1 Appendix↩︎

1.1 Relations and phrases↩︎

Table 1 shows the relations among entities and their corresponding synonyms. The relation \(\mathrm{identical}\) links a entity and a synonym to claim that the synonym refers to the entity. During training, the relation \(\mathrm{identical}\) links \(\mathrm{[MASK]}\) and a synonym or entity to extract the textual feature. Among other relations, \(\mathrm{is\_a}\) is the most common relation, which describes the subsumption relation between a child entity and a parent entity. We transform these relations into phrases to put them in templates used by our Prompt-based model.

1.2 Implementation details↩︎

Details about prompt-based methods For prompt-based methods (Prompt, GraphPrompt (w/o \(\mathcal{T}^2\)), and GraphPrompt), we trained the model with \(\mathcal{L}_\mathrm{c}\) for 400 iterations to warm-up entity embedding \(\boldsymbol{x}_v\). For zero-shot setting, we followed the bi-encoder architecture that uses two encoders for entities and synonyms. Every time we updated the embeddings of entities \(\boldsymbol{x}_{v'}, v'\in \mathcal{V}\), we had to run the encoder for every entity. For few-shot learning, we found that the entity embedding can be directly trained with an embedding layer. We used the entity side of the bi-encoder to generate entity embedding \(\boldsymbol{x}_{v'}^0\), and used this embedding to initialize the embedding layer. Then we used embeddings from this trainable embedding layer to replace the \(\mathrm{sg}(\boldsymbol{x}_{v'})\) and \(\mathrm{sg}(\boldsymbol{x}_{v})\) term in the loss.

Details about second-order GraphPrompt The second-order GraphPrompt (GraphPrompt in Table 1) actually didn’t include zeroth-order and first-order templates, since we considered that they are sub-templates of second-order templates. We achieved this by padding a \(\mathrm{[MASK]}\) neighbor. For example, \(\mathcal{T}^0(\mathrm{[MASK]}, v)\) is implemented as \(\mathcal{T}_\mathrm{identical}^2(\mathrm{[MASK]}, v, \mathrm{[MASK]})\), and \(\mathcal{T}_r^1(\mathrm{[MASK]}, v)\) is implemented as \(\mathcal{T}_r^2(\mathrm{[MASK]}, v, \mathrm{[MASK]})\). To get \(\boldsymbol{x}_v\) and \(\boldsymbol{y}_v\) from this template, you only need to ignore the output of the second mask.

Details about the base model The base model is a BioSyn-like model with some important modifications. We trained the model for 30 epochs with initial learning rate 1e-5, and decayed it to about 1e-6 when the model converged. We used sparse features [1] during candidate generation. During encoding, we didn’t add sparse features, since we found that sparse features had no significant impact on the results, and even caused a slight decrease of accuracy. Besides, we found that BioSyn [1] sometimes failed to retrieve positive candidates due to limited candidate size and the inaccuracy of the model. Therefore, we manually added positive candidate in order to make full use of training data. The procedure for inference is the same as BioSyn [1].

Details about baselines NormCo [2] was initially introduced to perform bio-entity linking with inputs being text corpora. Central to its proposed method is the modeling of coherence leveraging concept co-mentions in each text corpus. However, as NormCo is not designed to learn the semantics of concepts, it is not capable of zero-shot learning in our dataset. Therefore, we did not report its results under zero-shot setting.

In addition, to construct a coherence sequence analogous to the co-appearances of mentions in the original setting, we took the mentions (entities and synonyms) of neighbor concepts of each training concept (excluding validation and test mentions when training), where the mentions are arranged in order based on their distance from the central concept we build this sequence for.

As Sieve-Based [3] is a rule-based entity normalization method which does not need the training data, we treated the model as a zero-shot model. Besides, Sieve-Based does not include a scoring mechanism, so we could only report the results of Acc@1.

References↩︎