Awesome

Material Parsers (and other tools)

Previously this project was released as grobid-superconductors-tools, born as aister project of grobid-superconductors containing a web service that interfaces with the python libraries (e.g. Spacy).

The service provides the following functionalities:

• Convert material name to formula (e.g. Lead -> Pb, Hydrogen -> H): /convert/name/formula
• Decompose formula into structured dict of elements (e.g. La x Fe 1-x O7-> {La: x, Fe: 1-x, O: 7}): /convert/formula/composition
• Classify material in classes (from the superconductors domain) using a rule-base table (e.g. "La Cu Fe" -> Cuprates): /classify/formula
• Tc's classification (Tc, not-Tc): /classify/tc for information please open an issue
• Relation extraction given a sentence and two entities: /process/link for information please open an issue
• Material processing using Deep Learning models and rule-based processing /process/material

Usage

The service is deployed on huggingface spaces, and can be used right away. For installing the service in your own environment see below.

Convert material name to formula

curl --location 'https://lfoppiano-grobid-superconductors-tools.hf.space/convert/name/formula' \
--form 'input="Hydrogen"'

output:

{"composition": {"H": "1"}, "name": "Hydrogen", "formula": "H"}

Decompose formula in a structured dict of elements

Example:

curl --location 'https://lfoppiano-grobid-superconductors-tools.hf.space/convert/formula/composition' \

--form 'input="CaBr2-x"'

output:

{"composition": {"Ca": "1", "Br": "2-x"}}

Classify materials in classes

Example:

curl --location 'https://lfoppiano-grobid-superconductors-tools.hf.space/classify/formula' \
--form 'input="(Mo 0.96 Zr 0.04 ) 0.85 B x "'

output:

['Alloys']

Process material

This process includes a combination of everything listed above, after passing the material sequence through a DL model

Example:

curl --location 'https://lfoppiano-material-parsers.hf.space/process/material' \
--form 'text="(Mo 0.96 Zr 0.04 ) 0.85 B x "'

output:

[
    {
        "formula": {
            "rawValue": "(Mo 0.96 Zr 0.04 ) 0.85 B x"
        },
        "resolvedFormulas": [
            {
                "rawValue": "(Mo 0.96 Zr 0.04 ) 0.85 B x",
                "formulaComposition": {
                    "Mo": "0.816",
                    "Zr": "0.034",
                    "B": "x"
                }
            }
        ]
    }
]

Evaluation

The model uses DeLFT's model BidLSTM_CRF.

Evaluated on the 23/12/25

                  precision    recall  f1-score   support

        <doping>     0.6926    0.6377    0.6640       265
   <fabrication>     0.3333    0.0909    0.1429        44
       <formula>     0.8348    0.8459    0.8403      2569
          <name>     0.7346    0.7935    0.7629       949
         <shape>     0.9089    0.9608    0.9341       841
     <substrate>     0.5875    0.3176    0.4123       148
         <value>     0.8844    0.8920    0.8882       463
      <variable>     0.9645    0.9710    0.9677       448

all (micro avg.)     0.8321    0.8385    0.8353      5727

Installing in your environment

docker run -it lfoppiano/grobid-superconductors-tools:2.1

References

If you use our work, and write about it, please cite our paper:

@article{doi:10.1080/27660400.2022.2153633,
    author = {Luca Foppiano and Pedro Baptista Castro and Pedro Ortiz Suarez and Kensei Terashima and Yoshihiko Takano and Masashi Ishii},
    title = {Automatic extraction of materials and properties from superconductors scientific literature},
    journal = {Science and Technology of Advanced Materials: Methods},
    volume = {3},
    number = {1},
    pages = {2153633},
    year = {2023},
    publisher = {Taylor & Francis},
    doi = {10.1080/27660400.2022.2153633},
    URL = {
    https://doi.org/10.1080/27660400.2022.2153633
    },
    eprint = {
    https://doi.org/10.1080/27660400.2022.2153633
    }
}

Overview of the repository

• Converters TSV to/from Grobid XML files conversion
• Linking module: A rule based python algorithm to link entities
• Commons libraries: contains common code shared between the various component. The Grobid client was borrowed from here, the tokenizer from there.

Developer's notes

Set up on Apple M1

conda install -c apple tensorflow-deps

pip install -r requirements.macos.txt 

conda install scikit-learn=1.0.1

We need to remove tensorflow, h5py, scikit-learn from the delft dependencies in setup.py

pip install -e ../../delft 

pip install -r requirements.txt 

Finally, don't forget to install the spacy model

python -m spacy download en_core_web_sm

Release

bump-my-version bump patch|minor|major