Ner4Opt: Named Entity Recognition for Optimization

Given an optimization problem in natural language, Ner4Opt extracts optimization related entities from free-form text.

Check out this HugginFace Space Demo to test it yourself 🤗.

The Ner4Opt model (Constraints'24, CPAIOR'23, NeurIPS'22) is trained to detect six named entities:

• CONST_DIR: Constraint direction
• LIMIT: Limit
• OBJ_DIR: Objective direction
• OBJ_NAME: Objective name
• PARAM: Parameter
• VAR: Variable

Here are the details of our pre-trained models and the training procedure. Our pretrained models are hosted on HuggingFace Models.

Quick Start

from ner4opt import Ner4Opt

# Input optimization problem description as free-form text
problem_description = "The Notorious Desk company wants to promote a new brand of wine and wants to market it using a total market budget of $ 87,000 . To do so , the company needs to decide how much to allocate on each of its two advertising channels : ( 1 ) morning TV show and ( 2 ) social media . Each day , it costs the company $ 1,000 and $ 2000 to run advertisement spots on morning TV show and social media respectively . The expected daily reach , based on past ratings , is 15,000 viewers for each morning show spot and 30,000 internet users for a social media spot . The chief marketer knows from her experience that both channels are key to the success of the product launch . She wants to plan at least 4 but no more than 7 morning show spots . In addition , the social media spots needs to be at least 30 due to pricing tier policy . How many times should each of the media channels be used to maximize the reach of the campaign ?"

# Ner4Opt Model with options lexical, lexical_plus, semantic, hybrid (default). 
ner4opt = Ner4Opt(model="hybrid")

# Extract a list of dictionaries corresponding to entities found in the given problem description.
# Each dictionary holds keys for the following: 
# start (starting character index of the entity), end (ending character index of the entity)
# word (entity), entity_group (entity label) and score (confidence score for the entity)
entities = ner4opt.get_entities(problem_description)

# Output
print("Number of entities found: ", len(entities))

# Example entity output
[
    {
        'start': 108,
        'end': 114,
        'word': 'budget',
        'entity_group': 'CONST_DIR',
        'score': 0.9919970308651846
    },
    {
        'start': 120,
        'end': 126,
        'word': '87,000',
        'entity_group': 'LIMIT',
        'score': 0.9993724035912778
    },
    { ... },
]

Installation

Ner4Opt requires Python 3.8 and can be installed from PyPI using pip install ner4opt or by building from source

git clone https://github.com/skadio/ner4opt.git
cd ner4opt
pip install .

Testing

To run tests, execute the following from the root folder:

python -m unittest

Citation

If you use Ner4Opt, please cite the following paper:

@article{Kadioglu24,
  author       = {Serdar Kadioglu, Parag Pravin Dakle, Karthik Uppuluri, Regina Politi, Preethi Raghavan, SaiKrishna Rallabandi, Ravisutha Srinivasamurthy},
  title        = {Ner4Opt: named entity recognition for optimization modelling from natural language},
  journal      = {Constraints},
  volume       = {29},
  number       = {3-4},
  pages        = {261--299},
  year         = {2024},
  url          = {https://doi.org/10.1007/s10601-024-09376-5},
  doi          = {10.1007/S10601-024-09376-5}
}

@inproceedings{dakle23,
  title 	= {Ner4Opt: Named Entity Recognition for Optimization Modelling from Natural Language}
  author 	= {Parag Pravin Dakle, Serdar Kadıoğlu, Karthik Uppuluri, Regina Politi, Preethi Raghavan, SaiKrishna Rallabandi, Ravisutha Srinivasamurthy}
  journal 	= {The 20th International Conference on the Integration of Constraint Programming, Artificial Intelligence, and Operations Research (CPAIOR 2023)},
  year 	    = {2023}
}