How does the Sentiment Analysis work?

Player XP’s sentiment analysis technology is built utilising the latest machine learning techniques to provide accurate, games industry focused, sentiment output. The sentiment system works on a scale of -99 to 99, where -99 is very negative sentiment and 99 is very positive sentiment.

To determine the position a single record of feedback falls on the scale, Player XP uses machine learning confidence of sentiment based on the appearance of words found to be in mostly positive or negative reviews from our games focused training data.

This confidence is used to determine how positive/negative something is by weighing the confidence against a neutrality confidence using a stochastic process.


What is meant by Constructive Feedback Filtering?

Player XP’s technology offers dynamic filtering of data to help you identify the feedback that matters the most. Player XP considers this filtered data to be constructive feedback, in that this feedback is determined to have semantic value. To identify this semantic value, Player XP runs a number of processes:

– Utilising machine learning classifiers to identify if a piece of semantic data falls inside one of Player XP’s core subject categories and is therefore something of interest about a specific title.

– The remaining data is then checked to see if the content has more than 12 characters or 2 words to ensure that the data is not too ambiguous to derive meaning from reliably.


How do Player XP’s data classifiers work?

Classification of data works by using a neural network trained on a vast set of games data. The neural net uses a semantic map called “embeddings” to represent natural language. A training set of data tailored to the games industry is then used for both semantic and sentiment classification.

The neural net then uses both of these aspects to build a prediction model so that when new data is sent to the classifier, it can make a confident prediction.