"RESIDE: Improving Distantly-Supervised Neural Relation Extraction using Side Information"简略笔记

“RESIDE: Improving Distantly-Supervised Neural Relation Extraction using Side Information”

**本篇为阅读此文章的简略笔记，以下的所有资料与信息均来自于 Shikhar Vashishth, Rishabh Joshi, Sai Suman Prayagz, Chiranjib Bhattacharyya, Partha Talukdar的文章"RESIDE: Improving Distantly-Supervised Neural Relation Extraction using Side Information"

RESIDE:

1. Github:
   http://github.com/malllabiisc/RESIDE
2. Definition:
   A distantly-supervised neural relation extraction method which utilizes additional side information from KBs for improved relation extraction.
3. Implementation:
   1)It uses entity type and relation alias information for imposing soft constraints while predicting relations.
   2)It employs Graph Convolution Networks(GCN) to encode syntactic information from text and improves performance even when limited side information is available

an overview of RESIDE

1. Sections:
   1] Syntactic Sentence Encoding:
   RESIDE uses a Bi-GRU over the concatenated positional and word embedding for encoding the local con- text of each token.
   For capturing long-range dependencies, GCN over dependency tree is employed and its encoding is appended to the representation of each token.
   Finally, attention over tokens is used to subdue irrelevant tokens and get an embedding for the entire sentence.
   2] Side Information Acquisition:
   we use additional supervision from KBs and utilize Open IE methods for getting relevant side information.
   3] Instance Set Aggregation:
   sentence representation from syntactic sentence encoder is concatenated with the matched relation embedding obtained from the previous step.
   Then, using attention over sentences, a representation for the entire bag is learned. This is then concatenated with entity type em- bedding before feeding into the softmax classifier for relation prediction.
2. Test:
   In our experiments, we evaluate the models on Riedel and Google Distant Supervision (GDS) dataset.
   
   Comparison of Precision-recall curve.

Relation Extraction(RE)

1. Aim:
   Since the Knowledge Base(KB) cannot be exhaustive, RE attempts to fill this gap by extracting semantic relationships between entity pairs from plain text.

2. Implementation:
   Formally, given an entity pair (e1 ,e2) from the KB and an entity annotated sentence (or instance), we aim to predict the relation r, from a predefined relation set, that exists between e1 and e2. If no relation exists, we simply label it NA.

3. Disadvantage:
   Most supervised relation extraction methods require large labeled training data which is expensive to construct. Distant Supervision(DS) can help to deal with this problem.

Distant Supervision(DS) :

1. Definition:
   create large datasets by heuristically aligning text to a given Knowledge Base(KB).

2. Implementation:
   Under the assumption that if two entities have a relationship in a KB, then all sentences mentioning those entities express the same relation.

3. Disadvantage:
   some of the Sentences might be wrongly labeled.