Subscribe and receive the latest news from the industry.
Join 56,000+ members. Yes it's completely free.
Powering topic mining with GenAI
Telecoms.com periodically invites expert third parties to share their views on the industry’s most pressing issues. In this piece Harsha Angeri, VP, Corporate Strategy & Head, AI Business at Subex, talks us through the concept of topic mining.
Guest author
December 9, 2024
5 Min Read
Historically a rich source of unstructured data, including call centre transcripts, customer surveys, social media posts, network performance logs, and app store feedback, the telecom industry is witnessing a shift in how customers engage with text-based interactions. Using a variety of channels for service and support like chatbots, WhatsApp, and SMS, the tsunami of unstructured data from a multitude of sources has made traditional analytics obsolete. While the data holds critical insights into customer behaviour, service quality, and market trends; the sheer volume and complexity of the data makes conventional analysis methods inefficient. That is, until now…
Powered by generative artificial intelligence (GenAI), topic mining has become a critical tool for telecom operators. It automates the analysis of vast datasets by scanning call transcripts and categorising them into relevant themes, incident categories, and sub-categories. Topic mining allows operators to extract actionable insights from diverse data sources, alleviating the challenge of analysing data across multiple sources and enabling them to extract meaningful insights. These insights enable operators to uncover customer preferences, solve recurring issues before they escalate, make strategic decisions regarding product and services development, reduce costs, and improve the decision-making process.
Topic mining: The process
The journey of topic mining begins with the collection of customer feedback across a variety of channels. These channels provide a diverse and comprehensive dataset, capturing customer sentiments, complaints, and interactions from multiple touchpoints. The data that has been collected feeds into an advanced topic mining system, ensuring all relevant information is gathered to form a comprehensive view of the customer experience.
Once the data is collected, it undergoes four processing stages.
Text chunking and preparation: It is during this step that the raw feedback is broken down into manageable pieces and organised for deeper analysis. The text is cleansed and tokenised, ensuring the data is structured and ready for advanced processing.
Named entity recognition (NER): The NER identifies and classifies specific entities such as products, locations, or people mentioned in the feedback, enabling operators to perform more targeted analysis and expedite issue resolution.
Topic modelling: The system uses vector embedding, which converts text into numerical vectors in a high-dimensional space, capturing semantic relationships within the feedback. This is then followed by Large Language Model (LLM) interpretation, where LLMs extract context-aware topics, trends, and sentiments, enabling the system to identify both known and emerging issues.
Decision insights generation: This involves generating actionable insights from the processed data, enabling operators to make informed decisions regarding enhancing the customer experience and addressing key challenges.
The insights generated from topic mining feed into dynamic and iterative models that continuously improve customer experience and customer value management. For example, 1) segmentation models help in categorising customers into distinct groups based on behaviour and preferences; 2) by analysing customer data, next best offer models predict the most relevant products or services to offer the individual customer; 3) the models identify customers at risk of churn, enabling proactive retention strategies, and 4) anomalies in customer behaviour or network performance can indicate underlying issues, allowing operators to address potential problems before they escalate.
Unlock new opportunities
With vector embedding and LLMs, operators can now proactively uncover new product opportunities and identify emerging issues in a zero-shot manner. These advanced capabilities provide operators with actionable insights into unmet customer needs and dormant market demands. By analysing unstructured data from social media, app store reviews, or customer service transcripts, operators can:
Identify unmet needs: Insights from feedback can reveal areas where current services are lacking, suggesting features or net new offerings that will fill market gaps.
Detect latent demands: Patterns in customer behaviour and feedback may indicate demand for new services that have not yet been explicitly requested but are needed, providing operators the opportunity to develop innovative solutions.
Enhance products and services: By continuously mining feedback, operators can fine-tune existing offerings to ensure that customer expectations are met or exceeded, driving increased satisfaction and promoting loyalty.
Identifying issues or trends using traditional methods often rely on pre-labelled data. With today’s AI techniques, operators can use zero-shot learning to identify both new and unknown issues - without the need for labelled datasets. The ability to identify issues the zero-shot way provides several benefits, including:
Early detection of new problems: Using customer feedback, zero-shot learning allows the system to recognise new issues as they arise. This capability is underscored by zero-shot’s ability to detect new issues - without the need for model training.
Proactive problem solving: Operators can address emerging issues before they escalate, improving the customer experience and reducing churn.
Agility in a dynamic environment: Zero-shot learning provides the flexibility to adapt to new customer needs, regulatory requirements, and technology changes.
A powerful combination: Topic mining and RA/FM systems
A unique advantage of advanced topic mining lies in its ability to integrate with revenue assurance (RA) and fraud management (FM) systems. This combination is particularly powerful in addressing issues related to billing, recharge disputes, and fraud detection.
By merging customer feedback insights with operational and financial data, operators gain visibility into both customer experience and revenue impacts. The benefits of this integration include:
Pre-emptive billing and recharge issue detection: Customer feedback analysis from call centres, social media, and app stores can uncover patterns related to billing and recharge disputes. By correlating this feedback with data from RA/FM systems, operators can pinpoint the root causes of these issues - whether they are technical glitches, fraudulent activities, or process inefficiencies.
Fraud detection and prevention: Merging topic mining insights with RA/FM allows for the identification of fraudulent behaviour patterns emerging from customer complaints or suspicious activities.
End-to-end assurance: By combining topic mining insights with RA/FM data, operators can close the feedback loop on revenue-impacting issues, ensuring customer concerns are addressed and underlying fraud or assurance issues are proactively resolved.
Advanced topic mining is not just a tool for analysing customer feedback, it’s a critical asset to enhance customer satisfaction, cut costs, and improve operational efficiency. As advancements in AI continue, topic mining will become even more powerful, giving operators the insights needed to thrive in an ever-increasing data-driven world.
Harsha is responsible for steering the strategic direction of the company. His responsibilities encompass shaping corporate initiatives, developing and executing growth strategies, portfolio transformation and fostering innovation within our technology landscape. He runs the AI business of Subex driving the strategy & roadmaps across Conventional and Generative AI applications. Harsha comes aboard with a wealth of entrepreneurial leadership and strategic insight.
Read more about:
DiscussionAbout the Author
You May Also Like
.png?width=300&auto=webp&quality=80&disable=upscale)
.png?width=300&auto=webp&quality=80&disable=upscale)
_1.jpg?width=300&auto=webp&quality=80&disable=upscale)