File size: 7,609 Bytes
8e9c340 87fae06 8e9c340 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 |
---
language:
- de
- en
license: mit
tags:
- sentence_embedding
- search
- pytorch
- xlm-roberta
- roberta
- xlm-r-distilroberta-base-paraphrase-v1
- paraphrase
datasets:
- STSbenchmark
metrics:
- Spearman’s rank correlation
- cosine similarity
---
# Cross English & German RoBERTa for Sentence Embeddings
This model is intended to [compute sentence (text) embeddings](https://www.sbert.net/docs/usage/computing_sentence_embeddings.html) for English and German text. These embeddings can then be compared with [cosine-similarity](https://en.wikipedia.org/wiki/Cosine_similarity) to find sentences with a similar semantic meaning. For example this can be useful for [semantic textual similarity](https://www.sbert.net/docs/usage/semantic_textual_similarity.html), [semantic search](https://www.sbert.net/docs/usage/semantic_search.html), or [paraphrase mining](https://www.sbert.net/docs/usage/paraphrase_mining.html). To do this you have to use the [Sentence Transformers Python framework](https://github.com/UKPLab/sentence-transformers).
The speciality of this model is that it also works cross-lingually. Regardless of the language, the sentences are translated into very similar vectors according to their semantics. This means that you can, for example, enter a search in German and find results according to the semantics in German and also in English. Using a xlm model and _multilingual finetuning with language-crossing_ we reach performance that even exceeds the best current dedicated English large model (see Evaluation section below).
> Sentence-BERT (SBERT) is a modification of the pretrained BERT network that use siamese and triplet network structures to derive semantically meaningful sentence embeddings that can be compared using cosine-similarity. This reduces the effort for finding the most similar pair from 65hours with BERT / RoBERTa to about 5 seconds with SBERT, while maintaining the accuracy from BERT.
Source: [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://arxiv.org/abs/1908.10084)
This model is fine-tuned from [Philip May](https://may.la/) and open-sourced by [T-Systems-onsite](https://www.t-systems-onsite.de/). Special thanks to [Nils Reimers](https://www.nils-reimers.de/) for your awesome open-source work, the Sentence Transformers, the models and your help on GitHub.
## How to use
**The usage description above - provided by Hugging Face - is wrong for sentence embeddings! Please use this:**
To use this model install the `sentence-transformers` package (see here: <https://github.com/UKPLab/sentence-transformers>).
```python
from sentence_transformers import SentenceTransformer
model = SentenceTransformer('T-Systems-onsite/cross-en-de-roberta-sentence-transformer')
```
For details of usage and examples see here:
- [Computing Sentence Embeddings](https://www.sbert.net/docs/usage/computing_sentence_embeddings.html)
- [Semantic Textual Similarity](https://www.sbert.net/docs/usage/semantic_textual_similarity.html)
- [Paraphrase Mining](https://www.sbert.net/docs/usage/paraphrase_mining.html)
- [Semantic Search](https://www.sbert.net/docs/usage/semantic_search.html)
- [Cross-Encoders](https://www.sbert.net/docs/usage/cross-encoder.html)
- [Examples on GitHub](https://github.com/UKPLab/sentence-transformers/tree/master/examples)
## Training
The base model is [xlm-roberta-base](https://huggingface.co/xlm-roberta-base). This model has been further trained by [Nils Reimers](https://www.nils-reimers.de/) on a large scale paraphrase dataset for 50+ languages. [Nils Reimers](https://www.nils-reimers.de/) about this [on GitHub](https://github.com/UKPLab/sentence-transformers/issues/509#issuecomment-712243280):
>A paper is upcoming for the paraphrase models.
>
>These models were trained on various datasets with Millions of examples for paraphrases, mainly derived from Wikipedia edit logs, paraphrases mined from Wikipedia and SimpleWiki, paraphrases from news reports, AllNLI-entailment pairs with in-batch-negative loss etc.
>
>In internal tests, they perform much better than the NLI+STSb models as they have see more and broader type of training data. NLI+STSb has the issue that they are rather narrow in their domain and do not contain any domain specific words / sentences (like from chemistry, computer science, math etc.). The paraphrase models has seen plenty of sentences from various domains.
>
>More details with the setup, all the datasets, and a wider evaluation will follow soon.
The resulting model called `xlm-r-distilroberta-base-paraphrase-v1` has been released here: <https://github.com/UKPLab/sentence-transformers/releases/tag/v0.3.8>
Building on this cross language model we fine-tuned it for English and German language on the [STSbenchmark](http://ixa2.si.ehu.es/stswiki/index.php/STSbenchmark) dataset. For German language we used the dataset of our [German STSbenchmark dataset](https://github.com/t-systems-on-site-services-gmbh/german-STSbenchmark) which has been translated with [deepl.com](https://www.deepl.com/translator). Additionally to the German and English training samples we generated samples of English and German crossed. We call this _multilingual finetuning with language-crossing_. It doubled the traing-datasize and tests show that it further improves performance.
We did an automatic hyperparameter search for 33 trials with [Optuna](https://github.com/optuna/optuna). Using 10-fold crossvalidation on the deepl.com test and dev dataset we found the following best hyperparameter:
- batch_size = 8
- num_epochs = 2
- lr = 1.026343323298136e-05,
- eps = 4.462251033010287e-06
- weight_decay = 0.04794438776350409
- warmup_steps_proportion = 0.1609010732760181
The final model was trained with these hyperparameters on the combination of the train and dev datasets from English, German and the crossings of them. The testset was left for testing.
# Evaluation
The evaluation has been done on English, German and both languages crossed with the STSbenchmark test data. The evaluation-code is available on [Colab](https://colab.research.google.com/drive/1gtGnKq_dYU_sDYqMohTYVMVpxMJjyH0M?usp=sharing). As the metric for evaluation we use the Spearman’s rank correlation between the cosine-similarity of the sentence embeddings and STSbenchmark labels.
| Model Name | Spearman<br/>German | Spearman<br/>English | Spearman<br/>EN-DE & DE-EN<br/>(cross) |
|---------------------------------------------------------------|-------------------|--------------------|------------------|
| xlm-r-distilroberta-base-paraphrase-v1 | 0.8079 | 0.8350 | 0.7983 |
| [xlm-r-100langs-bert-base-nli-stsb-mean-tokens](https://huggingface.co/sentence-transformers/xlm-r-100langs-bert-base-nli-stsb-mean-tokens) | 0.7877 | 0.8465 | 0.7908 |
| xlm-r-bert-base-nli-stsb-mean-tokens | 0.7877 | 0.8465 | 0.7908 |
| [roberta-large-nli-stsb-mean-tokens](https://huggingface.co/sentence-transformers/roberta-large-nli-stsb-mean-tokens) | 0.6371 | 0.8639 | 0.4109 |
| [T-Systems-onsite/<br/>german-roberta-sentence-transformer-v2](https://huggingface.co/T-Systems-onsite/german-roberta-sentence-transformer-v2) | 0.8529 | 0.8634 | 0.8415 |
| **T-Systems-onsite/<br/>cross-en-de-roberta-sentence-transformer** | **0.8550** | **0.8660** | **0.8525** |
|