flowchart LR
subgraph "2021년 R 컨퍼런스: 챗GPT 이전"
A["R 컨퍼런스"] --> AB["키노트"] --> B["제작방법"]
end
subgraph "서울 R 미트업: 챗GPT 이후"
direction LR
Youtube["챗GPT 오정보"] --> STT["STT: Whisper"]
Youtube --> Summary["요약: Langchain"]
Youtube --> Embedding["임베딩: Pinecone"]
Youtube --> ETC["..."]
end
B -.-> Moat["We Have No Moat,<br> And Neither Does OpenAI"]
STT -.-> Moat
Summary -.-> Moat
Embedding -.-> Moat
ETC -.-> Moat
click A href "https://use-r.kr/"
click AB href "https://www.youtube.com/watch?v=clNN1xAiEkw"
click B href "https://statkclee.github.io/deep-learning/rconf-keynote.html"
click Youtube href "https://www.youtube.com/watch?v=csqOtqFBybk"
click STT href "https://r2bit.com/chatGPT/whisper.html"
click Summary href "https://r2bit.com/chatGPT/news_release.html"
click Embedding href "https://r2bit.com/chatGPT/misinfo_embedding.html"
click Moat href "https://www.semianalysis.com/p/google-we-have-no-moat-and-neither"
class A nodeStyle
class AB,AC,AD,AE theoryStyle
class B,C,D,E practiceStyle
class F taskStyle
classDef nodeStyle fill:#93c47d,stroke:#000000,stroke-width:0.7px,font-weight:bold,font-size:14px;
classDef theoryStyle fill:#f59542,stroke:#000000,stroke-width:0.7px,font-weight:bold,font-size:14px;
classDef practiceStyle fill:#4287f5,stroke:#000000,stroke-width:0.7px,font-weight:bold,font-size:14px;
classDef taskStyle fill:#42f548,stroke:#000000,stroke-width:0.7px,font-weight:bold,font-size:12px;
챗GPT와 데이터 과학
대통령 취임사 임베딩
2023년 5월 11일
서울 R 미트업
챗GPT API 가격
OpenAI 임베딩
| Models | Use Cases | |
|---|---|---|
| Text similarity: Captures semantic similarity between pieces of text. | text-similarity-{ada, babbage, curie, davinci}-001 | Clustering, regression, anomaly detection, visualization |
| Text search: Semantic information retrieval over documents. | text-search-{ada, babbage, curie, davinci}-{query, doc}-001 | Search, context relevance, information retrieval |
| Code search: Find relevant code with a query in natural language. | code-search-{ada, babbage}-{code, text}-001 | Code search and relevance |
작업흐름
원천 데이터
시각화
library(ggrepel)
extrafont::loadfonts()
inaugural_umap <- umap(embeddings_mat)
umap_df <- inaugural_umap$layout %>%
as.data.frame()%>%
rename(UMAP1="V1",
UMAP2="V2") %>%
bind_cols(api_embedding_tbl) %>%
mutate(구분명 = glue::glue("{대통령}_{str_remove(구분, ' ?취임사 ?')}")) %>%
select(UMAP1, UMAP2, 구분명)
umap_df %>%
ggplot(aes(x = UMAP1,
y = UMAP2))+
geom_point(size = 1.3, alpha = 0.8) +
geom_text_repel(aes(label=구분명)) +
theme_bw(base_family="NanumGothic")
# 영어만 추출
only_english_tbl <- api_embedding_tbl %>%
filter(str_detect(구분, "영문"))
english_embeddings_mat <- matrix(
unlist(only_english_tbl$임베딩),
ncol = 1536, byrow = TRUE
)
# 시각화
english_umap <- umap(english_embeddings_mat)
english_umap_df <- english_umap$layout %>%
as.data.frame()%>%
rename(UMAP1="V1",
UMAP2="V2") %>%
bind_cols(only_english_tbl) %>%
mutate(구분명 = glue::glue("{대통령}_{str_remove(구분, ' ?취임사 ?')}")) %>%
select(UMAP1, UMAP2, 구분명)
english_umap_df %>%
separate(구분명, into = c("대통령", "번역여부"), sep = "_") %>%
ggplot(aes(x = UMAP1,
y = UMAP2))+
geom_point(aes(color = 번역여부), size = 1.3, alpha = 0.8) +
geom_text_repel(aes(label=대통령)) +
theme_bw(base_family="NanumGothic") +
theme(legend.position = "top") +
labs(title = "대통령 취임사 유사도")
참고문헌
Reimers, N. (2022). OpenAI GPT-3 Text Embeddings - Really a new state-of-the-art in dense text embeddings? Medium.com. https://medium.com/@nils_reimers/openai-gpt-3-text-embeddings-really-a-new-state-of-the-art-in-dense-text-embeddings-6571fe3ec9d9
