Text Analysis - Sentiment Analysis: User Guide

Text Analysis features are available for text sessions and AV sessions using voice transcriptions upon completion of the session. It can be found in the reports section of Portal.

When entering the text analysis page for a session that has not previously has data generated, you will be presented with this screen:

This data can only be generated once, so please ensure that you are happy with the source data before clicking the button. Voice transcriptions are used for AV sessions, and the respondent chat log is used for text sessions.

Once the data has been generated, the main text analysis dashboard will be loaded. This is composed of five main components: word cloud, overall sentiment chart, recognized entities, sentiment over time chart and a table of all voice transcriptions or respondent chat logs. 

Word Cloud

The word cloud shows up to 100 key words and phrases used within the session. The size of each word is determined by the number of times it was used in the session. There is a default filter applied to the word cloud which removes some words which are generally deemed to not be useful. There is a filters button in the top left which allows you to customize this behaviour. Below is an example word cloud with the default filter applied. Looking at this, the words 'bit', 'kind', 'lot' and 'one' aren't particularly useful and so we should create a filter to remove them.

Clicking the filters button brings up a modal showing all the filters available for your current project. As we haven't created any filters yet for any sessions in this project, none are currently available.

Click the create filter button to be taken to the new filter interface.

Use the filter name box at the top to name your filter. The list of words that are found in the default filter are added to the new filter by default and can be seen here. If you wouldn't like to use any of these words and start completely from scratch, click the clear filter button. You can also remove individual words from the filter by clicking the delete icon next to the word. To add a new word to the filter, enter it into the 'Add New Filter Word' text box and either click the '+' button or press enter, this will then be added to the list of words below. Once you are happy with the word list, click 'Save Filter' to save this filter and be returned to the previous screen. 

Your new filter should be displayed in this list. Clicking apply will apply the filter to the word cloud which will automatically regenerate itself. A * will appear next to the filter name to indicate that it is currently applied to the word cloud. You can use the apply default button to revert to the default filter at any time. Clicking the edit button will allow you to make changes to your filter, and the delete button will remove it from the project completely. Below is our previous word cloud but with the new filter applied which removes the four words.

Clicking the button in the top right of the word cloud will expand it to 225% size, re-arranging the rest of the dashboard as necessary. You can click the button again to return it to it's regular size. Hovering over a word will tell you how many times this word appears in the session. Clicking on a word will apply a filter to the voice transcriptions or chat log below, showing only the entries that contain that word.

Overall Sentiment Chart

The overall sentiment chart shows how much of what was said in the session is deemed to be positive, negative or neutral by a sentiment analysis AI algorithm.

The Y-axis indicates what percentage of sentences from the session were assigned that sentiment. You can hover over one of the bars to get a precise value for this. Clicking on a bar will filter the voice transcriptions or chat logs below by the appropriate sentiment. 

Recognized Entities

The recognized entities section of the dashboard displays a list of named entities that were recognized in the session, divided into up to five categories - Locations, People, Organizations, Products and Skills. You can expand and hide each category by clicking on the arrow next to the category name. 

Locations, Organizations and People will have a button to the right of the entity name which takes you to a web page with more information about this entity. This will most likely just be a Wikipedia article. Clicking on an entity name will apply a filter to the voice transcriptions or chat logs below to show only those sentences containing that entity name. 

Sentiment Over Time Chart

The sentiment over time chart aims to give an idea of which parts of the session were deemed to be more positive and negative. 

The X-axis simply shows a timeline for the session in minutes. The Y-axis indicates what percentage of sentences in the five minutes surrounding that time were deemed to be positive or negative. The green line/area displays positive sentiment, and the red line/area displays negative sentiment. As an example, the data points at minute 35 would suggest that between the 32.5 and 37.5 minute mark, about 30% of things that were said were positive, and about 2% were negative. You can generally interpret this graph by looking for areas that stand out as being overwhelmingly one way or the other. For example there's a large green area between about minute 30 and 36, which would indicate that this period was deemed to be a lot more positive than it was negative. You can click and drag over a section of the graph to filter the voice transcriptions or chat logs by that time period.

Voice Transcriptions / Chat Logs

A table containing all voice transcriptions (for AV) or all chat logs (for text) from the session is displayed at the bottom. This is filtered by interacting with the above charts. You can remove these filters and return the table to it's default state by clicking the 'Reset Filters' button. There are some controls at the bottom of the table to switch page, or change how many sentences are displayed on each page. 

For AV sessions, the voice transcriptions that were said at a time when the recording was active will have a video icon at the end. Clicking this icon will open an inline video player displaying the recording at approximately the time of the voice transcription. This allows you to take a proper look at exactly what was going in the session when the sentence was said.