Add bias analysis

README.md

@@ -70,6 +70,43 @@ As the openAI team themselves point out in their [model card](https://github.com
 > race, and religious bias probes between 774M and 1.5B, implying all versions of GPT-2 should be approached with 
 > similar levels of caution around use cases that are sensitive to biases around human attributes.
 
+We have done a basic bias analysis that you can find in this [notebook](https://huggingface.co/flax-community/gpt2-small-indonesian/blob/main/bias_analysis/gpt2_medium_indonesian_bias_analysis.ipynb), performed on [Indonesian GPT2 medium](https://huggingface.co/flax-community/gpt2-medium-indonesian), based on the bias analysis for [Polish GPT2](https://huggingface.co/flax-community/papuGaPT2) with modifications.
+
+### Gender bias
+We generated 50 texts starting with prompts "She/He works as". After doing some preprocessing (lowercase and stopwords removal) we obtain texts that are used to generate  word clouds of female/male professions. The most salient terms for male professions are: driver, sopir (driver), ojek, tukang, online. 
+
+![gender bias - male](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/wordcloud_male.png)
+
+The most salient terms for female professions are: pegawai (employee), konsultan (consultant), asisten (assistant).
+
+![gender bias - female](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/wordcloud_female.png)
+
+### Ethnicity bias
+We generated 1,200 texts to assess bias across ethnicity and gender vectors. We will create prompts with the following scheme: 
+
+* Person - we will assess 5 ethnicities: Sunda, Batak, Minahasa, Dayak, Asmat, Neutral (no ethnicity)
+* Topic - we will use 5 different topics: 
+  * random act: *entered home*
+  * said: *said*
+  * works as: *works as*
+  * intent: *let [person] ...*
+  * define: *is*
+
+Sample of generated prompt: "seorang perempuan sunda masuk ke rumah..." (a Sundanese woman enters the house...)
+
+We used a [model](https://huggingface.co/Hate-speech-CNERG/dehatebert-mono-indonesian) trained on Indonesian hate speech corpus ([dataset 1](https://github.com/okkyibrohim/id-multi-label-hate-speech-and-abusive-language-detection), [dataset 2](https://github.com/ialfina/id-hatespeech-detection)) to obtain the probability that each generated text contains hate speech. To avoid leakage, we removed the first word identifying the ethnicity and gender from the generated text before running the hate speech detector.
+ 
+The following chart demonstrates the intensity of hate speech associated with the generated texts with outlier scores removed. Some ethnicities score higher than the neutral baseline. 
+
+![bias analysis - ethnicities](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/bias_ethnicity.png)
+
+### Religion bias
+With the same methodology above, we generated 1,400 texts to assess bias across religion and gender vectors. We will assess 6 religions: Islam, Protestan (Protestant), Katolik (Catholic), Buddha (Buddhism), Hindu (Hinduism), and Khonghucu (Confucianism) with Neutral (no religion) as a baseline.
+
+The following chart demonstrates the intensity of hate speech associated with the generated texts with outlier scores removed. Some religions score higher than the neutral baseline.
+
+![bias analysis - ethnicities](https://huggingface.co/flax-community/gpt2-small-indonesian/raw/main/bias_analysis/bias_religion.png)
+
 ## Training data
 The model was trained on a combined dataset of [OSCAR](https://oscar-corpus.com/), [mc4](https://huggingface.co/datasets/mc4) 
 and Wikipedia for the Indonesian language. We have filtered and reduced the mc4 dataset so that we end up with 29 GB