Advanced NLP: NER + Extractive QA

# Description

Advanced Natural Language Processing (NLP) covering two fundamental tasks: (1) Named Entity Recognition (NER) — detecting persons, organizations, locations and miscellaneous entities using dslim/bert-base-NER (fine-tuned CoNLL2003) for English and Davlan/bert-base-multilingual-cased-ner-hrl for Spanish. The model detects entities across 5 example sentences per language with high confidence (mean >0.95). Quantitative evaluation on 10 manually annotated sentences with seqeval: Accuracy 93.39%, Micro F1 81.36%, ORG F1 94.12%, LOC F1 92.31%, PER F1 75%. (2) Extractive Question Answering — literal answer extraction from context using distilbert-base-cased-distilled-squad with raw model inference (softmax over start/end logits, transformers v5). 2 real contexts (Turing Award, Deep Learning Revolution) with 8 questions: mean score 83%. 6 visualizations: entity distribution, token confidence, per-entity metrics, QA score histogram, per-question confidence, EN/ES comparison heatmap.

$ ./deploy.sh $ git clone

# Key features

$ Multilingual NER: English (dslim/bert-base-NER) and Spanish (Davlan/multilingual-ner-hrl)

$ Quantitative seqeval evaluation: Accuracy 93.39%, F1 ORG 94.12%, LOC 92.31%, PER 75%

$ Extractive QA with DistilBERT-SQuAD: 8 questions over 2 real contexts, mean score 83%

$ Raw model inference for transformers v5: softmax over start/end logits for QA

$ 10 manually annotated IOB2 test sentences for NER evaluation

$ 6 visualizations: bar charts, histograms, heatmap EN/ES comparison

$ Pretrained models: 3 ready pipelines without GPU fine-tuning

# Gallery

Project terminal