Unsupervised paradigm for information extraction from transcripts using BERT

Audio call transcripts are one of the valuable sources of information for\nmultiple downstream use cases such as understanding the voice of the customer\nand analyzing agent performance. However, these transcripts are noisy in nature\nand in an industry setting, getting tagged ground truth data is a challenge. In\nthis paper, we present a solution implemented in the industry using BERT\nLanguage Models as part of our pipeline to extract key topics and multiple open\nintents discussed in the call. Another problem statement we looked at was the\nautomatic tagging of transcripts into predefined categories, which\ntraditionally is solved using supervised approach. To overcome the lack of\ntagged data, all our proposed approaches use unsupervised methods to solve the\noutlined problems. We evaluate the results by quantitatively comparing the\nautomatically extracted topics, intents and tagged categories with human tagged\nground truth and by qualitatively measuring the valuable concepts and intents\nthat are not present in the ground truth. We achieved near human accuracy in\nextraction of these topics and intents using our novel approach\n