Danbnyn commited on
Commit
756a678
1 Parent(s): e9b9d17

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +99 -3
README.md CHANGED
@@ -1,3 +1,99 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - en
5
+ tags:
6
+ - finance
7
+ size_categories:
8
+ - 10K<n<100K
9
+ ---
10
+
11
+ # Dataset Card for Processed Financial News Articles from Reuters (2006-2013)
12
+
13
+ This dataset consists of 105,359 financial news articles originally sourced from Reuters, covering the period from 2006 to 2013. It includes processed texts with an additional 'Summary' field, suitable for use in NLP and financial trend analysis.
14
+
15
+ ## Dataset Details
16
+
17
+ ### Dataset Description
18
+
19
+ The dataset contains English-language financial news articles collected from Reuters. It is designed for natural language processing tasks, financial analysis, and trend detection during the specified period.
20
+
21
+ - **Curated by:** Dan Benayoun
22
+ - **Shared by:** Dan Benayoun
23
+ - **Language(s) (NLP):** English
24
+ - **License:** Apache-2.0
25
+
26
+ ### Dataset Sources
27
+
28
+ - **Paper:** [Financial News Dataset Analysis](https://emnlp2014.org/papers/pdf/EMNLP2014148.pdf) - This paper discusses methodologies for processing financial news datasets.
29
+
30
+ ## Uses
31
+
32
+ ### Direct Use
33
+
34
+ This dataset is intended for academic research, algorithm training, and development of NLP models that require financial context, particularly for tasks such as sentiment analysis, event detection, and trend tracking over time.
35
+
36
+ ### Out-of-Scope Use
37
+
38
+ The dataset should not be used for real-time financial decision-making or trading due to its historical nature and the inherent biases of media reporting.
39
+
40
+ ## Dataset Structure
41
+
42
+ ### Data Instances
43
+
44
+ A typical instance includes fields like `Headline`, `Journalists`, `Date`, `Link`, `Article`, and `Summary`.
45
+
46
+ ### Data Splits
47
+
48
+ The dataset is not divided into standard training, validation, or test splits.
49
+
50
+ ## Dataset Creation
51
+
52
+ ### Curation Rationale
53
+
54
+ This dataset was curated to assist researchers and developers in analyzing financial news trends and patterns across several years, with the addition of summaries to aid in quicker understanding and analysis.
55
+
56
+ ### Source Data
57
+
58
+ #### Data Collection and Processing
59
+
60
+ The articles were originally scraped from Reuters and processed for easier consumption in NLP tasks, including the generation of summaries for each article.
61
+
62
+ #### Who are the source data producers?
63
+
64
+ The text of the articles is produced by journalists and financial analysts at Reuters.
65
+
66
+ ## Annotations
67
+
68
+ ### Personal and Sensitive Information
69
+
70
+ The dataset contains information that could be considered personal or sensitive as it includes names of people, companies, and possibly their financial data.
71
+
72
+ ## Bias, Risks, and Limitations
73
+
74
+ The dataset reflects the potential biases of Reuters' reporting style and period-specific geopolitical influences on financial reporting.
75
+
76
+ ### Recommendations
77
+
78
+ Care should be taken to account for these biases when using the dataset for modeling or analysis. Researchers should also consider the implications of using summary data which may omit critical information or context.
79
+
80
+ ## Citation
81
+
82
+ **BibTeX:**
83
+
84
+ ```bibtex
85
+ @misc{reuters_financial_news_2006_2013,
86
+ author = {Dan Benayoun},
87
+ title = {Processed Dataset of Financial News Articles from Reuters (2006-2013)},
88
+ year = {Year of dataset release},
89
+ publisher = {Dataset publisher},
90
+ }
91
+
92
+ @misc{BloombergReutersDataset2015,
93
+ author = {Philippe Remy, Xiao Ding},
94
+ title = {Financial News Dataset from Bloomberg and Reuters},
95
+ year = {2015},
96
+ publisher = {GitHub},
97
+ journal = {GitHub repository},
98
+ howpublished = {\url{https://github.com/philipperemy/financial-news-dataset}},
99
+ }