A topic enables you to boost the recognition of a specific business issue by telling SpeechMiner UI what phrases to look for. For example, if you want to identify interactions in which the customer wishes to cancel an account, you could create a topic named Cancellation with the following phrases: "Close out my account" and "I need to cancel".
Topics can also be used to improve the overall speech recognition of the system in two main ways:
- Directly: For example, the word “cinnamon” may appear frequently in calls when “cinema” was actually said. You could enter a number of phrases in the topic related to “cinema” in order to improve the transcription results. This assumes that you do not need to report on calls related to “cinema”. That is, because the topic is a collection of mostly unrelated phrases, it is not very useful for reporting. In order to run reports on “cinema”, it is recommended to place the phrases in a topic that has a common business meaning.
- Indirectly: Consider the scenario in which “account number” is a phrase that is important to capture, but its accuracy is low due to a number of false positives being generated because “phone number” is recognized as “account number”. Adding the phrase “phone number” to the topic could potentially draw those false positives away from “account number”. Placing “phone number” in a topic implies that it is not important to run reports on “phone number”. For the incorrectly-recognized and undesired “cinnamon” appearing in transcripts, it is recommended to try to remove it (and therefore correctly recognize “cinema”) by adding “cinnamon” into the topic with strictness set at 100. Various phrases with additional context (for example, “late night cinnamon”) may work better than using the single-word phrase “cinnamon”.
More specifically, topics are a collection of phrases defined globally for the SpeechMiner UI system, and included in one or more programs as required. When a topic is included in a program, SpeechMiner UI searches for all of the phrases included in the topic's definition, in all the interactions associated with the program. When one of the phrases is found, SpeechMiner UI identifies it as a linguistic event and registers the topic as found at a specific time during the interaction.
Each linguistic event identified by SpeechMiner UI in an interaction has a start time, an end time, a name, and a type. For example, if a topic called "Loan Offer" is identified by the phrase "can offer you a loan," the start time is when the agent began saying "can," the end time is when the agent finished saying "loan," the name is "Loan Offer," and the type of event is "Topic."
Although a topic can contain one phrase, the more phrases you add the more useful the topic.
Once a topic is created, you can always make modifications based on the accuracy of the results that the topic is achieving.