---

language: en
license: mit
datasets:
- midas/inspec
tags:
- keyphrase-extraction
metric:
- f1

---
** Work in progress **
# 🔑 Keyphrase Extraction model: KBIR-inspec
Keyphrase extraction is a technique in text analysis where you extract the important keyphrases from a text. Since this is a time-consuming process, Artificial Intelligence is used to automate it.
Currently, classical machine learning methods, that use statistics and linguistics, are widely used for the extraction process. The fact that these methods have been widely used in the community has the advantage that there are many easy-to-use libraries. 
Now with the recent innovations in deep learning methods (such as recurrent neural networks and transformers, GANS, …), keyword extraction can be improved. These new methods also focus on the semantics and context of a document, which is quite an improvement.


## 📓 Model Description
KBIR pre-trained model fine-tuned on the Inspec dataset. KBIR
Keyphrase Boundary Infilling with Replacement (KBIR) which utilizes a multi-task learning setup for optimizing a combined loss of Masked Language Modeling (MLM), Keyphrase Boundary Infilling (KBI) and Keyphrase Replacement Classification (KRC).
Paper: https://arxiv.org/abs/2112.08547

## ✋ Intended uses & limitations
### ❓ How to use
```python
# Define post_process functions
def concat_tokens_by_tag(keywords):
    keyphrase_tokens = []
    for id, label in keywords:
        if label == "B":
            keyphrase_tokens.append([id])
        elif label == "I":
            if len(keyphrase_tokens) > 0:
                keyphrase_tokens[len(keyphrase_tokens) - 1].append(id)
    return keyphrase_tokens


def extract_keyphrases(example, predictions, tokenizer, index=0):
    keyphrases_list = [
        (id, idx2label[label])
        for id, label in zip(
            np.array(example["input_ids"]).squeeze().tolist(), predictions[index]
        )
        if idx2label[label] in ["B", "I"]
    ]

    processed_keyphrases = concat_tokens_by_tag(keyphrases_list)
    extracted_kps = tokenizer.batch_decode(
        processed_keyphrases,
        skip_special_tokens=True,
        clean_up_tokenization_spaces=True,
    )
    return np.unique([kp.strip() for kp in extracted_kps])

```

```python
# Load model and tokenizer
model_name = "DeDeckerThomas/keyphrase-extraction-kbir-inspec"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
```
```python
# Inference
text = """
Keyword extraction is a technique in text analysis where you extract the important keywords 
from a text. Since this is a time-consuming process, Artificial Intelligence is used to automate it. 
Currently, classical machine learning methods, that use statistics and linguistics, are widely used 
for the extraction process. The fact that these methods have been widely used in the community has 
the advantage that there are many easy-to-use libraries. Now with the recent innovations in 
deep learning methods (such as recurrent neural networks and transformers, GANS, …), 
keyword extraction can be improved. These new methods also focus on the semantics 
and context of a document, which is quite an improvement.
""".replace("\n", "")

encoded_input = tokenizer(
    text,
    truncation=True,
    padding="max_length",
    max_length=max_length,
    return_tensors="pt",
)

output = model(**encoded_input)
logits = output.logits.detach().numpy()
predictions = np.argmax(logits, axis=2)

extracted_kps = extract_keyphrases(encoded_input, predictions, tokenizer)

print("***** Input Document *****")
print(text)

print("***** Prediction *****")
print(extracted_kps)
```

```
***** Input Document *****
Keyword extraction is a technique in text analysis where you extract the important keywords 
from a text. Since this is a time-consuming process, Artificial Intelligence is used to automate it. 
Currently, classical machine learning methods, that use statistics and linguistics, are widely used 
for the extraction process. The fact that these methods have been widely used in the community has 
the advantage that there are many easy-to-use libraries. Now with the recent innovations in 
deep learning methods (such as recurrent neural networks and transformers, GANS, …), 
keyword extraction can be improved. These new methods also focus on the semantics 
and context of a document, which is quite an improvement.

***** Prediction *****
['Artificial Intelligence' 'GANS' 'Keyword extraction'
 'classical machine learning methods' 'context' 'deep learning methods'
 'keyword extraction' 'linguistics' 'recurrent neural networks'
 'semantics' 'statistics' 'text analysis' 'transformers']
```

### 🛑 Limitations
* The model performs very well on abstracts of scientific papers. Please be aware that this model very domain-specific.
* Only works in English.

## [📚 Training Dataset](https://huggingface.co/datasets/midas/inspec)
## 👷‍♂️ Training procedure
The model is fine-tuned as a token classification problem where the text is labeled using the BIO scheme.
- B => Begin of a keyphrase
- I => Inside of a keyphrase
- O => Ouside of a keyphrase

For more information, you can take a look at the training notebook.

### Preprocessing
```python
def preprocess_fuction(all_samples_per_split):
    tokenized_samples = tokenizer.batch_encode_plus(
        all_samples_per_split[dataset_document_column],
        padding="max_length",
        truncation=True,
        is_split_into_words=True,
        max_length=max_length,
    )
    total_adjusted_labels = []
    for k in range(0, len(tokenized_samples["input_ids"])):
        prev_wid = -1
        word_ids_list = tokenized_samples.word_ids(batch_index=k)
        existing_label_ids = all_samples_per_split[dataset_biotags_column][k]
        i = -1
        adjusted_label_ids = []

        for wid in word_ids_list:
            if wid is None:
                adjusted_label_ids.append(lbl2idx["O"])
            elif wid != prev_wid:
                i = i + 1
                adjusted_label_ids.append(lbl2idx[existing_label_ids[i]])
                prev_wid = wid
            else:
                adjusted_label_ids.append(
                    lbl2idx[
                        f"{'I' if existing_label_ids[i] == 'B' else existing_label_ids[i]}"
                    ]
                )

        total_adjusted_labels.append(adjusted_label_ids)
    tokenized_samples["labels"] = total_adjusted_labels
    return tokenized_samples
```
## 📝Evaluation results

The model achieves the following results on the Inspec test set:

| Dataset           | P@5  | R@5  | F1@5 | P@10 | R@10 | F1@10 | P@M  | R@M  | F1@M |
|:-----------------:|:----:|:----:|:----:|:----:|:----:|:-----:|:----:|:----:|:----:|
| Inspec Test Set   | 0.53 | 0.47 | 0.46 | 0.36 | 0.58 | 0.41  | 0.58 | 0.60 | 0.56 |

For more information on the evaluation process, you can take a look at the keyphrase extraction evaluation notebook.
### Bibliography
Debanjan Mahata, Navneet Agarwal, Dibya Gautam, Amardeep Kumar, Sagar Dhiman, Anish Acharya, & Rajiv Ratn Shah. (2021). LDkp Dataset [Data set]. Zenodo. https://doi.org/10.5281/zenodo.5501744

Kulkarni, Mayank, Debanjan Mahata, Ravneet Arora, and Rajarshi Bhowmik. "Learning Rich Representation of Keyphrases from Text." arXiv preprint arXiv:2112.08547 (2021).

Sahrawat, Dhruva, Debanjan Mahata, Haimin Zhang, Mayank Kulkarni, Agniv Sharma, Rakesh Gosangi, Amanda Stent, Yaman Kumar, Rajiv Ratn Shah, and Roger Zimmermann. "Keyphrase extraction as sequence labeling using contextualized embeddings." In European Conference on Information Retrieval, pp. 328-335. Springer, Cham, 2020.