Update README.md
Browse files
README.md
CHANGED
@@ -113,7 +113,7 @@ German, French, Italian and Romansh documents in the [Swissdox@LiRI database](ht
|
|
113 |
|
114 |
This model was finetuned via unsupervised [SimCSE](http://dx.doi.org/10.18653/v1/2021.emnlp-main.552). The same sequence is passed to the encoder twice and the distance between the two resulting embeddings is minimized. Because of the drop-out, it will be encoded at slightly different positions in the vector space.
|
115 |
|
116 |
-
The fine-tuning script can be accessed [here](
|
117 |
|
118 |
#### Training Hyperparameters
|
119 |
|
@@ -143,14 +143,14 @@ The two evaluation tasks make use of the [20 Minuten dataset](https://www.zora.u
|
|
143 |
|
144 |
Embeddings are computed for the summary and content of each document. Subsequently, the embeddings are matched by minimizing cosine similarity scores betweend each summary and content embedding pair.
|
145 |
|
146 |
-
The performance is measured via accuracy, i.e. the ratio of correct vs. incorrect matches.
|
147 |
|
148 |
|
149 |
#### Evaluation via Text Classification
|
150 |
|
151 |
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
152 |
|
153 |
-
Articles with the topic tags "movies/tv series", "corona" and "football" (or related) are filtered from the corpus and split into training data (80%) and test data (20%). Subsequently, embeddings are set up for the train and test data. The test data is then classified using the training data via a k-nearest neighbor approach.
|
154 |
|
155 |
Note: For French and Italian, the training data remains in German, while the test data comprises of translations. This provides insights in the model's abilities in cross-lingual transfer.
|
156 |
|
|
|
113 |
|
114 |
This model was finetuned via unsupervised [SimCSE](http://dx.doi.org/10.18653/v1/2021.emnlp-main.552). The same sequence is passed to the encoder twice and the distance between the two resulting embeddings is minimized. Because of the drop-out, it will be encoded at slightly different positions in the vector space.
|
115 |
|
116 |
+
The fine-tuning script can be accessed [here](https://github.com/jgrosjean-mathesis/swissbert-for-sentence-embeddings/tree/main).
|
117 |
|
118 |
#### Training Hyperparameters
|
119 |
|
|
|
143 |
|
144 |
Embeddings are computed for the summary and content of each document. Subsequently, the embeddings are matched by minimizing cosine similarity scores betweend each summary and content embedding pair.
|
145 |
|
146 |
+
The performance is measured via accuracy, i.e. the ratio of correct vs. incorrect matches. The script can be found [here](https://github.com/jgrosjean-mathesis/swissbert-for-sentence-embeddings/tree/main).
|
147 |
|
148 |
|
149 |
#### Evaluation via Text Classification
|
150 |
|
151 |
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
152 |
|
153 |
+
Articles with the topic tags "movies/tv series", "corona" and "football" (or related) are filtered from the corpus and split into training data (80%) and test data (20%). Subsequently, embeddings are set up for the train and test data. The test data is then classified using the training data via a k-nearest neighbor approach. The script can be found [here](https://github.com/jgrosjean-mathesis/swissbert-for-sentence-embeddings/tree/main).
|
154 |
|
155 |
Note: For French and Italian, the training data remains in German, while the test data comprises of translations. This provides insights in the model's abilities in cross-lingual transfer.
|
156 |
|