Lipika Dey

Short CV: Dr. Lipika Dey has made key contributions in the areas of language technologies and text and data mining for deriving operational and strategic insights from business communications to aid business decision making. One of her pioneering contributions has been towards conceptualization and development of an architecture for “real time contextually aware enterprises”. This platform can amass and analyze large volumes of heterogeneous information including natural language text, in near real time using novel algorithms, to deliver enterprise level solutions for problems like customer satisfaction management, improving Nett Promoter Scores for brands, automation of support center operations, detecting early signals for future risks and strategic planning for research management. Aided by her deep expertise in ontology-based reasoning with text data Lipika and her team also provided, for the first time, an effective solution for extracting high quality insights from consumer generated noisy texts that can be directly mapped to a business ontology.

Under her active mentorship, variants of this platform, with few patented and few patent-pending, technologies, have been integrated into multiple deliverables built by TATA CONSULTANCY SERVICES (TCS) for some of its most-valued customers in the verticals of Banking and Financial services, Manufacturing and Telecom sectors. A key differentiator of these deliverables has been the imbibed methodologies designed by her that help in establishing a direct linkage between the analytical outcomes and measurable business KPIs. This enabled faster adoption of language technologies by industrial clients. It has led to significant reduction in terms of manpower and operational costs as well as increase in Customer Satisfaction Index (CSI) for all customers. Lipika has been instrumental in developing innovative and niche applications based on deep learning architectures for solving some of the critical natural language processing problems of the industry today - like adverse event detection, role-disambiguation of actors mentioned in text, causality detection and regulatory violation detection from streaming text content. These technologies have been used in high-value Customer verification and Sustainability Assessment for financial sectors as well as for risk assessment and sustainability scoring of organizations based on their GRI reporting as well as real-time tracking of public mentions for Environment - Social - Governance (ESG) related activities.

Title of Talk 1: Language Technologies for Healthcare Industry

Synopsis: Though most of the communication within the Healthcare industry is predominantly textual in nature, exploitation of language technologies to extract insights from the text content is very minimal. However, the content can yield a goldmine of insights related to prognostic factors for frequent and rare diseases, about adverse drug effects, reveal patient cohorts which may provide additional predictive insights about hospital admission requirements and so on. Given the unique nature of clinical and bio-medical languages, special purpose language models have been built for analysis of these texts. A wide range of predictive models have been subsequently built for insight extraction and their use in decision making systems. This talk will cover some of these works, done across the globe and also by the speaker with her teams.

Title of Talk 2: Augmenting Decision Making with Causal Explanations

Synopsis: Despite the success of Large Language Models in generating content, they are known to often fail in generating causal explanations. Text documents however are rich repositories of causal knowledge. While journal publications typically contain analytical explanations of observations on the basis of scientific experiments conducted by researchers, analyst reports, News articles or even consumer generated text contain not only viewpoints of authors, but often contain causal explanations for those viewpoints. As interest in data science shifts towards understanding causality rather than mere correlations, there is also a surging interest in extracting causal constructs from text to provide augmented insights for better decision making. Causality extraction from text is usually viewed as a relation extraction problem which requires identification of causal sentences as well as detection of cause and effect clauses separately. Sequence modelling using transformer based architectures have proved to be very successful in in-context learning and recovering a wide variety of causal structures.

Title of Talk 3: NLP applications for Sustainability Analysis

Synopsis: Along with implementing sustainable practices within an organization, tracking sustainability practices of vendors, partners, clients and other stakeholders is also an important aspect of business management. Keeping abreast of sustainability related research and technology literature is also an important aspect of knowledge management within an organization. Given the vastness and complex nature of this content, it is impossible to track and assimilate all the content manually. Deployment of advanced language processing technologies dedicated to analysis of sustainability related content including simultaneous analysis of structured and unstructured data is an emerging area of research. It not only includes extraction of insights but also reasoning with heterogeneous data, detecting contradictions, fallacies etc. This talk will focus on discussing the challenges and advances in the area of sustainability content analysis.