Cognitive comment analysis for preventing inappropiate online appearance/image.
Publication Date: 2016-Jul-07
The IP.com Prior Art Database
Disclosed is a method for cognitive comment analysis in order to prevent inappropriate online appearance/image
Page 01 of 3
Cognitive comment analysis for preventing inappropiate online appearance /
Disclosed is a method for cognitive comment analysis in order to prevent inappropriate online appearance/image.
Users often need to leave their comments on auction sites or just need to reply/write messages on forums/social media. Our solution is to prevent users from posting content not suitable because not fitting to their image - or personality they want to be perceived.
Comments/messages are passed to set of classifiers - each analyses certain tone and we can calculate if given message fits to defined "personality". if not it helps user to moderate the message to fit such defined "personality". You can imagine it for example as system helping support people appear "cold and professional" in their contacts with customers.
System is positioned between business logic and UI but may as well be part of business logic. We can classify comment with group of classifiers later referred to also as tone analysers. We assume that each of tones is returned in normalised range and is linked with text fragment that mostly influenced the tone value. Having this information we can check if comment fits acceptable user "personality profile". We do this by comparing a vector of "tones" to what user has defined (or what is pre-defined). When the returned value for given tone is outside the range defined we check hint given by classifier and based on it we try to modify the text to fit given range. If we are unable to supply corrected text automatically we indicate to user the text that influenced the decision most. This flow is shown on fig 1.
Page 02 of 3
set of coefficients showing how user wants to be seen in various aspects - like how much various emotions can be visible in his/her comments. In addition each tone or coefficient will be connected with weight that will describe how much user cares about given tone (see embodiment below).
Let's imagine we have the following profiles profiles:
- tough guy
- cold professional
Each would have list of emotions/language/social tones and their weight and a flag indicating that this particular tone should not be taken under considerations. Example ton...