Processing

Natural Language Processing simulates the human ability to understand language by tagging parts of speech, such as named entities, keywords, and concepts, in order to help machines "read" the text.

For example.

• "It's amazing" is more positive than "It's good."

• "It's very good" is classed as more positive than "It's good", given the multiplier effect words such as 'very' produce.

• "It's not very good" is classed as negative. Although the phrase contains 'very good,' the word 'not' is taken into consideration.

• "It's amazing, but", receives a lower score than "It's amazing" as 'but' introduces a caveat to the equation.

The software assigns a sentiment score for each minute of transcript. It also produces an aggregate, overall score of positive, neutral or negative.

Sentiment Analysis only takes the transcript text into account and does not factor in additional data such as Facial Emotion Analytics or Filters data.

Accuracy is dependent on the quality of transcripts and the quality of the English translation. Cultural variations, slang, sarcasm, and other variations in the spoken word may complicate analysis.

Visualization

Sentiment levels are displayed in the Sentiment timeline chart on the Media View page in Video. Sentiment is displayed on a minute-by-minute basis. The analysis also provides an overall sentiment score displayed at the top of the chart.

Click anywhere on the Sentiment timeline to jump to the moment in the video.

Sentiment appears as a Cognitive filter in the Media Library. Each media item has one Sentiment filter representing the overall sentiment emotion.

The Analytics page can display an aggregate view of Sentiment Analysis data. The chart lists every sentiment emotion identified in the channel and the number of videos with that identified sentiment. For more information, see Analytics Module display.