Analysis using NER, Quintuple Extraction, Comparative Opinion Analysis, and Coreference Resolution
BITS MTECH DISSERTATION PROJECT
This system is specifically trained on annotated Air India flight reviews using vector search with cosine similarity for context retrieval. While it can analyze any feedback text, it performs best with airline and flight-related content.
Domain Adaptability: The same methodology can be extended to any industry by replacing the training data.
Traditional feedback analysis systems lack comprehensive understanding of customer sentiments, entity relationships, and comparative opinions. This system addresses the need for multi-dimensional analysis including Named Entity Recognition (NER), sentiment quintuple extraction, coreference resolution, and comparative opinion mining to provide deeper insights into customer feedback.
System architecture showing the complete data flow from input processing through NLP agents to final outputs.
Identification and classification of entities (PER, ORG, GPE, DATE, etc.)
Target Objects, Features, Sentiments, Opinion Holders, and Extracted Times
Resolving pronouns and references to their actual entities
Extracting comparative opinions between different entities
Dataset: Air India Flight Reviews from Kaggle (Annotated)
Source: Air India Reviews EDA - Kaggle Dataset open_in_new
Methodology: The system uses annotated Air India reviews stored as vector embeddings. When new input is received, cosine similarity search retrieves the most relevant annotated examples, which are then provided as context to the LLM for generating accurate NLP analysis results.
This domain-specific approach enables the system to understand airline industry terminology, service aspects, and customer sentiment patterns. The same methodology can be extended to any domain by replacing the training data - simply annotate domain-specific feedback (e.g., restaurant reviews, product feedback, healthcare comments) and the system will adapt accordingly.