Semantic Relation Classification——via Convolution Neural Network

From statwiki
Revision as of 15:51, 22 November 2020 by R6gong (talk | contribs) (Created page with " After featurizing all words in the sentence. The sentence of length N can be expressed as a vector of length $N$, which looks like $$e=[e_{1},e_{2},\ldots,e_{N}]$$ and eac...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.

After featurizing all words in the sentence. The sentence of length N can be expressed as a vector of length $N$, 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 $W\in\mathbb{R}^{(d^{w}+2d^{wp})\times k}$ to produce a new feature, defiend as $$c_{i}=tanh(W\cdot e_{i:i+k-1}+bias)$$ This process is applied to all subsets of features with length $k$ starting from the first one. Then a mapped feature factor $$c=[c_{1},c_{2},\ldots,c_{N-k+1}]$$ is produced.\\ \\ The max pooling operation is used, the $\hat{c}=max\{c\}$ was picked. With different weight filter, different mapped feature vectors can be obtained. Finally, the original sentence $e$ can be converted into a new representation $r_{x}$ with a fixed length. For example, if there are 5 filters, then there are 5 features ($\hat{c}$) picked to create $r_{x}$ for each $x$.\\ Then, the score vector $$s(x)=W^{classes}r_{x}$$ is obtained which represented the score for each class, given $x$'s entities' relation will be classified as the one with the highest score. The $W^{classes}$ here is the model being trained.\\ \\ To improve the performance, “Negative Sampling" was used. Given the trained data point $\tilde{x}$, and its correct class $\tilde{y}$. Let $I=Y\setminus\{\tilde{y}\}$ represent the incorrect labels for $x$. 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 $m_{+}$, $m_{-}$, and penalty scale factor $\gamma$. 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.