User interface for content moderation of voice chat
A content moderation system analyzes speech, or characteristics thereof, and determines a toxicity score representing the likelihood that a given clip of speech is toxic. A user interface displays a timeline with various instances of toxicity by one or more users for a give session. The user interface is optimized for moderation interaction, and shows how the conversation containing toxicity evolves over the time domain of a conversation.
1. A computer-implemented method for displaying toxicity information within a timeline window of a graphical user interface, the method comprising:
calculating a toxicity score for a plurality of speech segments over the course of an audio chat session;
displaying a detailed session timeline showing a plurality of users in at least a portion of an audio chat session in a user interface, the detailed session timeline including a time axis having one or more toxicity indicators that represent a severity of toxicity and correspond to a given user.
2. The method of claim 1 , wherein each of the plurality of users has a time axis simultaneously displayed in the timeline.
3. The method of claim 2 , wherein each of the independent time axes is a horizontal axis.
4. The method of claim 2 , wherein each of the independent time axes displays a speech indicator that represents a time when the corresponding user is speaking.
5. The method of claim 2 , wherein each of the independent time axes has a session start time indicator and a session end time indicator for the user.
6. The method of claim 2 , wherein each of the independent time axes has a user indicator.
7. The method of claim 1 , wherein the toxicity indicator includes a toxicity score for the speech segment.
8. The method of claim 1 , wherein the toxicity indicators provide a heatmap for toxicity scores.
9. The method of claim 1 , wherein a moderator may select a user view for a particular user.
10. The method of claim 1 , wherein the length of the one or more vertical toxicity indicators is a function of a toxicity score.
11. The method of claim 1 , wherein selecting a toxicity indicator displays a transcript and audio file for the associated speech segment within the user interface.
12. The method of claim 1 , further comprising display an entire session timeline window showing the entire audio chat session.
13. The method of claim 12 , further comprising selecting a portion for the detailed session timeline window from the entire session timeline window; and
displaying the selected portion in the detailed session timeline window.
14. The method of claim 13 , further comprising selecting a different portion for the detailed session timeline window from the entire session timeline window; and
displaying the different portion in the detailed session timeline window.
15. The method of claim 1 , further comprising displaying a session details window, the session details window including a speaker identification, a number of speaker offenses for the speaker, a max toxicity score for the speaker, and a classification of offenses for the speaker during the session.
16. The method of claim 15 , wherein a moderator selection of the speaker identification causes the graphical user interface to display a user view with a window with showing all of the selected user's activity.
17. The method of claim 16 , wherein the user view displays the length of each session for the user, the maximum toxicity for the user for each session, and the offense category for each session.
18. The method of claim 1 , further comprising selected a portion of the detailed timeline view.
19. The method of claim 1 , further comprising displaying toxicity indicators for toxicity that meets a toxicity score threshold.
20. The method of claim 1 , further comprising receiving an input in the detailed session timeline window, the input corresponding to an interval of time; and
displaying a transcription of toxic speech that meets a toxicity score threshold for the corresponding interval of time.
21. The method of claim 20 , further comprising displaying a link to audio for the toxic speech for the corresponding interval of time.
22. The method of claim 1 , further comprising displaying toxicity indicators for a plurality of users.
23. The method of claim 1 , wherein the horizontal time axis has a thickness along its length that is a function of an amount of speech.
24. The method of claim 1 , further comprising display a moderator action window.
25. The method of claim 24 , wherein the moderator action window includes options for muting, suspending, or banning the speaker.
26. A multi-stage content analysis system comprising:
a first stage trained using a database having training data with positive and/or negative examples of training content for the first stage,
the first stage configured to:
receive speech content,
analyze the speech content to categorize the speech content as having first-stage positive speech content and/or first-stage negative speech content;
a subsequent stage configured to receive at least a portion, but less than all, of the first-stage negative speech content,
the subsequent stage further configured to analyze the first-stage positive speech content to categorize the first-stage positive speech content as having subsequent-stage positive speech content and/or subsequent-stage negative speech content, the subsequent stage further configured to update the database using the subsequent-stage positive speech content and/or the subsequent-stage negative speech content,
the subsequent stage including a toxicity scorer configured to provide a toxicity score for subsequent-stage positive speech content;
a user interface configured to display the toxicity score for the subsequent-stage positive speech content.
27. The multi-stage content analysis system of claim 26 , wherein the user interface is configured to display the toxicity score for multiple users within a timeline.
28. The multi-stage content analysis system of claim 27 , wherein each of the toxicity scores for the multiple users includes a toxicity indicator and an associated user indicator within the timeline.