Slide 14
Slide 14 text
14
上位チームの解法 | magic
金メダル圏のほとんどのチームはあるトリック(magic)で大幅にスコアを向上させた
同一anchor,contextの異なるtargetの文字列を連結してbertに突っ込む
何をしているか?
1. 同一anchor,contextのtargetを文字列として連結(以下targetsと呼ぶ)
df.groupby(["anchor","context"])["target"].apply(lambda x :" ".join(list(x)))
2. targetsをinput textに連結
text = anchor[SEP]target[SEP]CPC_TEXT[SEP]targets
どんな入力になるか?
abatement [SEP] abatement of pollution [SEP] HUMAN NECESSITIES. FURNITURE; DOMESTIC ARTICLES OR APPLIANCES; COFFEE MILLS; SPICE
MILLS; SUCTION CLEANERS IN GENERAL [SEP] abatement of pollution,act of abating,active catalyst,eliminating process,forest
region,greenhouse gases,increased rate,measurement level,minimising sounds,mixing core materials,multi pollution abatement device,noise
reduction,pollution abatement,pollution abatement incinerator,pollution certificate,rent abatement,sorbent material,source items pollution
abatement technology,stone abutments,tax abatement,water bodies
https://www.kaggle.com/competitions/us-patent-phrase-to-phrase-matching/discussion/332273