This article introduces Performance Profiling's key concepts and walks through the sentiment analysis process.
You have large volumes of consumer-generated opinions for both products and service but understanding trends and extracting actionable insight from the valuable reviews is a huge challenge.
Performance Profiling focuses on the review content and using industry-leading natural language processing algorithms, we can pull out the words that your consumers are using in real-time and, irrespective of the review rating, determine if these words are suggesting positive or negative content. As part of this analysis, our algorithms combine the words found with associated words and plurals to gain a more complete picture and discard words that have little value to the analysis. We then focus on the sentences these words are used in, breaking the text up into multiple sentences to identify positive, neutral and negative sentiment through grading and tracking where and how often the words are used.
- Topics: A word or group of words which are found in your customer’s reviews that are considered significant in the context of the review.
- Topic coverage: The percentage of your customer’s reviews that mention the topic.
- Mentions: The number of times that the topic is found in your customer’s reviews.
- Sentiment: A label or numerical value that indicates a topic's strength of sentiment. Each topic is given a value between +100 (very positive) and -100 (very negative) which is then mapped to a graphic that easily identifies the sentiment score and the number of topic mentions of a certain sentiment. Note that data for service and product are analysed and managed separately.
Let's take an example and see how these concepts apply to real data. Here is a holiday review example:
The resort was fantastic. The food in all the restaurants was lovely, always plenty of choice. Delivery to our room was quick and efficient. You cannot fault the rooms, plenty of space and all the facilities required. We will certainly return. The only area which could be improved is the speed of service at the bar, sometimes you are waiting a long time at busy times. Maybe one or two more staff are required. We look forward to coming back in the future.
From this review, Performance Profiling picks out the following topics:
As part of the analysis, we group words, for example, for the topic 'delivery' we include words such as deliver, delivered and delivery driver, and calculate how many mentions there are for each topic and associated/plural words across all of the reviews being examined. The discarded words in this example are 'long', 'all', 'only' and 'maybe'. The next stage is to identify positive, neutral and negative sentiment.
Using the above holiday review as an example:
- Positive sentiment mentions:
The resort was fantastic.
The food in all the restaurants was lovely, always plenty of choice.
Delivery to our room was quick and efficient.
You cannot fault the rooms, plenty of space and all the facilities required.
- Neutral sentiment mentions:
We will certainly return.
Maybe one or two more staff are required. We look forward to coming back in the future.
- Negative sentiment mentions:
The only area which could be improved is the speed of service at the bar, sometimes you are waiting a long time at busy times.
Here we are examining just one review. If this review was the only review to be analysed, the coverage for each topic would be 100%. This is because all reviews (in this case, just one) contain every topic. If, however, we were analysing 100 reviews and the topics listed above only featured in one review, the coverage for each topic would be 1% (one review out of 100 reviews).
The analysis also returns a score for the sentiment of each mention, this being between -100 (most negative) and +100 (most positive), which is then published as a set of results within either the Service Performance Report (for service review mentions) or Product Performance Report (for product review mentions). These reports allow you to immediately identify the most positive and negative opinions that customers are using to describe your service and products giving you insight that was previously not possible.