2023-02-18干活记录

MathBERT:

耗时：2-3hours(昨天和人聊天聊完了，今天九点才到实验室，呜呜呜一早上就看了个论文)

读论文：BERT-Based Embedding Model for Formula Retrieval

Corpus Description：

resource:

from MSE;the formulas extracted from the question, answer, and comment posts

the formulas are represented in three different formats:

L ATE X, Presentation MathML,Content MathML format.

Each format has five distinct attributes:

formula_id, post_id,thread_id, type, formula

early system:

variable-size vector motivated by Bit Position Information Table(BPIT) and Term-Document matrix

The prime objective of this system :

to transform the formula into the variable size vector.

Each weight of the vector represents the occurrence count of a particular entity in a formula and corresponds to entity position in BPIT .

the proposed system in ARQMath2:

 

 

无语，看了半天这篇工作的预训练好像是照搬的BERT。我也是醉了...

爬虫 and NER 加上 TC：

耗时：4-6hours

学会了怎么做一份粗糙的csvNER打标数据。

分完了数据集，构建了词表，明天写模型。

尽量一个月内学完这本鱼书：

回家看了一下深度学习的花书。