knowing them in advance. It involves breaking down reviews into smaller segments, allowing more granular and accurate insights from data. It is best for open ended responses, like in customer calls, voice notes, podcasts and video content analysis.
Went to Bar Chef last night and loved their drinks, especially the martinis, but the food was horrible. My nacho tasted microwaved and the calamari was rubbery.
This review needs to be analyzed at the aspect level sentiment with further aspect insights on Drinks via martinis, and Food insights are revealed through the aspects of nachos and calamari.
In typical reviews, consumers often touch on many aspects of a product or service. Complaints or praise for price, quality, or ease of use can all be mentioned in one comment. Aspect based sentiment analysis determines first which categories are being mentioned and then calculates the sentiment for each of those categories. When compiled in aggregate across a large number of reviews, the strengths, and weaknesses of a business' aspects like product or service feature surface quickly and actionable insights become obvious instantly.
When it comes to sentiment analysis, there is really no “best” methodology but only different approaches for various use cases. In hospitality or restaurant reviews, the best approach is always the one that provides the greatest degree of granular results and tangible insights that can be used to make a real difference to your business. Our data tells us that about 15% of restaurant reviews are complex, demonstrating two or more sentiments. That is a lot of vital intelligence that requires a more sophisticated approach such as topic driven or aspect based. Ultimately, aspect based sentiment is going to provide you the best results if your business attracts customers who tend to write long, complex, and detailed reviews. The most important point is that you begin to apply sentiment analysis to your text data if you have not already. That is the first best step.
Repustate’s sentiment analysis API processes thousands of hospitality reviews per day for hundreds of clients. It enables real-time social media sentiment analysis and even saves the unforeseen PR crisis. Our AI-powered software provides both Topic Driven and Aspect Based Sentiment Analysis for the best, most accurate results in 23 languages. Our API can process 1,000 reviews per second or 3.6 million per minute.Repustate uses AI-powered machine learning and deep learning techniques to identify, classify and score text that specifically expresses sentiment. With Repustate, you can understand your data, customers, & employees with 12X the speed and accuracy.