Comparison of TF-IDF and Sentence-Transformer NLP Methods for a Perfume Recommendation System Based on User Descriptions with Streamlit Visualization

Leveraging Language Models for Textual Retrieval in the TREC 2025 DRAGUN Track

Large Language Models for Token-Efficient and Semantic-Preserving Opinion Summarization

Opinionated text - spanning product reviews, hotel feedback, and social posts - captures rich signals about user experiences, preferences, and concerns. However, the scale, redundancy, and imbalance of such corpora make it challenging to analyze opinions effectively, particularly when the goal is to generate summaries that remain faithful to the diversity of viewpoints expressed. This paper presents a framework that preserves semantics in LLM-based opinion summarization while minimizing token usage. We combine multidimensional classification (e.g., sentiment, topics) with a family of stratified sampling strategies to select compact yet representative subsets of opinions before prompting the LLM. Tailored prompts then produce balanced summaries that surface the salient aspects expressed in the opinions (e.g., strengths and weaknesses of products/hotels). Experiments on Amazon product reviews, Tripadvisor hotel reviews, and X/Twitter posts demonstrate that our method significantly reduces token usage and computational cost while consistently outperforming traditional AI-based and standard LLM summarization baselines in terms of content coverage, balance, and semantic preservation.

Performance Evaluation of Embedding-Based and Keyword-Based Retrieval in Text Description-Based Hotel Recommendation

The massive volume of textual descriptions on hotel booking platforms makes it difficult for recommendation systems to accurately match user preferences. Traditional keyword-based retrieval methods, such as TF-IDF, often struggle to capture semantic relationships when relevant terms do not explicitly overlap. This study evaluates the performance of keyword-based (TF-IDF) and embedding-based (paraphrase-multilingual-MiniLM-L12-v2) retrieval approaches in a content-based hotel recommendation system using a small-scale dataset. The dataset consists of 30 unique Traveloka hotels in Yogyakarta collected from Kaggle, representing a resource-constrained experimental setting. Evaluation was conducted using a black-box approach with 10 dynamic synthetic queries and assessed through Precision@3 (P@3) and Mean Average Precision (MAP). The results indicate that MiniLM achieved higher retrieval effectiveness than TF-IDF, with a mean P@3 of 0.3667 and a mean MAP of 0.1378, compared with 0.3000 and 0.1333, respectively. These findings suggest that embedding-based retrieval is more effective in capturing semantic information, including synonym usage and implicit contextual relationships, within the evaluated dataset. Therefore, compact embedding models such as MiniLM may provide an alternative approach to traditional keyword-based retrieval methods for small-scale recommendation systems.

Optimizing sample selection for large language model-based entity matching using AssistEM

AssistEM, a framework for efficient LLM adaptation to EM via principled data selection, demonstrates that selective fine-tuning not only accelerates adaptation but also improves training efficiency (requiring fewer GPU hours), enabling open-source LLMs to rival–and in some cases outperform–closed-source models.

Advanced Sentiment Analysis of IMDB Movie Reviews Using NLP Preprocessing, TF-IDF and Classical Machine Learning Models

Sentiment analysis (or opinion) is a core problem in Natural Language Processing (NLP) that aims to automatically determine the polarity of opinions expressed in text. As online review platforms continue to expand, the volume of user generated content has grown beyond what can be examined manually. Movie reviews are particularly challenging because they are subjective, vary widely in length, and use diverse writing styles. This paper presents a sentiment classification pipeline for IMDb movie reviews that combine rigorous text cleaning with Term Frequency Inverse Document Frequency (TF–IDF) feature extraction and classical supervised learning models. Exploratory analysis is used to examine sentiment balance and review-length characteristics. The pre-processing stage removes HTML/URLs, normalizes text, filters stop words, and applies lemmatization to reduce noise and vocabulary sparsity. TF–IDF vectors with unigram and bigram features are then constructed to capture both individual sentiment cues and short contextual phrases. Logistic Regression, Linear SVM, Multinomial Naïve Bayes, and Random Forest are evaluated under a consistent experimental setup. Performance is reported using accuracy, precision, recall, F1-score, and ROC-based analysis. The results show that Logistic Regression achieves the best overall performance, with an accuracy of 88.68% and an F1-score of 88.62%, followed closely by Linear SVM. Overall, TF–IDF with linear classifiers remains an interpretable and computationally efficient baseline for practical sentiment classification.

An Intelligent PDF Question-Answering System; A Retrieval-Augmented Generation Approach

Due to the surge in digital document creation, intelligent systems that can effectively retrieve accurate information from unstructured PDF content have become a necessity. Current keyword-based search techniques lack semantic meaning, while large language models (LLMs), when utilized individually, produce hallucinated results when they are not contextually aware of document content. In this paper, we introduce a novel real-time question answering system for PDF content using a lightweight approach to Retrieval-Augmented Generation (RAG). Our system combines a HuggingFace sentence transformer model for semantic embeddings with a FAISS vector similarity search for efficient retrieval, leveraging a Groq-hosted LLaMA 3.3 70B model for context-grounded answer generation. The system is built using LangChain for orchestration and Streamlit for web-based deployment, and requires no GPU, no fine-tuning, and no pre-indexed corpus. Evaluation across five complementary metrics—Exact Match, Token-level F1, BLEU, ROUGE, and semantic similarity—demonstrates strong alignment between generated and reference answers, with a peak semantic similarity of 0.8709 and a mean Token F1 of 0.4701, confirming the viability of this approach for real-world document intelligence.