Update README.md

README.md

@@ -9,13 +9,6 @@ size_categories:
 - 100M<n<1B
 ---
 
-## Dataset Description
-- **Homepage:** [Dell Research Harvard](https://dell-research-harvard.github.io/)
-- <!--- **Repository:** [GitHub repository]()--->
-- <!--- **Paper:** [NeurIPS Submission]()--->
-- **Point of Contact:** Melissa Dell
-
-
 ![headline_image_with_title.jpg](https://s3.amazonaws.com/moonup/production/uploads/61654589b5ec555e8e9c203a/fJiAy43PFD3FBa1aMSzHz.jpeg)
 
 ## Dataset Summary
@@ -37,13 +30,34 @@ It contains headlines from all 50 states.
 
 
 ## Dataset Structure 
+Each year in the dataset is divided into a distinct file (eg. 1952_headlines.pkl), giving a total of 70 files.  
+
+Each of these contains a list of dictionaries of the form: 
 
+```python
+{
+    "headline": "...",
+    "date": "...",
+    "newspaper": "...",
+    "state": "...",
+    "cluster": "..."
+}
+```
 
-## Data Instances
+** Headline: headline text.
+** Date: the date of publication of the newspaper article, as a string in the form YYYY-MM-DD.
+** Newspaper: name of the newspaper that published the headline.
+** State: state of the newspaper that published the headline.
+** Cluster: a number that is shared with all other headlines for the same article. This number is unique across all year files. 
+
+We chose to present the dataset in the form of clusters, rather than pairs to eliminate duplication of text data and minimise the storage size of the datasets. Below we give an example of how to convert the dataset into pairs.
 
 
 ## Usage
 
 
+
+
 ## Contacts
-This dataset was created by Emily Silcock and Melissa Dell. For more information, see https://dell-research-harvard.github.io/
+This dataset was created by Emily Silcock and Melissa Dell. For more information, see [Dell Research Harvard](https://dell-research-harvard.github.io/). 
+