Update README.md

README.md

@@ -1,8 +1,10 @@
-## Dataset Summary
-
-A dataset for benchmarking keyphrase extraction and generation techniques from abstracts of english scientific papers. For more details about the dataset please refer the original paper - [https://dl.acm.org/doi/pdf/10.3115/1119355.1119383](https://dl.acm.org/doi/pdf/10.3115/1119355.1119383)
+A dataset for benchmarking keyphrase extraction and generation techniques from abstracts of English scientific papers. For more details about the dataset please refer the original paper - [https://dl.acm.org/doi/pdf/10.3115/1119355.1119383](https://dl.acm.org/doi/pdf/10.3115/1119355.1119383)
 Data source - [https://github.com/boudinfl/ake-datasets/tree/master/datasets/Inspec](https://github.com/boudinfl/ake-datasets/tree/master/datasets/Inspec)
 
+## Dataset Summary
+The Inspec dataset was originally proposed by *Hulth* in the paper titled - [Improved automatic keyword extraction given more linguistic knowledge](https://aclanthology.org/W03-1028.pdf) in the year 2003. The dataset consists of abstracts of 2,000 English scientific papers from the [Inspec database](https://clarivate.com/webofsciencegroup/solutions/webofscience-inspec/). The abstracts are from papers belonging to the scientific domains of *Computers and Control* and *Information Technology* published between 1998 to 2002. Each abstract has two sets of keyphrases annotated by professional indexers - *controlled* and *uncontrolled*. The *controlled* keyphrases are obtained from the Inspec thesaurus and therefore are often not present in the abstract's text. Only 18.1% of the *controlled* keyphrases are actually present in the abstract's text. The *uncontrolled* keyphrases are those selected by the indexers after reading the full-length scientific articles and 76.2% of them are present in the abstract's text. There is no information in the original paper about how these 2,000 scientific papers were selected. It is unknown whether the papers were randomly selected out of all the papers published between 1998-2002 in the *Computers and Control* and *Information Technology* domains or were there only 2,000 papers in this domain that were indexed by Inspec. The train, dev and test splits of the data were arbitrarily chosen. 
+
+One of the key aspect of this dataset which makes it unique is that it provides keyphrases assigned by professional indexers, which is uncommon in the keyphrase literature. Most of the datasets in this domain have author assigned keyphrases as the ground truth. The dataset shared over here does not explicitly presents the *controlled* and *uncontrolled* keyphrases instead it only categorizes the keyphrases into *extractive* and *abstractive*. **Extractive keyphrases** are those that could be found in the input text and the **abstractive keyphrases** are those that are not present in the input text. In order to get all the meta-data about the documents and keyphrases please refer to the [original source](https://github.com/boudinfl/ake-datasets/tree/master/datasets/Inspec) from which the dataset was taken from. The main motivation behind making this dataset available in the form as presented over here is to make it easy for the researchers to programmatically download it and evaluate their models for the tasks of keyphrase extraction and generation. As keyphrase extraction by treating it as a sequence tagging task and using contextual language models has become popular we have also made the token tags available in the BIO tagging format.  
 
 ## Dataset Structure
 
@@ -20,7 +22,7 @@ Data source - [https://github.com/boudinfl/ake-datasets/tree/master/datasets/Ins
 
 |Split| #datapoints  |
 |--|--|
-| Train | 1000 |
+| Train | 1,000 |
 | Test | 500 |
 | Validation | 500 |
 
@@ -165,22 +167,6 @@ print("Abstractive/absent Keyphrases: ", test_sample["abstractive_keyphrases"])
 print("\n-----------\n")
 ```
 
-## Model Benchmarking
-
-Below are the performance of the different supervised models on this corpus as reported by [https://arxiv.org/pdf/2112.08547.pdf](https://arxiv.org/pdf/2112.08547.pdf)
-
-### Keyphrase Extraction
-
-<br>
-<p align="center">
-  <img src="https://huggingface.co/datasets/midas/inspec/resolve/main/inspec_performance.png" alt="Model Benchmarks"  width="50%"/>
-  <br>
-</p>
-<br>
-
-
-### Keyphrase Generation
-
 
 
 ## Citation Information