iperbole commited on
Commit
754b4db
1 Parent(s): 2fb367f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +90 -0
README.md CHANGED
@@ -83,3 +83,93 @@ configs:
83
  - split: test
84
  path: physics/test-*
85
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83
  - split: test
84
  path: physics/test-*
85
  ---
86
+
87
+ # Prerequisite RElation LEARNing (PRELEARN)
88
+
89
+ Original Paper: https://ceur-ws.org/Vol-2765/paper164.pdf
90
+
91
+ This dataset contains a collection of binary-labelled concept pairs (A,B) extracted from textbooks on four domains: **data mining**, **geometry**, **physics** and **precalculus**.
92
+ Then, domain experts were asked to manually annotate if pairs of concepts showed a prerequisite relation or not, therefore the dataset consists of both positive and negative concept pairs.
93
+
94
+ We obtained the data from the original repository, making only one modification: undersampling the training data. To evaluate generative models in in-context learning, it's essential to have a balanced distribution for sampling examples in a few-shot setting. The undersampling process was carried out randomly, and separately for each domain.
95
+
96
+ ## Example
97
+
98
+ Here you can see the structure of the single sample in the present dataset.
99
+
100
+ ```json
101
+ {
102
+ "concept_A": string, # text of the concept A
103
+ "wikipedia_passage_concept_A": string, # paragraph of wikipedia corresponding to concept A
104
+ "concept_B": string, # text of the concept B
105
+ "wikipedia_passage_concept_B": string, # paragraph of wikipedia corresponding to concept B
106
+ "target": int, # 0: B non è preconcetto di A, 1: B è preconcetto di A
107
+ }
108
+ ```
109
+
110
+ ## Statitics
111
+
112
+ | PRELEARN Data Mining | 0 | 1 |
113
+ | :--------: | :----: | :----: |
114
+ | Training | 109 | 109 |
115
+ | Test | 50 | 49 |
116
+
117
+ | PRELEARN Physics | 0 | 1 |
118
+ | :--------: | :----: | :----: |
119
+ | Training | 315 | 315 |
120
+ | Test | 100 | 100 |
121
+
122
+ | PRELEARN Geometry | 0 | 1 |
123
+ | :--------: | :----: | :----: |
124
+ | Training | 332 | 332 |
125
+ | Test | 100 | 100 |
126
+
127
+ | PRELEARN Precalculus | 0 | 1 |
128
+ | :--------: | :----: | :----: |
129
+ | Training | 408 | 408 |
130
+ | Test | 100 | 100 |
131
+
132
+ ## Proposed Prompts
133
+
134
+ Here we will describe the prompt given to the model over which we will compute the perplexity score, as model's answer we will chose the prompt with lower perplexity.
135
+ Moreover, for each subtask, we define a description that is prepended to the prompts, needed by the model to understand the task.
136
+
137
+ Description of the task: "Dati due concetti, indica se primo concetto è un prerequisito o meno per il secondo.\nUn concetto A è prerequisito per un concetto B, se per comprendere B devo prima aver compreso A.\nI seguenti concetti appartengono al dominio: {{domain}}.\n\n"
138
+
139
+ ### Cloze Style:
140
+
141
+ Label (**B non è prerequisito di A**): "{{concept_B}} non è un prerequisito per {{concept_A}}"
142
+
143
+ Label (**B è prerequisito di A**): "{{concept_B}} è un prerequisito per {{concept_A}}"
144
+
145
+ ### MCQA Style:
146
+
147
+ ```
148
+ Domanda: il concetto {{concept_B}} è un prerequisito per la comprensione del concetto {{concept_A}}? Rispondi sì o no:
149
+ ```
150
+
151
+ ## Some Results
152
+
153
+ The following results are given by the Cloze-style prompting over some english and italian-adapted LLMs.
154
+
155
+ | PRELEARN (AVG) | ACCURACY (15-shots) |
156
+ | :-----: | :--: |
157
+ | Gemma-2B | 60.12 |
158
+ | QWEN2-1.5B | 57.00 |
159
+ | Mistral-7B | 64.50 |
160
+ | ZEFIRO | 64.76 |
161
+ | Llama-3-8B | 60.63 |
162
+ | Llama-3-8B-IT | 63.76 |
163
+ | ANITA | 63.77 |
164
+
165
+ ## Aknwoledge
166
+
167
+ We want to thanks this resource's authors for publicly releasing such an interesting dataset.
168
+
169
+ Further, We want to thanks the student of [MNLP-2024 course](https://naviglinlp.blogspot.com/), where with their first homework tried different interesting prompting strategies.
170
+
171
+ The data is freely available through this [link](https://live.european-language-grid.eu/catalogue/corpus/8084).
172
+
173
+ ## License
174
+
175
+ The data come under the license [Creative Commons Attribution Non Commercial Share Alike 4.0 International](https://creativecommons.org/licenses/by-nc-sa/4.0/)