Naandhu/bert-resume-classifier

Model Card: Resume Classification Using BERT

Model Overview

This model is a fine-tuned version of `bert-base-uncased` designed for multiclass classification. It categorizes resumes into one of 24 predefined job categories, making it suitable for automated resume screening and classification tasks.

Dataset

The dataset used for fine-tuning consists of 2400+ resumes in string and PDF formats. These resumes are categorized into 24 job categories. The dataset is available at https://www.kaggle.com/competitions/jarvis-calling-hiring-contest/data

• —Classes: ['ACCOUNTANT', 'ADVOCATE', 'AGRICULTURE', 'APPAREL', 'ARTS', 'AUTOMOBILE', 'AVIATION', 'BANKING', 'BPO', 'BUSINESS-DEVELOPMENT', 'CHEF', 'CONSTRUCTION', 'CONSULTANT', 'DESIGNER', 'DIGITAL-MEDIA', 'ENGINEERING', 'FINANCE', 'FITNESS', 'HEALTHCARE', 'HR', 'INFORMATION-TECHNOLOGY', 'PUBLIC-RELATIONS', 'SALES', 'TEACHER']

The dataset underwent significant preprocessing to remove noise and improve text quality for tokenization. Preprocessing steps include:

• —Removal of HTML tags, URLs, punctuation, unicode characters, escape sequences, stop words, and irrelevant white spaces.
• —All the functions available in preprocessing.py

Model Configuration

• —Base Model: bert-base-uncased
• —Fine-tuning Task: Multiclass classification (24 classes)
• —Preprocessing Summary: The preprocessing steps applied to the training data have been encapsulated in the preprocess_function to simplify and standardize usage.

• —Model Output: The raw output consists of logits for each class. To obtain probabilities, you can apply the sigmoid activation function using torch.nn.Sigmoid().

• —Postprocessing: A postprocessing utility, included as the postprocess_function, converts the raw logits into the corresponding classified class names in text format for easier interpretation.

Training Details

The fine-tuning process involved:

• —Input tokenization using bert-base-uncased tokenizer.
• —Feeding preprocessed text into the BERT model for contextual understanding.
• —Output logits normalized using the sigmoid activation function to produce probabilities for each class.
• —The entire training code is available in kaggle: https://www.kaggle.com/code/naandhu/bert-base-uncased-fine-tuned-for-classification

Model Output

The model provides raw output logits for each job category. These logits can be converted into probabilities using:

import torch.nn as nn

sigmoid = nn.Sigmoid()
probs = sigmoid(logits)

The highest probability corresponds to the predicted job category.

Use Cases

• —Automated resume classification for HR platforms.
• —Sorting resumes into industry-specific categories for targeted hiring processes.
• —Candidate profiling and analysis for recruitment agencies.

Limitations

• —Model performance is reliant on the quality and diversity of the dataset. Biases in the dataset may affect predictions.
• —Preprocessing removes non-textual elements, which might strip out context-critical features.
• —PDFs with poor formatting or heavy graphical content may not preprocess effectively.

Citation

If you use this model in your work, please cite: "Resume Classification Model using BERT for Multiclass Job Categorization."