Client, an FMCG major, partners with market research agencies to conduct and analyse focus group discussions to better understand consumer behaviour and expectations. The outcome of these studies is often a summary report which captures the major themes/observations as obtained from the FGDs but doesn’t provide a mechanism to correlate insights with the raw conversations of the FGD participants. To achieve the same, the client either needs to peruse the raw transcript of the FGD or generate a coherent summary from the transcript, both of which are manual, time-consuming and expensive. Consequently, the client partnered with Inxite Out to create a solution that could enable them to correlate the insights from research agencies with the raw information extracted from FGDs autonomously and inexpensively.
INXITE OUT APPROACH
An end-to-end solution was developed leveraging MEGHNAD, Inxite Out’s Conversation Intelligence Framework. The solution consisted of the following steps:
FGD Summarization: End-to-end process using Meghnad
MEGHNAD’s transcription engine was used to generate transcribed text files from multi-lingual FGD discussions with a typical duration of 1 – 2 hours. The transcribed text files generated had the following elements making them amenable for downstream processing:
- Speaker-Segmentation: Clearly demarcates the utterances of the moderator of the FGD from the participants.
- Captures granular time stamps with the utterances to enable lineage tracking.
- Transcribed text files are all in English, even though the audio files had a mixture of English and Hindi.
Chunking & Summarization
Leveraged MEGHNAD’s Summarization engine to generate a topical and coherent summary of the FGD. It was achieved in the following steps:
- Chunking of transcribed text files based on temporal information and topic content.
- LangChain-based map-reduce prompt execution strategy on those chunked texts stored in a vector database post recursive splitting to align with token lengths allowed by large language models used underneath.
- Summarization of each independent chunk and intelligent aggregation to ensure a coherent summary across chunks.
Incorporation of Lineage Tracking
Facilitated lineage tracking to enable deep dives from the summary to the raw transcript for granular conversation understanding.
- Focus group discussion summarization instituted at the client to facilitate cross-referencing of market research findings with raw customer conversations.
- 100% FGD summarization coverage as compared to 20% coverage with legacy methods.
- 75%+ reduction in FGD summarization cycle time as compared to legacy methods