Semantic Relation Classification——via Convolution Neural Network

Presented by

Rui Gong, Xinqi Ling, Di Ma,Xuetong Wang

Introduction

One of the emerging trends of natural language technologies is their use for the humanities and sciences (Gbor et al., 2018). SemEval 2018 Task 7 mainly solves the problem of relation extraction and classification of two entities in the same sentence into 6 potential relations. The 6 relations are USAGE, RESULT, MODEL-FEATURE, PART WHOLE, TOPIC, and COMPARE.

Data comes from 350 scientific paper abstracts, which has 1228 and 1248 annotated sentences for two tasks. For each data, an example sentence was chosen with its right and left sentences, as well as an indicator showing whether the relation is reserved, then a prediction is made.

Three models were used for the prediction: Linear Classifiers, Long Short-Term Memory(LSTM), and Convolutional Neural Networks (CNN). In the end, the CNN model category performed the best, so the article specifically submitted the final submission for this model. By using the learned custom word embedding function, the research team added a variant of negative sampling, thereby improving performance and surpassing ordinary CNN.

Previous Work

SemEval 2010 Task 8(Hendrickx et al., 2010) explored the classification of natural language relations and studied the 9 relations between word pairs. However, it is not designed for scientific text analysis, and their challenge differs from the challenge of this paper in its generalizability; this paper’s relations are specific to ACL papers (e.g. MODEL-FEATURE) whereas the 2010 relations are more general, and might necessitate more common-sense knowledge than the 2018 relations. Xu et al. (2015a) and Santos et al. (2015) , both of them applied CNN with negative sampling to finish task7. The 2017 SemEval Task 10 also featured relation extraction within scientific publications.

Algorithm

This is the architecture of CNN. We first transform a sentence via Feature embeddings. Basically, we transform each sentence into continuous word embeddings:

$$ (e^{w_i}) $$

And word position embeddings: $$ (e^{wp_i}): e_i = [e^{w_i}, e^{wp_i}] $$

In the word embeddings, we got a vocabulary ‘V’, and we will make an embedding word matrix based on the position of the word in the vocabulary. This matrix is trainable and needs to be initialized by pre-trained embedding vectors. In the word position embeddings, we first need to input some words named ‘entities’ and they are the key for the machine to determine the sentence’s relation. During this process, if we have two entities, we will use the relative position of them in the sentence to make the embeddings. We will output two vectors and one of them keeps track of the first entity relative position in the sentence ( we will make the entity recorded as 0, the former word recorded as -1 and the next one 1, etc. ). And the same procedure for the second entity. Finally, we will get two vectors concatenated as the position embedding.

After the embeddings, the model will transform the embedded sentence to a fix-sized representation of the whole sentence via the convolution layer, finally after the max-pooling to reduce the dimension of the output of the layers, we will get a score for each relation class via a linear transformation.

After featurizing all words in the sentence. The sentence of length N can be expressed as a vector of length [math]\displaystyle{ N }[/math], which looks like $$e=[e_{1},e_{2},\ldots,e_{N}]$$ and each entry represents a token of the word. Also, to apply convolutional neural network, the subsets of features $$e_{i:i+j}=[e_{i},e_{i+1},\ldots,e_{i+j}]$$ is given to a weight matrix [math]\displaystyle{ W\in\mathbb{R}^{(d^{w}+2d^{wp})\times k} }[/math] to produce a new feature, defiend as $$c_{i}=\text{tanh}(W\cdot e_{i:i+k-1}+bias)$$ This process is applied to all subsets of features with length [math]\displaystyle{ k }[/math] starting from the first one. Then a mapped feature factor is produced: $$c=[c_{1},c_{2},\ldots,c_{N-k+1}]$$

The max pooling operation is used, the [math]\displaystyle{ \hat{c}=max\{c\} }[/math] was picked. With different weight filter, different mapped feature vectors can be obtained. Finally, the original sentence [math]\displaystyle{ e }[/math] can be converted into a new representation [math]\displaystyle{ r_{x} }[/math] with a fixed length. For example, if there are 5 filters, then there are 5 features ([math]\displaystyle{ \hat{c} }[/math]) picked to create [math]\displaystyle{ r_{x} }[/math] for each [math]\displaystyle{ x }[/math].

Then, the score vector $$s(x)=W^{classes}r_{x}$$ is obtained which represented the score for each class, given [math]\displaystyle{ x }[/math]'s entities' relation will be classified as the one with the highest score. The [math]\displaystyle{ W^{classes} }[/math] here is the model being trained.

To improve the performance, “Negative Sampling" was used. Given the trained data point [math]\displaystyle{ \tilde{x} }[/math], and its correct class [math]\displaystyle{ \tilde{y} }[/math]. Let [math]\displaystyle{ I=Y\setminus\{\tilde{y}\} }[/math] represent the incorrect labels for [math]\displaystyle{ x }[/math]. Basically, the distance between the correct score and the positive margin, and the negative distance (negative margin plus the second largest score) should be minimized. So the loss function is $$L=\log(1+e^{\gamma(m^{+}-s(x)_{y})})+\log(1+e^{\gamma(m^{-}-\mathtt{max}_{y'\in I}(s(x)_{y'}))})$$ with margins [math]\displaystyle{ m_{+} }[/math], [math]\displaystyle{ m_{-} }[/math], and penalty scale factor [math]\displaystyle{ \gamma }[/math]. The whole training is based on ACL anthology corpus and there are 25,938 papers with 136,772,370 tokens in total, and 49,600 of them are unique.

Results

In machine learning, the most important part is to tune the hyper-parameters. Unlike traditional hyper-parameter optimization, there are some modifications to the model in order to increase performance on the test set. There are 5 modifications that we can apply:

1. Merged Training Sets. It combined two training sets to increase the data set size and it improves the equality between classes to get better predictions.

2. Reversal Indicate Features. It added a binary feature.

3. Custom ACL Embeddings. It embedded word vector to an ACL-specific corps.

4. Context words. Within the sentence, it varies in size on a context window around the entity-enclosed text.

5. Ensembling. It used different early stop and random initializations to improve the predictions.

These modifications performances well on the training data and they are shown in table 3.

As we can see the best choice for this model is ensembling. Because the random initialization made the data more natural and avoided the overfit. During the training process, there are some methods such that they can only increase the score on the cross-validation test sets but hurt the performance on the overall macro-F1 score. Thus, these methods were eventually ruled out.

There are six submissions in total. Three for each training set and the result is shown in figure 2.

The best submission for the training set 1.1 is the third submission which did not use the cross-validation as the test set. Instead, it runs a constant number of training epochs, and based on the training data it can be chosen by cross-validation. The best submission for the training set 1.2 is the first submission which extracted 10% of the training data as validation accuracy on the test set predictions. All in all, early stopping cannot always be based on the accuracy of the validation set since it cannot guarantee to get better performance on the real test set. Thus, we have to try new approaches and combine them together to see the prediction results. Also, doing stratification will certainly improve the performance of the test data.

Conclusions

Throughout the process, linear classifiers, sequential random forest, LSTM, and CNN models are tested. Variations are applied to the models. Among all variations, vanilla CNN with negative sampling and ACL-embedding has significantly better performance than all others. Attention-based pooling, up-sampling, and data augmentation are also tested, but they barely perform positive increment on the behavior.

Critiques

- Applying this in news apps might be beneficial to improve readability by highlighting specific important sections.

- In the section of previous work, the author mentioned 9 natural language relationship between the word pairs. Among them, 6 potential relationships are USAGE, RESULT, MODEL-FEATURE,PART WHOLE, TOPIC, and COMPARE. It would help the readers to better understand if all 9 relationships are listed in the summary.

-This topic is interesting and this application might be helpful for some educational websites to improve their website to help readers focus on the important points. I think it will be nice to use Latex to type the equation in the sentence rather than center the equation on the next line. I think it will be interesting to discuss applying this way to other languages such as Chinese, Japanese, etc.

References

Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.

DragomirR. Radev, Pradeep Muthukrishnan, Vahed Qazvinian, and Amjad Abu-Jbara. 2013. The ACL anthology network corpus. Language Resources and Evaluation, pages 1–26.

Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013a. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781.

Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013b. Distributed representations of words and phrases and their compositionality. In Advances in neural information processing systems, pages 3111–3119.

Kata Gbor, Davide Buscaldi, Anne-Kathrin Schumann, Behrang QasemiZadeh, Hafa Zargayouna, and Thierry Charnois. 2018. Semeval-2018 task 7:Semantic relation extraction and classification in scientific papers. In Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval2018), New Orleans, LA, USA, June 2018.