When to Train
This topic describes part of the functionality of Genesys Content Analyzer.
When does your training object have enough categorized text objects to make training worthwhile?
Here are some possible situations and comments on them.
Uniformly Low Feedback
In this situation, all categories have a small amount of feedback (less than about 12 text objects per category). This object is not fully ready for training. You can still try training a model, but you should be aware that the results probably will not be very good.
Unbalanced Feedback: Mostly Low
In this situation, all categories except a small group have a small amount of feedback (less than about 12 text objects per category). The small group (one to five categories) may have several hundred or even thousands of feedback objects per category. You can train a model, but the resulting model will mostly return the categories from the small group. A situation of this type may have these causes:
It may be an accurate reflection of the situation. For example, your company may sell 25 products but just three of them may account for 90 per cent of its business.
It may reflect shortcomings in the system:
Agents may not use standard responses properly.
- The standard responses and/or the category tree may be poorly designed.
- Agents may not use standard responses properly.
To determine which of these causes obtains, inspect your category tree, standard responses, and agents’ use of them. If the situation arises because of shortcomings in the system, consider doing the following:
Bring some balance into the training object by deleting some of the text objects associated with categories of the high-feedback group.
Modify the low-feedback categories.
Unbalanced Feedback: Mostly High
In this situation, some categories have a small amount of feedback (less than about 12 text objects per category), but a significant number (over 50) of categories have a large amount of feedback (over 30–50 text objects or more per category). This is a rather common situation. You can train a model and it will work acceptably on the high-feedback categories. But consider modifying the low-feedback categories.
Uniformly High Feedback
In this situation, almost all categories have significant feedback (over 50 text objects per category). This is the best situation. It means that agents are frequently using almost all standard responses. You can train the model and it should perform well on all categories.