NLP is computer understanding, analysis, manipulation and/or generation of natural (human) language. We use NLP to analyze user-generated content within specific semantic categories.
What is semantics?
Formally, semantics is the linguistic study of the intended structure and meaning of words. When we train JuLiA, we focus our training on particular problems, such as identifying abusive blog comments. In this case, "abusiveness" is the meaning we want to identify.
What is machine learning?
Machine learning refers to the ability of a program, in this case JuLiA, to learn from experience, that is, to modify and improve its performance on the basis of newly acquired information.
What is sentiment analysis?
This refers to the categorization of content by its emotional intent or tonality. We conduct sentiment analysis within particular subcategories of abusiveness. For example, JuLiA might find discriminatory or inflammatory sentiment in an incoming comment and tag it accordingly.
How do you customize JuLiA?
We train JuLiA according to any new client's editorial standards using historical data (such as published or deleted comments) from their community. We also have predefined models of general abusiveness that can be modified to suit a client's needs.
How long does a typical customization take?
It typically takes about six weeks to build a new model from scratch. Modifying an existing model takes significantly less time.
Can I use JuLiA out of the box?
Yes. Our generalized version can be easily adapted to any client's needs. However, customization is recommended.
How much does JuLiA cost?
We have a tiered pricing model that deeply undercuts the cost of human moderation. Please contact us for more information.
What do you mean by "JuLiA learns over time"?
Every model is updated on a monthly basis using feedback from the client's own community. In this way the model continuously learns new language and vocabulary, and refines its own semantic "understanding".
How accurate is JuLiA?
Our models have consistently proven to be more accurate, more consistent, and less biased than human moderators.
How are you evaluating accuracy?
For each model we generate standard metrics of accuracy, precision, and recall against a test set of live comments. We then compare these against standard measures of human performance.
How fast does JuLiA moderate content?
JuLiA moderates content in real time and tends to eliminate backlogged and pending content on even the busiest sites. Our current benchmark is 2 million comments a month and rising.
Can JuLiA do more than moderate individual comments?
In addition to comment moderation, we also provide a full suite of community reporting tools that allow you to identify abusive members, track feedback on published articles, identify experts within your community, and so much more.
