matthewdicks98 commited on
Commit
6a99fcf
·
verified ·
1 Parent(s): efadc00

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -7
README.md CHANGED
@@ -27,15 +27,12 @@ size_categories:
27
 
28
  - [**NOSIBLE Forward Looking v1.1 Base**](https://huggingface.co/NOSIBLE/forward-looking-v1.1-base)
29
 
30
-
31
  ## What is it?
32
 
33
- The NOSIBLE Forward Looking Dataset is an open collection of **100,000** cleaned, deduplicated, and forward-looking labeled news. Each label reflects the forward looking nature of a short text snippet, categorizing it based on whether the described events are likely to be **forward** or **not-forward** looking.
34
 
35
  All text is sourced from the **NOSIBLE Search Feeds** product using a curated set of finance-related queries. Forward looking labels are assigned through a multi-stage, LLM-based annotation pipeline (described below).
36
 
37
-
38
-
39
  ## How to use it
40
  Using the [HuggingFace datasets library](https://huggingface.co/docs/datasets/):
41
 
@@ -78,9 +75,9 @@ The following is an example sample from the dataset:
78
 
79
  ### Data Fields
80
 
81
- - `text` (string): A text chunk from a document.
82
- - `label` (string): The temporal orientation label of the text, sourced from LLMs and refined with active learning (an iterative relabeling process).
83
- - `netloc` (string): The network location (domain) of the document.
84
  - `url` (string): The URL of the document.
85
 
86
  ## Dataset creation
@@ -118,6 +115,8 @@ The algorithm outline is as follows:
118
  - Repeat until no additional samples require relabeling.
119
  5. This is the final dataset used for training the [NOSIBLE Forward Looking v1.1 Base](https://huggingface.co/NOSIBLE/forward-looking-v1.1-base) model.
120
 
 
 
121
  ## Additional information
122
 
123
  ### License
 
27
 
28
  - [**NOSIBLE Forward Looking v1.1 Base**](https://huggingface.co/NOSIBLE/forward-looking-v1.1-base)
29
 
 
30
  ## What is it?
31
 
32
+ The NOSIBLE Forward Looking Dataset is an open collection of **100,000** cleaned, deduplicated, and forward-looking labeled news. Each label reflects the temporal orientation of a short text snippet, categorizing it based on whether the statement is **forward** or **not forward** looking.
33
 
34
  All text is sourced from the **NOSIBLE Search Feeds** product using a curated set of finance-related queries. Forward looking labels are assigned through a multi-stage, LLM-based annotation pipeline (described below).
35
 
 
 
36
  ## How to use it
37
  Using the [HuggingFace datasets library](https://huggingface.co/docs/datasets/):
38
 
 
75
 
76
  ### Data Fields
77
 
78
+ - `text` (string): A text chunk from a search result.
79
+ - `label` (string): The temporal orientation label.
80
+ - `netloc` (string): The domain name of the source document.
81
  - `url` (string): The URL of the document.
82
 
83
  ## Dataset creation
 
115
  - Repeat until no additional samples require relabeling.
116
  5. This is the final dataset used for training the [NOSIBLE Forward Looking v1.1 Base](https://huggingface.co/NOSIBLE/forward-looking-v1.1-base) model.
117
 
118
+ We used [`OpenAI: GPT-5.1`](https://openrouter.ai/openai/gpt-5.1) as the oracle.
119
+
120
  ## Additional information
121
 
122
  ### License