IDF impact on categories
When analyzing categories, the system uses the IDF method to calculate and prioritize the category Grouping of calls in Speech Analytics that pertain to specific business issues among all transcribed interactions. results.
Each interaction In Speech Analytics, an interaction represents a single part of the contact between one employee and the same customer. In Text Analytics, an interaction is the communication session between one or more employees and the same customer with a unifying contextual element. is ranked according to its relevancy to the category, indicated by a defined number of stars. Five stars is the highest rank. The higher the rank, the more relevant that interaction is to the category. Depending on the minimum rank setting, interactions below the rank threshold, for example, a 2-star rank, are excluded from the category results.
The ranking Rating of a specific interaction in the search results in Speech Analytics, which is determined by the number of instances of the search term within the interaction, and the interaction’s individual score. of an interaction in a category, however, is not fixed. Because new interactions are continuously added to the index Repository of transcribed interactions, which are used for analysis in Speech Analytics., every hour or so, the system reevaluates categories. The addition of new interactions to the index can influence the ranking of existing interactions. Therefore, the same interaction can be ranked differently, depending on when you searched the category. In addition, the number of interactions can change slightly in the result set.
If category term One to five words that are meaningful to a specific type of business, or phrases that stand out in interactions in Speech and Text Analytics. frequencies change drastically following the indexing Process in Speech Analytics in which the Speech Application Service retrieves transcribed interactions, and builds a semantic index based on this data. of new interactions, the IDF weight changes and updates the overall rank of interactions. In this case, theoretically, an interaction that currently belongs to a specific category can no longer belong to that same category later.
However, because the IDF calculation is based on a time frame of 30 days backward from each interaction date, term frequencies are expected to be stable. A change in the frequency of a term due to new interactions would not affect the scoring of interactions older than 30 days. For the same reason, category statistics are expected to be stable as well.